Technology

Looking for a different day?

A new NYT Strands puzzle appears at midnight each day for your time zone – which means that some people are always playing 'today's game' while others are playing 'yesterday's'. If you're looking for Friday's puzzle instead then click here: NYT Strands hints and answers for Friday, October 23 (game #964).

Strands is the NYT's latest word game after the likes of Wordle, Spelling Bee and Connections – and it's great fun. It can be difficult, though, so read on for my Strands hints.

Want more word-based fun? Then check out my NYT Connections today and Quordle today pages for hints and answers for those games, and Marc's Wordle today page for the original viral word game.

SPOILER WARNING: Information about NYT Strands today is below, so don't read on if you don't want to know the answers.

NYT Strands today (game #965) - hint #1 - today's themeWhat is the theme of today's NYT Strands?

• Today's NYT Strands theme is… Re-fresh-ing

NYT Strands today (game #965) - hint #2 - clue words

Play any of these words to unlock the in-game hints system.

• DRAKE
• READY
• SUNDRY
• CASK
• DETER
• GREED

NYT Strands today (game #965) - hint #3 - spangram lettersHow many letters are in today's spangram?

• Spangram has 10 letters

NYT Strands today (game #965) - hint #4 - spangram positionWhat are two sides of the board that today's spangram touches?

• First side: left, 4th row

• Last side: right, 4th row

Right, the answers are below, so DO NOT SCROLL ANY FURTHER IF YOU DON'T WANT TO SEE THEM.

NYT Strands today (game #965) - the answers

(Image credit: New York Times)

The answers to today's Strands, game #965, are…

• BASKET
• WASHER
• SINK
• RACK
• DETERGENT
• IRON
• SPANGRAM: LAUNDRYDAY

• My rating: Easy
• My score: Perfect

I didn’t need any hints, but I didn’t exactly smash this game out of the park…

Instead of seeing DETERGENT I added “deter” and “gent”, I bailed out of connecting LAUNDRYDAY at “laundry” and tried to tap out “cark” before RACK.

Having made such inelegant progress, it was somewhat ironic to be rewarded with the “perfect!” pop-up, as much like my own efforts at clothes washing it was far from it.

Yesterday's NYT Strands answers (Friday, October 23, game #964)

• NOSY
• EAVESDROPPING
• INQUISITIVE
• SNOOPY
• PRYING
• SPANGRAM: BUSYBODY

What is NYT Strands?

Strands is the NYT's not-so-new-any-more word game, following Wordle and Connections. It's now a fully fledged member of the NYT's games stable that has been running for a year and which can be played on the NYT Games site on desktop or mobile.

I've got a full guide to how to play NYT Strands, complete with tips for solving it, so check that out if you're struggling to beat it each day.

It's a big day for both Apple and Major League Baseball (MLB). Today, October 23, 2026, Apple launched its biggest iPhone change ever: the iPhone Duo. But in just a few hours, baseball's most important series will kick off, with the Tampa Bay Rays facing the Milwaukee Brewers in the World Series. Game 1 starts at 8 p.m. ET.

And there's a commonality here: MLB has launched a redesigned app experience built to take advantage of the iPhone Duo's foldable form factor. So, if you were lucky enough to pick one up, you can tune into the World Series on your Duo's 5.4-inch cover screen. Or, let's be real, you'll open it up and watch on that enormous 7.6-inch inner display, complete with its Nano-texture finish.

But this experience is more than just giving baseball a bigger screen. MLB has been working on the new experience since Apple announced the iPhone Duo on September 9, getting it ready for launch in just 44 days.

You can fold the iPhone Duo like a mini laptop — ahem, MacBook — or prop it up in a seated position, with the livestream on top and real-time, updating box scores on the bottom. Watch in full screen on either display, or split the screen with another app. Whether it's X (formerly Twitter), Threads, a group chat to discuss the game, or even Apple's Sports app, you can keep up with the conversation while keeping an eye on the action.

The idea is to meet fans where they are and make the most of the hardware they have in their hands. The iPhone Duo gives MLB a new canvas, but the real opportunity is to deliver something useful with it—an experience that makes following baseball feel more natural, not just bigger.

(Image credit: Apple)

Sean Curtis, MLB's chief technology officer, told TechRadar that he's particularly excited about combining the league's Game Day experience with live video.

"I'm really excited about, you know, the full-screen nature of just like how good our video looks on Apple devices and just really take that in as like, 'hey, this just came from my phone, but it really feels like just such a quality experience,'" Curtis said.

"I really like our Game Day experience merged with video, so watch video [and] live stats."

The combination lets fans follow the action and keep tabs on the numbers in one place, rather than bouncing between the two. MLB is also using AI to generate additional information about individual plays, players, and game situations, adding context to the action as it unfolds.

"When you get into Game Day, and you're actually applying context to the events, we're using AI to actually generate more information about each play, each player, each situation," Curtis said.

For a sport with so much happening between pitches, that combination could be particularly compelling. The video gives you the action, while the data helps explain it. Rather than forcing fans to choose between watching and digging into the numbers, MLB wants to bring those two parts of the experience closer together.

Kenny Gersh, MLB's executive vice president of Media, Business Development, and Product, highlighted the simple appeal of opening the device and immediately having a much bigger screen ready for video.

"You're on a phone that's in your pocket, you're sitting down somewhere, and just to be able to go like that — whoops, I messed it up with my finger — but to have it just open right up into that is amazing," Gersh told TechRadar.

It's a straightforward idea, but one that makes sense for live sports. The phone can stay compact in your pocket until you're ready to settle in and watch, at which point the larger inner display gives the MLB app more room to shine.

(Image credit: Jacob Krol/Future)

MLB's relationship with Apple goes back to the early days of the App Store. The league was among the first 500 apps available on the original iPhone, and its iPad app was ready for launch day, earning a spotlight during the original iPad keynote.

As Greg Joswiak, Apple's senior vice president of Worldwide Marketing, told TechRadar, "Sports are important to us; it's important to our users.” Over the years, the partnership has evolved alongside Apple's products and platforms. Gersh recalled working with Apple on HTTP Live Streaming (HLS), helping advance the technology needed to deliver live video reliably, including advertising insertion.

More recently, Apple and MLB have explored new ways to bring fans closer to the action, including spatial sports broadcasts on Apple Vision Pro and the use of iPhone cameras to capture footage for baseball broadcasts on Apple TV.

The iPhone Duo is the latest chapter in that relationship, and it feels particularly interesting. MLB isn't just showing fans that baseball works on a foldable phone; it's exploring how a different kind of device could change the way they follow the game.

(Image credit: Future/Jacob Krol)

And this is only the beginning. As the new experience arrives and the World Series gets underway, MLB has some of baseball's biggest games ahead to see how fans respond. Those early learnings could shape what comes next, with the off-season offering a chance to refine the experience and build something even more compelling for next year's Opening Day.

Curtis hinted at further developments he can't discuss yet, particularly around keeping fans engaged with the app without pulling them away from a live game.

For now, a 7.6-inch inner display, live video, real-time stats, and AI-generated context give baseball fans a new way to follow the action on their phones. The World Series offers an ideal first test, and the off-season gives MLB time to build on what it learns. If the league gets that balance right, the next version of its app could be even more compelling when baseball returns for Opening Day.

An Apple Music easter egg will let you experience the sound and the scroll of the click wheel.

• Stranger Than Heaven director Mikinobu Abe says the game's Showbiz system was "a real challenge" for RGG Studio
• Abe says lots of research was put into finding out what a show promoter does and how to depict that in-game
• Abe adds that "At some point we did have a small rhythm game," but the idea didn't match the team's vision

Stranger Than Heaven director Mikinobu Abe has revealed how RGG Studio designed the game's Showbiz narrative system and how difficult it was to get right.

In an interview with TechRadar Gaming, Abe said the Showbiz, an entertainment management system requiring the main character, Makoto Daito, to build a business, find sources of inspiration for music, and plan productions as a promoter, "was a real challenge for us."

The director explained that RGG started by asking: what does a promoter actually do? Because "you know what a singer does, and you know what a band does, but you don't exactly know what a promoter does in the music industry," so research began from there.

Abe revealed that "At some point we did have a small rhythm game, but it didn't exactly match the feeling that you're promoting a show," so the team went back to the whiteboard phase to try different things.

"One thing that we noticed is that a promoter is the person who brings in not just the money, but also the inspiration and the whole vision of the show," Abe said. "We thought that's an interesting aspect that we can put into this game."

In Stranger Than Heaven, players can walk around cities and interact with things in the environment, like a grandfather clock or a bonfire, and collect sounds as inspiration for music. Daito can then take these samples and collaborate with a composer to create new music and plan a production for the singer and band.

"The inspiration sources are basically like when you have a certain job, you always have these antennas looking for something," Abe said. "So, if you're a musician, you always have this antenna for the music you're hearing, and that can inspire people to create new stuff, and that was no different for Makoto as well.

Abe added, "That's why this game system of finding and then using that inspiration to create music came to life in the game."

Stranger Than Heaven releases on January 15, 2027, for PS5, Xbox Series X and Series S, and PC.

• Samsung Display is developing more efficient OLED pixel tech to rival LG's FLiPP tech
• It could start production very soon, but the prototype is only 7 inches
• The tech seems to target watch and phone screens over larger panels

As we reported last week, LG Display is moving forward with next-gen Fine Metal Mask-free OLED pixels that could mean brighter, more efficient and maybe more affordable OLED panels. And it looks like Samsung Display is thinking along very similar lines.

A new report by trade site The Elec says that Samsung Display has been working on a prototype seven-inch FMM-free OLED panel using its proprietary Micro Patterned OLED technology, or MPO for short.

The Elec has been speaking to unnamed industry sources, who said that Samsung has been working on MPO product development since March 2026 with a target date at the end of 2026, although that may slip into early 2027.

As with LG's FLiPP technology, the plan is to focus on a particular screen size at first. Samsung is aiming at phone or watch-sized displays, while LG is aiming at laptop or monitor-sized displays. (Neither is aiming at TVs in the near future, alas.)

But while the two companies are taking slightly different approaches to FMM-free manufacturing, the result is likely to be the same: brighter, more efficient and more affordable OLED panels.

LG's FLiPP process promises to reduce waste and make OLED production more efficient (Image credit: LG Display)What is Samsung Display's MPO technology, and how does it differ from LG Display's FLiPP?

First, why do these companies want to drop 'Fine Metal Mask' tech? This is currently how the material for OLED pixels is deposited into each pixel, and while the tech is useful and mature, it makes for pixels that aren't very efficient.

The primary goal of moving to a more efficient tech is make it so that more of the area of each pixel is filled with light-emitting elements, which means the panel will be brighter, or can be the same brightness but uses less power to do it, which great for battery life and longevity of devices. FMM-free pixels could also be cheaper to make, long-term.

LG's FLiPP technology (FMM-Less innovative Pixel Patterning) essentially covers the panel with organic materials and then removes the bits it doesn't want via etching and photolithography. MPO does things slightly differently, creating microscopic structures that fence in the organic materials so they're only laid where they need to go. Those structures are then removed. That means less waste and simpler manufacturing, which will ultimately mean lower manufacturing costs, in theory.

The most important difference may not be technical, however: as The Elec points out, there are various active patents regarding FMM-free technologies, and MPO may enable Samsung Display to avoid them.

What we don't know yet is how well the technology will scale. As one of the unnamed industry insiders explained, "MPO is being developed in a direction better suited to high-resolution OLED displays than medium- and large-sized panels. Once a 7-inch-class sample is available, it will be possible to assess how effectively MPO’s technical advantages can be realized at an actual product size."

The big problems for OLED screens, and the thing that keeps them more expensive than LCD panels, is yield rates — that means how many panels you produce are actually usable. The new FMM-free technologies might work perfectly and be cheaper in terms of the material used, but if only 50% of the panels you make are good enough to ship, you haven't actually saved anything.

That's why the production test runs of both technologies will be the make-or-break moment — though both are behind the roll-out of inkjet-printed OLED, arriving in monitors now, which hopes to deliver all the same advantages.

Try our TV size and model finder! You tell it how far you sit from your TV, we'll tell you what size to buy based on viewing angle advice from image quality experts, and we'll recommend our three top TVs at that size for different prices.

Apple Watch Series 12: One minute review

While the Apple Watch Series 12 is a decidedly modern-looking smartwatch in a pleasingly thin profile, it’s not a major redesign from the Cupertino-based tech giant. And if you're upgrading from the Series 11, you'll probably notice more of a difference in what the watch can do rather than in how it looks. Apple has updated the Health Sensing System (its array of LEDs and sensors used to collect your health data), introduced Readiness Score, and made everyday tracking more useful, with real-time steps and heart-rate information easier to access.

I've been wearing the Series 12 for more than a month, and Readiness Score is the addition I've found myself using the most. It takes your sleep and other health data and turns it into a score out of 10, giving you a quick idea of how ready you might be for the day. I've generally found the results match how I feel, and it's a feature that can genuinely influence what I do, whether that's heading out for a workout or taking a longer walk with my dog, Rosie.

I've also been comparing the Series 12 with the Series 11, Apple Watch Ultra 3 and Ultra 4, as well as the Oura Ring 5 and Pixel Watch 5. That has helped me put Apple's latest watch into context, particularly when looking at health tracking and heart-rate readings. The Series 12 has generally delivered the sort of everyday experience I'd expect from an Apple Watch, with battery life that's worked well for me most of the time.

A couple of things to consider, though. The design is very familiar, so there's little reason to upgrade purely for its appearance, and I did have two nights when the battery ran out before morning. Apple's upcoming AI features are also still a reason to wait rather than buy immediately. If they're what you're most interested in.

For anyone using an older Apple Watch, particularly a Series 9 or earlier, the Series 12 makes a much stronger case. The combination of newer health features, Readiness Score, performance, and battery life makes this an upgrade worth considering.

Apple Watch Series 12: Specs Table

Component

Apple Watch Series 12 (42mm)

Apple Watch Series 12 (46mm)

Price

$399 / £369 / AU$679 for the 42mm

$429 / £399 / AU$699 for the 46mm

Dimension

46 x 40 x 9.7mm

42 x 37 x 9.7mm

Weight

31.5g (without band)

38.6g (without band)

Case/bezel

Aluminum, Titanium, or Ceramic

Aluminum, Titanium, or Ceramic

Display

42mm Always-On Retina LTPO3 OLED display (up to 2,000 nits)

46mm Always-On Retina LTPO3 OLED display (up to 2,000 nits)

Processor

Apple-made S11 chip

Apple-made S11 chip

GPS

Precision L1 GPS (GPS, GLONASS, Galileo, QZSS, and BeiDou)

Precision L1 GPS (GPS, GLONASS, Galileo, QZSS, and BeiDou)

Battery life

Up to 24 hours with normal use

Up to 24 hours with normal use

Connections

4G LTE, Satellite SOS, Wi‑Fi 4 (802.11n), Bluetooth 5.3

Wi‑Fi 4 (802.11n), Bluetooth 5.3, and 4G LTE

Water resistant

Water resistance 50m

Water resistance 50m

Apple Watch Series 12: Price and availability

• $399 / £369 / AU$679 for the 42mm
• $429 / £399 / AU$699 for the 46mm
• Prices unchanged from last year

The Apple Watch Series 12 starts at $399 / £369 / AU$679 for the 42mm with a recycled aluminum case and standard strap, and $429 / £399 / AU$699 for the 46mm. LTE-capable models, which can be added to your data plans, as well as optional improvements to the finish — titanium or ceramic — are available at an additional cost.

It doesn’t represent outstanding value in the same way as, let’s say, the Amazfit Active 3 Premium might, but it’s fine — the price has actually held steady on Apple Watches for the last couple of years, which is encouraging given the electronics industry is undergoing a supply chain crisis at the moment.

• Value score: 4.5/5

Apple Watch Series 12: Design

(Image credit: Future/Jacob Krol)

• Same ‘squircle’ and digital crown design
• Retains slimmer look from Series 11
• Redesigned sensor array and GPS architecture

If you expect the Apple Watch Series 12 to look substantially different from the Series 11, you'll be disappointed. Apple has kept the familiar rectangular ‘squircle’ design, with the Digital Crown and side button on the case, and the overall look remains very much the same.

After using the Series 12 for more than a month and comparing it with the Series 11, I haven't found the exterior to be a major reason to upgrade. It doesn't look dramatically different on your wrist, and Apple hasn't tried to reinvent a design that already works.

That's not necessarily a problem, though. The Apple Watch Series 12 remains a comfortable everyday smartwatch, and wearing it all day and overnight matters when you use it for sleep tracking and the Readiness Score. The biggest changes are on the rear of the case, as Apple has redesigned the LEDs and internal architecture to make its data collection more accurate. However, a watch can have all the health features in the world, but they're less useful if you don't want to wear it consistently.

(Image credit: Future/Jacob Krol)

The Series 12 comes in 42mm and 46mm sizes, with aluminum and titanium options alongside ceramic, which is returning and still adds a bit more heft. That gives you some choice in look and finish, though the core experience stays the same. The aluminum model is the more straightforward choice for most people, while titanium and ceramic offer alternatives if materials and appearance matter more to you.

The always-on OLED display is familiar, too. It lets you glance at the time, check a notification, or see your workout metrics without having to wake the screen manually. That's one of those features that quickly becomes part of your routine, particularly if you use complications to keep useful information on the watch face.

One addition I've found handy is the real-time step-count complication. I check steps regularly, and seeing the number at a glance saves me from opening the Fitness app. The heart rate complication works the same way, especially when I want to keep an eye on my readings during the day.

Compared with the Ultra 3 and Ultra 4, the Series 12 feels more like a familiar everyday watch. The Ultra models have their own appeal, particularly if you want a larger, more rugged watch, but the Series 12 is easier to recommend to someone who doesn't need that design.

The main point is that the Series 12 doesn't feel like a new design generation. If you're upgrading, look at the back of the case, with its improved health features enabling the Readiness Score and overall experience, rather than expecting a noticeable aesthetic change to the watch itself.

• Design score: 4.5/5

Apple Watch Series 12: Features

(Image credit: Future/Jacob Krol)

• Readiness score is a major boon
• More frequent heart rate data collection
• Upcoming Audio Intelligence features

The Readiness Score is the biggest reason I'd consider the Series 12, and it stood out most during my testing. It uses overnight sleep and HRV data, along with other health and activity information and some custom processing, to produce a score from 1 to 10. The idea is to give you a quick indication of how ready you might be for the day, rather than leaving you to interpret a collection of separate metrics yourself. Tested against the Garmin Cirqa, our senior fitness editor Matt Evans found both devices provided identical Readiness scores after wearing both for one week’s worth of sleep.

I like how Apple presents it: a half-moon graphic that’s easy to read, with an accompanying explanation that gives context for the score. It feels grounded in the data, rather than relying on vague wellness language that doesn't explain why you got a particular result.

My scores have generally been high, and they've mostly matched how I've felt. Getting my first 10 out of 10 was satisfying, of course, but the more interesting part is how I can use the information. If I've slept well and feel good, a strong score might encourage me to fit in a workout or take Rosie on a longer walk. It won't make every decision for me, but it gives me another useful piece of information to consider.

I've used an Oura Ring before, so I know the appeal of a daily Readiness score and the help it can provide beyond raw data. I appreciate that Apple built the feature into the watch I already use. I don't need to add another device to my routine, and Readiness Score doesn't require a separate subscription.

That makes it a particularly interesting addition for existing Apple Watch owners who want more from the health data their watch collects. Instead of simply checking how long I slept or looking at the previous day's activity, I get a more immediate summary I can use to plan my day. However, we have to acknowledge the wait for this feature — most other devices have had Readiness for years, but Apple has stated the feature has been enabled by its redesigned sensors taking heart rate and HRV readings more frequently, up to 60 times more often than previous Apple Watches.

(Image credit: Future/Jacob Krol)

I've also been paying attention to heart rate readings while comparing the Series 12 with the Apple Watch Series 11, Ultra 3, and Ultra 4, along with the Oura Ring 5 and Pixel Watch 5. In my everyday use, the Series 12's readings have been broadly consistent with what I've seen from those other devices.

Having heart rate information readily available is useful during workouts, too. I've used the watch for yoga, outdoor walks and runs, and indoor cycling, and it has fit naturally into those sessions. The real-time heart rate complication also makes it easier to check my readings without navigating away from the watch face.

The Series 12 retains the wider Apple Watch health feature set, including ECG, heart-rate notifications, wrist-temperature sensing, sleep tracking, respiratory-rate tracking, cycle tracking, hearing-health features, hypertension notifications and blood-oxygen tracking where supported. These aren't all new to this generation, but they contribute to the overall package.

Apple's upcoming AI (‘Audio Intelligence’ this time) features aren't a strong reason to buy right now. Live Rewind and Siri Recap sound interesting, but I haven't been able to test them because they aren’t out yet. If you're considering the Series 12 for those, I'd wait until Apple confirms when they're shipping and what will be available in your region, as there is some contention with Siri AI in certain regions such as the EU.

For now, Readiness Score is the feature I'd focus on. It's useful today, fits into the existing Apple Watch experience, and has made the biggest difference in how I use the Series 12. Best of all, it’s not locked behind another subscription … and that’s simply refreshing.

• Design score: 5/5

Apple Watch Series 12: Performance

(Image credit: Future/Jacob Krol)

• S11 chip drives excellent performance
• Accurate compared to other wearables
• Responsive with no stutters

While the Series 12 still feels and looks like an Apple Watch in pretty much every sense of the word, you can’t judge a book by its cover, especially when it comes to performance. Under the hood, the Series 12 runs on Apple’s new S11 chip, and the new health-sensing system on the back enables the Readiness score and HRV features. All of this comes together to make the Series 12 feel responsive, with none-to-minimal stutters when calling on a particular feature set.

I've now used the Series 12 for more than a month, and I've compared it directly alongside the Series 11, Ultra 3, and Ultra 4. These comparisons have helped me understand how the Series 12 fits into the wider wearable landscape, and while the latest generation doesn’t run laps around the previous gen or the one before that, the responsiveness is definitely noticeable.

Whether you’re customizing a Watch Face with your favorite complications, calling on Siri to remember a piece of information or drafting a message, controlling media playback, making a call, recording a voice memo, checking your activity score, monitoring your heart rate, or starting a workout, the Series 12 feels very responsive. We’ve also tested the same heart rate system, albeit on the Ultra 4, against a chest-mounted heart rate monitor, the Frontier X2, alongside the Garmin Fenix 9 Pro, and found Apple’s latest gen is highly accurate.

Battery life has been good overall, and it certainly meets Apple's all-day battery life promise. On a normal day with plenty of interaction points and one extended workout, I typically have around 30%–45% battery remaining by bedtime. That's enough to get through a normal day, and faster charging makes it easier to top up when I need to. Depending on how I use it, I can get through a full day and sometimes stretch it to a day and a half.

So far, the Series 12 has run out of battery overnight twice, and I think it’s especially worth mentioning because you need a full night of sleep data to wake up with a Readiness Score. And that means an unexpected shutdown can interrupt that routine. My general solution, even if the percentage hasn’t dropped super low, is to give it a quick charge before bed rather than assuming it will always have enough power to last the night.

The faster charging helps here. A short top-up with the USB-C to Apple Watch dongle that ships with the Series 12 can make it easier to keep the watch going overnight while still wearing it.

If you’re coming from an older Apple Watch, the all-day battery life paired with fast-charging will likely make a big difference. It’s not just a quality-of-life upgrade from, say, a Series 10 or older; compared with the rest of the features, it's a significant leap in what watchOS can accomplish and how long you go before you need to charge.

• Performance score: 4.5/5

Should you buy the Apple Watch Series 12?Buy it if..

You're upgrading from an Apple Watch Series 9 or older and want better health tracking, Readiness Score, and faster charging.

You want a responsive everyday smartwatch with useful at-a-glance health metrics and no extra subscription for Readiness Score.

Don’t buy it if..

You already own an Apple Watch Series 11 and want a major design change or a substantial leap in performance.

You're buying primarily for Apple's upcoming AI features, or you need battery life you can count on for every night of sleep tracking without a top-up.

Also consider

Garmin Fenix 9 Pro

The Apple Watch Ultra 4's biggest competitor in the premium adventure watch space.

Read our full Garmin Fenix 9 Pro review

How we tested

We tested the Apple Watch Series 12 by wearing it day and night, tracking workouts, monitoring sleep, and putting its features to the test over multiple weeks. We compared its health and fitness readings with other wearables and an electrical heart rate monitor, tested battery life, and explored as many features as possible to see how it holds up in everyday use.

First reviewed October 2026.

The mobile app is giving users more control of the videos they get.

• Marvell reveals Photonic Fabric Memory Module, the building block of a 16-module appliance sharing 32TB of DDR5 and 1,152GB of HBM3e across 16 hosts over fiber
• The design dates back to Celestial AI's 2025 presentations, currently remains pre-silicon, and is targeting customer sampling by 2027
• Nvidia is already hedging with a $2 billion Marvell stake in addition to its own BlueField-4 STX context memory, Enfabrica's pooling tech, and a co-packaged optics roadmap

Marvell has put physical hardware behind one of the more ambitious ideas in AI infrastructure.

In a video posted to X, and earlier in a theCUBE interview recorded around its October 6 Investor Day, the company showed off its Photonic Fabric Memory Module, which it bills as the "first optically interconnected high-capacity, high-bandwidth memory."

The pitch is simple: use light instead of copper so that dozens of terabytes of memory can sit in a separate box, racks away, and be shared by many processors at once, something that could effectively birth the widespread use of communal or shared memory across multiple AI servers.

Two terabytes at a time

Some clarity, however, is warranted: the 32TB figure belongs not to the module but to the Photonic Fabric Memory Appliance, which gangs 16 modules together and shares the pool across up to 16 hosts.

Those hosts can be whole servers or individual accelerators, so 16 GPUs might understate the potential reach in one sense and at the same time, overstate the bandwidth any single GPU will see in another. Nothing is shipping just yet, either: Marvell is targeting customer sampling in 2027.

Each module pairs 72GB of HBM3e with slots for up to eight DDR5 DIMMs, a Photonic Fabric ASIC, and 7.2Tbps of optical I/O, according to Marvell's August 2026 Hot Interconnects presentation.

Hitting 2TB per module means filling all eight slots with 256GB RDIMMs and sixteen modules, joined by a passive all-to-all fiber shuffle rather than an electrical switch, yields 32TB of DDR5 plus 1,152GB of HBM3e, which Marvell rounds to about 1TB and runs as a faster write-through cache in front of the DRAM.

Marvell is going after the economics of today's accelerators. HBM sits roughly 1.5 millimeters from the processor inside the same package, so the only way to buy more of it is to buy more processors. This is increasingly expensive because HBM is extremely costly, and it also means a GPU essentially acts as "the world's most expensive memory controller," as per Dave Lazovsky.

The target workload is the KV cache, the working memory a large language model keeps for each conversation, which balloons with long contexts and agentic sessions. Once it outgrows local HBM and host DRAM, servers either evict it and recompute, or fetch it from much slower storage. This has significant performance implications, and Marvell plans to mitigate it differently.

Ravi Mahatme, who leads product management for the Photonic Fabric unit, said the biggest payoff of a shared tier is "a cross-node KV cache," letting one accelerator reuse context another has already computed. Will Chu, Marvell's EVP of Custom Cloud Solutions, made the broader case in the company's FMS announcement: "As AI scales, memory must scale more independently of compute."

The CAPEX argument is rooted in this: If operators can buy memory separately from accelerators, they no longer need to buy GPUs just to get capacity. This makes sense, but Marvell has yet to put a number to it, even as a May 2026 company blog talks about lower memory cost per gigabyte and lower inference costs without figures, and Celestial AI's 2025 conference slides promised benefits beyond TCO reduction with no supporting numbers.

The timing also isn't ideal for harnessing cheap memory: the approach's 128 DIMM slots draw on the same DDR5 caught up in the current memory crunch, where server RDIMM prices have climbed faster than consumer memory, now around 5.5 times their pre-crisis price in the US.

The appliance design is not entirely new either; Celestial AI presented the same architecture, 16 modules holding 32TB of DDR and about 1TB of HBM, at Hot Interconnects in August 2025, and TechRadar has covered its DDR5-plus-HBM approach back in 2024. What has changed is the Marvell badge, a physical module, and a sampling date.

Nvidia is already hedging, both as an investor and a pioneer

The idea that Nvidia won't sit idle is more than an assumption, because it has been moving for more than a year, including taking a stake in Marvell itself. In March 2026, Nvidia invested $2 billion in Marvell in a deal covering NVLink Fusion and silicon photonics, with Jensen Huang declaring that "The inference inflection has arrived."

It is also important to understand that every NVLink Fusion platform must include at least one Nvidia product, so Nvidia gets paid even when custom silicon wins the socket, but that doesn't mean it doesn't want a piece of that pie anyway.

It is bringing its own answers to the same memory problem in multiple ways; A single Vera Rubin NVL72 rack carries 54TB of LPDDR5X, more than Marvell's entire appliance.

At Nvidia GTC 2026, the company launched BlueField-4 STX, a flash-based context memory tier built for KV cache, claiming up to five times the token throughput of CPU-based storage, with Huang saying reasoning AI needs "a new class of storage." Partner systems are due in the second half of 2026. Back in 2025, Nvidia reportedly spent more than $900 million on Enfabrica, whose EMFASYS system pooled up to 18TB of memory for GPU clusters.

On optics, Nvidia has put $4 billion into Lumentum and Coherent, unveiled photonics switches at GTC 2025, and will offer co-packaged optics as an NVLink option in Feynman-generation systems due in 2028. On the choice between copper and optical scale-up, Huang has said: "We're going to do both."

Marvell still has a long way to go: working silicon, a named customer, a memory partner, and measured throughput on real hardware rather than simulated curves all need to play out exactly as planned.

Until then, the appliance is a well-argued blueprint, and the CAPEX savings remain an argument rather than a number. The irony is hard to miss: the company whose GPUs Lazovsky casts as the world's priciest memory controller owns a $2 billion slice of Marvell, and through NVLink Fusion, stands to earn on the racks either way.

We’ll fold Apple’s iPhone Duo 900 times an hour to see how its hinge holds up.

Khadas Mind Pro: 30-second review

Having reviewed the Khadas Mind 2S just over a year ago, I was impressed by just how much processing power Khadas had managed to squeeze into such a small machine. Now, with the arrival of the Mind Pro, the company has retained much the same compact form factor and innovative modular design, while introducing Intel’s latest Core Ultra processor and Arc integrated graphics.

The result is an exceptionally powerful mini PC that feels just as capable as many considerably larger desktop systems, especially when it comes to everyday productivity as well as creative applications.

It once again features the premium metal construction, and with dimensions of just 146 × 105 × 28mm, it’s small enough to slip into a bag and carry between home and office. What makes this particularly useful is the integrated 7.98Wh standby battery, which allows the machine to maintain your Windows session while moving between those locations, much the same as using a laptop.

During testing, I was able to leave applications open, disconnect the computer, throw the Khadas Mind Pro in my backpack and travel home. Then I was able to reconnect to a monitor, mouse and keyboard at home and continue the session without needing to re-open applications and work. This is what we have come to expect from laptops, but not with a Mini PC, and it’s a surprisingly useful feature, boosted by the option of extras such as the Dock and GPU modules.

Performance is equally impressive as it features an Intel Core Ultra X7 358H, combined with 64GB of RAM and integrated Arc B390 graphics. This combination proved to handle Microsoft Office, Photoshop and Lightroom extremely well. Even Adobe Premiere Pro and DaVinci Resolve proved more usable than I had anticipated with 4K footage shot on the GoPro Mission 1 ILS.

Gaming was another surprise, with Cyberpunk 2077 and Indiana Jones and the Great Circle both running at medium-quality settings; although as I extended play there were occasional frame drops, there was nothing too much to impact gameplay. The other point here was that the fans kicked in and, while audible, they remained relatively quiet.

The machine also has considerable AI potential, with a dedicated NPU and 64GB of shared memory. My initial CPU-based Lemonade AI testing showed that it could handle practical local AI tasks, although switching between models wasn’t always straightforward. Using the dedicated NPU with the right models gained some impressive results, but again selecting the right LLM manager and model takes some time, this process took me from Lemonade to the LM Studio.

Through the test, my main issue with the machine was the connectivity. With only two Thunderbolt 4 ports, it’s easy to run out of high-speed connections when using a USB-C monitor, power supply and external storage. If you are using the Mind Pro as your main desktop, this is where the optional Mind Dock starts to make sense and is more like an essential purchase.

Ultimately, the Mind Pro is an impressive example of innovation and how far mini PCs have progressed. It won't be the best mini PC for most users. It isn’t cheap, and you’ll need to invest further to see the full potential of its modular design, but the impressive processing power, portability and modular extras make it expensive but very desirable.

Khadas Mind Pro: Price and availability

• How much does it cost? From $2,559
• When is it out? Available now
• Where can you get it? Directly from Khadas

The Khadas Mind Pro costs $2,599 for the Core Ultra X7 358H, 64GB RAM and 2TB SSD configuration that I’ve looked at in this review and is available directly from Khadas. Alongside the Mini PC, you can also purchase the optional Mind Dock, Mind xPlay and Mind Graphics 2, although these will add considerably to the overall cost.

• Value: 4 / 5

(Image credit: Alastair Jennings)Khadas Mind Pro: Specs

CPU: Intel Core Ultra X7 358H, 16 cores / 16 threads
Graphics: Intel Arc B390 integrated graphics
NPU: Intel AI Boost NPU 5, up to 50 TOPS
RAM: 64GB LPDDR5X, 8533MT/s
Storage: 2TB PCIe 4.0 M.2 2230 SSD
Additional storage: M.2 2230 NVMe expansion slot
Rear Ports: 2 x Thunderbolt 4, 2× USB-A 3.2 Gen 2, HDMI 2.1 FRL
Expansion: Khadas Mind Link proprietary modular interface
Connectivity: Wi-Fi 7, Bluetooth 6.0
Audio: Audio output via supported display, USB and compatible expansion accessories. Battery: 7.98Wh internal standby battery
Dimensions: 146 × 105 × 28mm
Weight: 562g
Operating System: Windows 11 Home
Power: 100W USB-C power adapter
Optional Accessories: Mind Dock, Mind xPlay and Mind Graphics 2

Khadas Mind Pro: Design

Unpacking the compact AX9 Max, the metal casing and detailed plastic base of this mini PC instantly reinforced its price point. Only after carefully checking the detailing and the large, shiny logo on top did I start to question whether this was a premium machine or just one trying to look premium.

Checking the specifications and that the price point was justified, but there’s still something about the design that feels a little over the top. The rest of the design is solid, and at 134.9 × 131.8 × 57.7mm and 777.5g, it’s a great size and weight for this style of machine.

Around the casing, there are four USB 3.2 Gen 2 Type-A ports and a 3.5mm stereo headphone jack. On the left side of the casing is an SD 4.0 memory card slot, while the back has two USB4 Type-C ports, a USB 2.0 port, two HDMI 2.1 ports, two 2.5Gb RJ45 sockets, a USB 3.2 Gen 2 Type-A port, and the DC-in port.

(Image credit: Alastair Jennings)

Having reviewed the Khadas Mind 2S just over a year ago, I was impressed by the combination of its exceptionally small size, processing power and innovative modular design. With the arrival of the Mind Pro, it’s immediately apparent that the company has retained much of the ethos of those initial machines and what made them so interesting.

The form factor is familiar, the design remains incredibly minimalist, and once again, it’s the modular expansion ecosystem that differentiates this from the usual mini PCs I look at.

The first impression is that this is an exceptionally well-made and small mini PC. The metal casing gives it a premium feel, with solid construction, and it’s just understated when it comes to styling, which I really like.

The front just features the power button mounted into a subtle cutaway, and all connectivity options are lined up at the rear.

The compact dimensions are great if you have limited desk space or you need to take the machine between locations. This portability is one of the key features and, like a laptop that enables you to close the lid without powering down, the Mind Pro enable you to do much the same when you move from one location to another. Simply unplug the monitor, keyboard and mouse, and pop it in your bag; then, when you get to the next location, connect up and off you go. All the apps and work that you were working on will still be there and ready for you to continue with.

The idea behind the machine is that you have the core, which is the Mind Pro, then you have the accessories, so in the office where you need to focus on high-demand work, you’ll have the Mind Graphics 2 eGPU Module and Mind Dock All-In-One Docking Station.

While at home, where you may just want to check email and do some light work, you may just need the Mind xPlay Portable Display + Keyboard, which will, in low-power mode, essentially give you the same abilities as a Windows Home laptop. You can also plug in the Tone2 for audio. The point is the Mind Pro is just the core, and how and where you use it will dictate what additional extras you want and need.

What makes the portability possible is the integrated 7.98Wh standby battery. This isn’t designed primarily to power the machine in the same way as a laptop battery and instead is used to maintain the machine's working state when disconnected from external power.

Taking a look at the limited ports on the back, there are two Thunderbolt 4 ports, an HDMI 2.1 connection and two USB-A 3.2 Gen 2 ports. While this matches most premium laptops, I found that the two Thunderbolt connections became a bit of an issue, particularly when using the Mind Pro for creative work.

With a Thunderbolt monitor and power connected, there was no fast connection left for connecting high-speed external SSDs or card readers without using an additional adapter. The two USB-A connections are useful, but their 10Gbps maximum transfer rate is far slower than the Thunderbolt 4’s 40Gbps connection bandwidth.

Considering the performance on offer here, a third Thunderbolt connection would have been welcome. For anyone intending to use the Mind Pro as their main desktop computer, the Mind Dock is essential, and at $199, that's quite an additional cost on an already expensive machine.

The all-metal construction does make the Mind Pro look like it should be passively cooled. However, there is an active cooling system inside, and while the fans become audible on many of the more demanding tasks, I was generally well controlled.

During intensive creative work, the casing also became warm to the touch, which isn’t particularly surprising given the amount of processing power packed into such a small space.

The design of this small machine is stunning, and the portability of the design with the integrated standby battery is actually a great idea. However, when it comes to the design, the limited port selection is frustrating, especially at this price, but it has been designed to be used as a modular setup, and the dock and GPU are available to boost the performance further at a cost.

• Design: 5 / 5

(Image credit: Alastair Jennings)Khadas Mind Pro: Features

The Khadas Mind Pro builds on the modular design of the machines that I have looked at in the past from the company and combines powerful processing with a proprietary expansion interface and an integrated standby battery.

At the heart of the system is Intel’s Core Ultra X7 358H, a 16-core processor supported by Intel Arc B390 integrated graphics and a dedicated AI Boost NPU. In this review, I have looked at the standard Pro machine with 64GB of LPDDR5X memory and a 2TB PCIe 4.0 SSD.

While those specifications are impressive, what separates this from most other mini PCs is that it has been designed so that the Mind Pro can easily move between different work and home environments, with a range of Mind Link connection expansion modules that adapt its use depending on your needs.

These modules include the Mind Dock that provides additional connectivity for when it’s used as a desktop workstation, and this comes in at around $200. Then there's the Mind xPlay portable display at $99, or with keyboard at $399, which offers a more mobile configuration.

There’s also the Mind Graphics 2 that adds dedicated graphics processing for creative applications, gaming and potentially supported AI workloads, and this will set you back a further $1449.

Unfortunately, none of these modules was available during my testing, so I wasn’t able to assess the complete ecosystem. Even without the expansion options, however, the Mind Pro has plenty of raw processing power.

The internal 7.98Wh battery is the most unique, and as I found this on a normal working day when I regularly travel between office, home and college, with a journey time that can vary from ten minutes up to an hour and a half.

With a conventional mini PC, that would mean saving everything, closing applications and shutting down before leaving, then re-opening everything when I get to my next destination. Which is one of the reasons I use a laptop, as I can keep everything open; however, the laptop isn’t light.

With the Mind Pro, I could simply disconnect the power, place the machine in my backpack and reconnect it at the other end, just like a laptop, just minus the screen, keyboard, and trackpad.

Throughout testing, this worked exceptionally well. I could leave Adobe Premiere Pro and other applications open, move between locations and return to the same Windows session without having to reopen everything.

It certainly makes for a lightweight alternative, although the internal 2TB SSD provides plenty of space for applications and smaller projects; I generally edit video from external storage.

This makes the portable workflow slightly more complicated, as unplugging an external SSD while Premiere Pro is accessing footage isn’t particularly advisable, but then that’s the same on a laptop. For shorter productions, however, the system really comes into its own.

During testing, I was editing approximately five-minute videos using footage from the GoPro Mission One, with the files stored directly on the internal SSD. This enabled me to start editing at the office, disconnect the computer and finish the project at home without closing Premiere Pro or shutting down Windows.

The battery also allows the Mind Pro to operate independently of external power for short periods, although this is very much a secondary feature. Using the Khadas Mind app, it’s possible to select the Use Battery option under Device Mode When Unplugged, allowing the computer to continue operating instead of immediately entering a standby state.

To see how practical this was, I tested the machine using three power settings, with a mixture of Microsoft Word, web browsing and Adobe Lightroom. At maximum power, the system lasted approximately seven minutes before displaying a low-battery warning and entering its low-power state.

Switching to the medium power setting in the Mind application was able to extend this working time to around 14 minutes, while the lowest power setting was around 30 minutes. These figures are far less than the 60 minutes quoted by Khadas for light active use. Most importantly, this internal battery feature means the machine wasn’t losing my work when the battery ran low.

• Features: 4.5 / 5

Khadas Mind Pro: Performance

(Image credit: Alastair Jennings)Benchmark scores

CrystalDiskMark Read: 5,224.09 MB/s
CrystalDiskMark Write: 4,860.54 MB/s
Geekbench 6 CPU Multi: 16,660
Geekbench 6 CPU Single: 2,883
Geekbench 6 GPU: 58,502
Geekbench 7 CPU Multi: 18,118
Geekbench 7 CPU Single: 2,587
Geekbench 7 GPU: 55,027
PCMark Overall: 9,429
Cinebench R23 CPU Multi (Threads): 20,252
Cinebench R23 CPU Single (Threads): 2,084
Cinebench 2026 CPU Multi (Threads): 4,835
Cinebench 2026 CPU Single (Threads): 505
Fire Strike Overall: 15,128
Fire Strike Graphics: 18,110
Fire Strike Physics: 29,411
Fire Strike Combined: 5,105
Time Spy Overall: 7,647
Time Spy Graphics: 7,049
Time Spy CPU: 14,738
Wild Life Overall: 44,322
Steel Nomad Overall: 1,601
Windows Experience Overall: 8.9
Procyon: 1,485

Getting started with the Khadas Mind Pro follows the usual process, and the machine arrives with Windows 11 Home preinstalled, alongside the Khadas Mind software, and the initial Windows setup takes around 10 to 15 minutes.

Aside from the usual process of accepting or declining Microsoft’s various additional services, there is very little to do before you’re up and running. Then the Khadas Mind application provides access to the machine’s power management and battery options, highlighting that this is far more than a normal mini PC.

While the modular ecosystem enables you to expand the system with a docking station or external GPU, even as the core unit when used as a standalone computer, the Mind Pro offers impressive performance, especially considering its small size.

I started the test by installing Microsoft Office; there was more than enough performance to run Word, Excel and PowerPoint all at the same time, alongside multiple browser windows. Everything remained fast and responsive, and even when adding Photoshop and Lightroom into the mix, there was little indication that the system was struggling.

As is now common, there’s also a dedicated Intel NPU, and that adds potential for supported local AI features, although cloud-based Microsoft Copilot functionality shouldn’t be confused with processing directly on the NPU. Opening up the task manager indicates when the NPU comes into play and often as a support role to cloud-based connections and some app integration.

After seeing that standard office applications were well within the machine's capabilities, I moved on to Adobe’s creative applications, and running these showed just how powerful this machine is even without the boost that the optional modules would obviously offer.

Lightroom Classic ran extremely smoothly, opening large catalogues, processing high-resolution photographs and exporting batches of images without difficulty. The internal 2TB SSD also performed well, keeping up with the demands of loading, editing and saving substantial image files.

Loading Photoshop and the performance was equally impressive. I worked with high-resolution photographs from the Canon EOS R5 C and Sony A7 IV, and the Mind Pro handled them without any noticeable problems. Opening Adobe Bridge, thumbnails appeared quickly, and I could select batches of up to 30 files and open them in Adobe Camera Raw for editing.

If you’re a professional photographer, then the small size of the machine and its portability make it a great option, especially when connected to a large monitor. I also found that the 64GB RAM provided substantial headroom, and with the Arc B390 integrated graphics, all image editing and processing was again well within the machine's abilities.

One of the features that I really liked was the ability to plug into my monitor and dock in the office and use the machine as I do with any Mini PC; then, when I headed home, I was able to leave several video projects and other documents open, pop the machine in my bag and then reconnect to another monitor in the home office, with all apps and documents still open and ready to go. The machine in that respect is just like a laptop without the screen and keyboard, of course. The benefit here is that it’s far smaller and lighter, so even more portable than a laptop, especially a laptop with this kind of power.

While the performance is good, I did have a connectivity issue. My test monitor uses USB-C rather than HDMI, so connecting the display and power supply occupied both Thunderbolt 4 ports. This left only the two USB-A 3.2 Gen 2 connections for external storage and accessories. While these are rated at 10Gbps, they’re considerably slower than the 40Gbps connection bandwidth offered by Thunderbolt 4.

For photographers and videographers using fast external SSDs, the optional Mini Dock would make a difference on this front.

When video editing, this machine again proved surprisingly capable and while I initially had reservations about how comfortably a mini PC of this size would handle 4K footage in Adobe Premiere Pro, editing a short five-minute project went well with few problems.

There were occasional slowdowns, particularly as the edit became more complex, but the overall workflow was better than I expected. After switching the video editing application to DaVinci Resolve, I found through the edits that the performance was similar. Straightforward 4K timelines were fine, although increasing the complexity of the edit resulted in additional slowdowns, particularly during rendering.

This is where an external GPU would potentially make a significant difference, and the Mind Graphics 2 looks like an interesting expansion option if you are planning to use this system for more intensive video production. Although, again, this does come at quite a financial cost.

Before moving on to the AI potential, I tried out a few games, and again I was surprised by the ability of the machine. I tested Cyberpunk 2077 and Indiana Jones and the Great Circle, and both titles loaded successfully, with no issues with the integrated CPU.

At medium graphics settings, gameplay was surprisingly good considering that all the graphics processing was being handled by that integrated Intel Arc B390. While I wouldn’t position the Mind Pro as a replacement for a dedicated gaming PC, the ability to run demanding modern titles on a computer this small is impressive.

The final test was to take a closer look at the AI potential of the Khadas Mind Pro, loading Lemonade AI and the Qwen model to see just how capable this compact mini PC was when running AI workloads locally.

I started by testing several AI models to code a website, generate images, organise files, recognise image content and handle general chatbot and text-generation tasks. The results were mixed, although the 64GB of RAM provided enough headroom to run some relatively large models locally.

One issue with the stability of the Lemonade and the AI process that I came up against was when switching between tasks and therefore LLM models. While there was enough memory to load and run individual models, running two larger models side by side seemed to cause issues and checking Task Manager, it showed the RAM maxing out. Instead, I found myself working through different stages, unloading one model before loading the next.

For example, when organising images, I first used PowerShell to create a CSV file containing a list of images before handing that information over to the Qwen model to help manage the files. I then switched to an image recognition model to identify image contents and add the relevant tagging information. For small batches, this worked; however, increasing the count from 20 to 100 images, something between the CPU, Lemonade, and the model kept falling over.

Through this process, I also found that switching models wasn't always straightforward. Rather than simply unloading one model and loading another, it was often more reliable to close Lemonade Server completely, restart it and then load the next model. Without restarting, the software could become confused, and the process was noticeably less reliable.

For a measurable test, I ran through a batch of 20 images, using the AI tools to assist with basic file organisation and CSV-based content management. Because the workflow involved switching between several models, it was difficult to calculate the total number of tokens used across the entire process.

However, during one of the main processing stages, the system recorded a time to first token of 9.08 seconds and generated 1,340 tokens in total. CPU utilisation reached 100%, while RAM usage fluctuated between 60% and 80%, showing the machine was capable of handling the workload but was also being pushed relatively hard.

The synthetic AI benchmark results backed up the practical testing, with the Khadas Mind Pro achieving a Procyon AI performance score of 1,485. This highlights that when configured correctly with the AI LLM management software, it will have impressive AI processing capabilities, although the real-world tests highlighted that available memory, model management and workflow optimisation are just as important as raw processing performance.

What impressed me most was that all of this was possible on such a small machine without an external GPU. The integrated graphics and 64GB of RAM provide a useful starting point for local AI development, experimentation and lighter production tasks, although experimentation with models and setup is at present a necessity.

That said, if you're planning to use the Mind Pro extensively for AI processing, the optional Mind Graphics 2 external GPU will be worth the upgrade.

• Performance: 4.5 / 5

Khadas Mind Pro: Final verdict

(Image credit: Alastair Jennings)

The Khadas Mind Pro is an exceptionally compact and powerful mini PC aimed at professionals, creatives and anyone who needs desktop performance across multiple locations without carrying a laptop.

What makes this small machine so versatile is the modular design, although I haven’t tested the additional modules in this review, focusing on the RAW power of the core unit.

As it stands, used in isolation, it’s surprisingly powerful and a small and stylish option for use in the office or studio; then, if you do need that boost, it can, at a price, be connected to the optional Mind Dock and Mind Graphics 2 external GPU, transforming it into a far more capable workstation for demanding creative applications, video editing and AI workloads.

Likewise at home, simply connect a monitor and keyboard, potentially alongside the Mind xPlay portable display, and you’re ready to continue working, or just the occasional check on email.

The solid metal construction gives the Mind Pro a premium feel, and its weight and small size make it easy to carry between locations. More importantly, the integrated standby battery means you can unplug the machine, slip it into your bag and reconnect elsewhere without closing applications or restarting Windows.

There are some issues, most notably the limited connectivity without a dock and the integrated GPU’s limitations under heavier workloads. However, the impressive CPU performance, 64GB of RAM and excellent portability do make this a highly versatile and desirable machine.

You are paying a premium for the quality of the build, the innovation, and the modular concept, but for professionals who regularly move between workspaces, the Mind Pro offers an attractive alternative to a conventional desktop or laptop.

Should I buy the Khadas Mind Pro?

(Image credit: Alastair Jennings)

Value

Expensive, but impressive versatility and upgrade potential.

4

Design

Exceptionally compact, solid construction with an innovative integrated battery.

4.5

Features

Excellent modular expandability, although onboard connectivity is limited.

4.5

Performance

Powerful CPU, plenty of RAM and impressive creative performance.

4.5

Overall

An innovative, portable workstation with genuine professional appeal.

4.5

Buy it if...

You want a powerful, portable mini PC

The Mind Pro combines impressive performance, compact metal construction, and an integrated standby battery, making it easy to transport between workspaces.

You like the idea of a modular workstation.

The Mind Dock expands connectivity, while the Mind Graphics 2 external GPU boosts performance for demanding creative workloads.

Don't buy it if...

You need plenty of ports without buying accessories.

Connectivity is limited, especially when using USB-C for both power and a display. The optional Mind Dock starts to feel like an essential purchase.

You need serious graphics performance out of the box.

The integrated GPU handles lighter workloads, but demanding video editing, gaming and AI tasks will benefit from an external GPU.

For more professional-grade desktops, we've tested the best business computers.

Banks are warning against giving too much access to AI agents.

• US civilian federal agencies must immediately remove DeepSeek from government systems
• Federal departments can no longer purchase DeepSeek products or affiliated services
• Military and intelligence organizations remain excluded

The US government has directed civilian federal departments to immediately eliminate DeepSeek artificial intelligence software and related services from their technology systems.

The Department of Homeland Security (DHS) issued the directive over concerns the Chinese company's technology could expose sensitive information to foreign intelligence operations.

The restriction also covers applications incorporating DeepSeek models, although privately owned computers and personal installations fall outside the federal government's directive.

DeepSeek faces removal over security concerns

The directive requires affected agencies to discontinue purchases involving DeepSeek technology, including offerings supplied by affiliates and external developers incorporating its models.

Government departments must also withdraw affected applications from approved purchasing catalogs, supplier inventories and online procurement platforms used for federal acquisitions.

Homeland Security Secretary Markwayne Mullin authorized the measure following a recommendation from the Federal Acquisition Security Council, which evaluates potential technology supply chain threats.

The decision relies on the Federal Acquisition Supply Chain Security Act of 2018, legislation allowing authorities to restrict technologies considered dangerous.

Officials cited Chinese legislation permitting state intelligence organizations to demand corporate information, raising concerns about government access to data processed through DeepSeek.

The department also identified privacy concerns involving account registration information, user monitoring and device fingerprinting associated with the company's hosted services.

Because conversational AI systems process submitted prompts, officials warned that government employees could inadvertently disclose confidential material through ordinary interactions with DeepSeek.

Information obtained through such exchanges could potentially support intelligence assessments involving American citizens, federal operations, or other sensitive government activities.

Security evaluations conducted by NowSecure also identified serious vulnerabilities affecting DeepSeek's iOS application, adding technical concerns to the government's assessment.

However, the directive concerns federal technology procurement and deployment rather than establishing a nationwide prohibition on individuals downloading or operating DeepSeek.

Agencies must comply, while exceptions remain possible

Military organizations and intelligence agencies are excluded because their procurement arrangements operate under separate authorities.

Departments seeking permission to retain affected technology must submit written exemption applications through the Cybersecurity and Infrastructure Security Agency (CISA).

Such applications must reach the relevant Homeland Security official within 60 days and explain why an exception is necessary for government operations.

Applicants must identify the relevant products, provide substantial operational or national security grounds, and propose safeguards addressing the identified security concerns.

While completing the removal of DeepSeek, agencies are expected to strengthen network surveillance, record suspicious information transfers, and investigate unusual activity involving affected systems.

CISA will assist departments with technical implementation, while the Homeland Security Secretary must reconsider the directive annually.

The notice appears in the Federal Register under docket DHS-2026-0628, establishing the administrative record supporting the government's exclusion and removal requirements.

Although officials describe several potential security exposures, the supplied notice does not establish that every DeepSeek deployment has transmitted confidential American information.

Its application to third-party implementations also raises practical questions about identifying affected software, particularly where developers operate DeepSeek models independently.

Get your Pip-Boy ready, Vault Dweller! Fallout Shelter: The Experiment is coming soon.

• A new report claims that Amazon is buying up more AI GPU racks than it can deploy in data centers
• An analyst links this report to Nvidia's GeForce GPU supply woes, theorizing that there's a connection
• The contention is that Nvidia is deliberately 'channel stuffing' or in other words, pushing an excess of AI GPUs to the detriment of gaming GPU chips

Nvidia's supply issues with its gaming GPUs are seemingly the result of a strategy to get AI giants to stock up on more graphics chips than they can handle, if fresh reports are to be believed.

You can't have failed to notice the stock issues around Nvidia's GeForce GPUs, and the spike in prices that has happened over the past few months, particularly for higher-end graphics cards.

Well, according to the musings of tech pundit and analyst Ed Zitron (via RespawnFirst) — who picked up on a report by Business Insider — the consumer GPU shortage is a direct result of Nvidia pushing AI graphics cards on the likes of Amazon.

Business Insider informs us that Amazon (or rather AWS) is buying more Nvidia GPUs than it can deploy. Apparently, racks of the AI graphics cards are being kept unused in holding warehouses until there's power available at the data centers where they'll be used (electricity capacity, not space, is the issue here).

Amazon is paying out huge sums for these GB300 racks from Nvidia, and according to an internal email that Business Insider got hold of, a new 'Early Rack Receipt' process (buying these GPUs upfront, before they're installed) saw some $5 billion worth of server racks paid for in August 2026. Seemingly $3 billion more are on order every month for the "foreseeable future" going by another email.

We're told one AWS team didn't have a "clear forecasted path" for getting the purchased racks installed — in other words, they're not even sure when these GPUs might be used.

Zitron's argument is based on a key phrase in the Business Insider report, where the author states that the reason for the switch in policy — Amazon only used to pay for racks when they were installed in a data center — is that "Amazon has to compete more for scarce resources as it jockeys with rivals for access to Nvidia GPUs and other AI components."

The conclusion Zitron draws is that this is basically indicating that Nvidia is making Amazon, and others buying up AI GPUs, compete with each other by agreeing to order a whole pile of them there and then — more than they possibly need — otherwise they won't get their allocation (presumably).

Analysis: GPU misery is clear enough to see

(Image credit: Shutterstock / aslysun)

In short, these AI GPUs aren't actually needed as such — not by Amazon in this quantity, as per the Business Insider report — but Nvidia is effectively strong-arming the company into buying the chips, or as Zitron puts it, this is "deliberate channel stuffing." Meaning Nvidia is forcing this AI demand artificially to the seeming detriment of GeForce GPUs, with lesser chip supply for those gaming cards causing price inflation and general misery for gamers.

Obviously we need to take this with a good deal of caution — that goes for the initial report from Business Insider, and doubly so with the theorizing that Zitron applies on top of that article.

There's no denying, however, that we have witnessed a frustrating shortage of consumer GPUs, and in particular Nvidia's high-end gaming graphics cards have become painfully thin on the ground — and pricey. The RTX 5090 is either unobtainable, save through third-party marketplace sellers trying to gouge you out of the best part of 10 grand (in the US), or it's hideously expensive to the tune of over £6,000 in the UK, for example.

What's worse is the RTX 5080 is now getting ridiculously pricey too, reaching £1,400 in the UK. And in the US on Newegg, there's now only one model for sale (direct from the retailer) — and it'll set you back $2,000. It's all getting silly, frankly, and with rumors that the RTX 5090 is being discontinued doing the rounds — speculation Nvidia has denied, I should make clear — it's currently a disappointingly gloomy picture for GPUs.

The expectation is that matters are only going to get worse for RAM, SSDs, and GPUs, as next year rolls into town — and that makes all the above theorizing seem a bit more plausible. It's certainly clear that Nvidia's priority is AI GPUs rather than GeForce graphics cards because of the profits involved with AI, but of course, Team Green is a business — so what do we expect? Prioritization is one thing, though, and the levelled accusation of the practice of 'channel stuffing' is entirely another.

Apple’s iPhone Duo brings iOS to a bigger tablet-like display, and these creative apps are already optimized to make the most of it.

• NATO's modernization command plans to buy six portable kits that can jam GPS, GLONASS, Galileo, and BeiDou, plus drone control and video links
• The blackout radius is adjustable from 500 m to 11 km, and each kit must deploy in 30 minutes, run for four hours on battery, and be controllable remotely over encrypted satellite links
• The purchase is a direct response to years of Russian-blamed GPS interference over the Baltic, including the 2025 jamming of the President of the European Commission's aircraft over Bulgaria

NATO is shopping for the ability to make satellite navigation disappear, on command, over an area the size of a small city.

The alliance's modernization command has signaled plans to buy six portable kits that can jam satellite navigation signals and the radio links drones use for control and video, so trainers can expose ships, aircrews, and drone operators to realistic interference during exercises.

For context, NATO wants to practice fighting in the same electromagnetic fog Russia has been generating over Europe's eastern flank for most of the past three years.

What is the NATO actually buying?

The buyer is Headquarters Supreme Allied Commander Transformation, the Norfolk, Virginia, command that runs NATO's training and modernization work. According to the notice, first reported by Defence Blog, the command intends to issue a request for proposals on October 26, 2026, with bids due November 30 and a resulting fixed-price contract expected on December 11.

The kits are framed as training gear, but the fine print leaves the door open for much more: they could also be used in operations. The requirements read like a wish list for electronic warfare.

Each kit must be light enough for one or two people to carry, ready to use within 30 minutes, and able to operate from a ship, rigid inflatable boat, uncrewed surface vessel, vehicle, tripod, or fixed ground site. It must also be able to target all satellites at once.

The system must be able to jam all four major satellite navigation constellations, America's GPS, Russia's GLONASS, Europe's Galileo, and China's BeiDou, across multiple frequency bands simultaneously, with each band individually selectable.

Operators must be able to set the area where satellite navigation is denied anywhere from 500 meters to 11 km (6 nautical miles), in increments of no more than 500 meters. That is a jamming bubble you can expand or shrink like a volume knob.

Because an invisible wall of interference near civilian aviation is potentially a genuine hazard, vendors must supply tables predicting how far the effect will reach at altitudes up to 10,000 ft (about 3,050 m), so exercise planners can keep it clear of civilian air traffic.

The drone-killing side is more complex but equally specific: beyond satellite signals, the kits must jam drone video links in the 5 to 6 GHz band and drone control links at 400 to 500 MHz and 2 to 3 GHz, along with several cellular and other bands. That maps neatly onto how small drones actually get beaten in the field. As the notice itself explains, interfering with a drone's navigation or control signals can push it off course, force it to hover, land, or fly home, or cut the operator's video feed.

NATO also does not want a system that only a signals-intelligence professional can run. The requirement calls for the kit to be operated from a laptop with an interface simple enough for non-specialists, with both omnidirectional and directional antennas, at least four hours of battery operation, and the ability to be controlled remotely anywhere in the world over encrypted satellite links, with a local override and a remote shutdown.

Not just a random wishlist for a GPS jammer

This procurement does not come out of nowhere. GPS interference has shifted from a rare curiosity to daily background noise along NATO's eastern edge. The alliance's own notice points to the pattern, noting that jamming has become a regular problem around the Baltic Sea, where airlines and shipping have reported widespread interference that several European governments have blamed on Russia.

The most politically pointed episode came in 2025. The NATO notice recalls that an aircraft carrying European Commission President Ursula von der Leyen lost GPS signals while approaching an airport in Bulgaria, an incident Bulgarian authorities attributed to suspected Russian interference.

An important distinction underlies all of this, and it explains why NATO wants crews to practice against the real thing rather than read about it. Jamming simply drowns the signal. Spoofing is nastier: it feeds a receiver a convincing but false position, so the crew never gets a clean warning that anything is wrong. As explained in our coverage of GNSS vulnerability, the real danger of spoofing is that the victim doesn't know what is happening and keeps acting on false information.

NATO argues that its training and evaluation teams need equipment that can create controlled interference on demand, so crews can be watched as they detect, report, and adapt.

For an alliance that has spent three years on the receiving end of Russia's electronic harassment, buying six suitcases that can do the same thing on cue is less about matching Moscow and more about making sure its own people are not blind when the signal drops. The kicker is obviously that the for-training component could just as easily be used aggressively by the same organization, that is, if the spec sheet holds.

Apple Watch Ultra 4: One minute review

The Apple Watch Ultra 4 isn't a radical aesthetic redesign, but it doesn't need to be. Apple's rugged smartwatch keeps the familiar 49mm titanium build, bright display, and customizable Action Button, while upgrades to its rear health sensors and internals make it a more capable companion for everyday wear and endurance workouts.

The biggest upgrade is arguably its battery life. Apple promises up to 50 hours of normal use, but I've easily managed two days with the Always-On display enabled and sleep tracking running overnight. That's seriously impressive for a fully featured smartwatch, even if some lower-power rivals like the best Garmin watches can last longer.

The new S11 chip keeps everything feeling responsive, with apps opening quickly and watchOS 27 running smoothly. Faster charging is another welcome addition, letting me replenish around 30% of the battery in just 15 to 20 minutes.

Health and fitness tracking also gets a meaningful boost. The new Health Sensing System improves measurements, and Readiness Score is one of my favorite additions to watchOS 27. It gives you a useful snapshot of how prepared you might be for the day, without requiring another subscription.

It also adds plenty of smaller quality-of-life improvements, including real-time step-count complications. Siri AI also brings more intelligence to the wrist, with features like Shazam and Sound Recognition. Siri Rewind and Live Recap have yet to arrive and prove themselves.

The downside is that the Ultra 4 looks and feels much like its predecessor, so Ultra 3 owners may not find enough here to justify an immediate upgrade. And while the battery life is excellent, the Ultra 4 still can't match every rival in sheer endurance.

The Apple Watch Ultra 4 costs $799 / £749 / AU$1,349 and is available now in Natural Titanium or Black Titanium. Both finishes come with Wi-Fi and cellular connectivity, so there's no need to choose between separate Bluetooth-only and cellular models, as you can with the Apple Watch Series 12.

Ultimately, the Apple Watch Ultra 4 is a refined, reliable smartwatch that builds on an already excellent foundation. Its multi-day battery life, capable fitness tracking, and smoother watchOS experience make it a compelling choice for endurance athletes and Apple fans who want a rugged watch that does it all.

Apple Watch Ultra 4: Video reviewApple Watch Ultra 4: Specs Table

Component

Apple Watch Ultra 4

Price

$799 / £749 / AU$1,349

Dimension

49 x 44 x 12mm

Weight

63g (without band)

Case/bezel

Grad 5 Titanium

Display

49mm Always-On Retina LTPO3 OLED display (up to 3,000 nits)

Processor

Apple-made S11 chip

GPS

Dual‑Frequency GPS (GPS, Galileo, QZSS, and BeiDou)

Battery life

Up to 50 hours with normal use

Connections

4G LTE, Satellite SOS, Wi‑Fi 4 (802.11n), Bluetooth 5.3

Water resistant

Water resistance 100m

Apple Watch Ultra 4: Price and availability

• $799 / £749 / AU$1,349
• No increase from last year
• You’re paying for an LTE model as usual

The Apple Watch Ultra 4 costs $799 / £749 / AU$1,349, and comes in Natural Titanium or Black Titanium, with both finishes including Wi-Fi and cellular connectivity as standard. This means you’re paying for your phone’s ability to add to a data plan whether you want to or not.

The watch is more expensive than other Apple Watches, which is standard for the Ultra series, but crucially, it’s the same price as the Apple Watch Ultra 3, with no annual price increase. In an era of an electronics components crisis, this is great news. It’s also absolutely stuffed with features as usual, with a premium build quality that, together, justifies the high price.

• Value score: 4.5/5

Apple Watch Ultra 4: Design

(Image credit: Future/Jacob Krol)

• Redesigned sensor array
• Identical display and case
• Useful programmable Action button

Like the Apple Watch Ultra 1, 2, and 3 before it, the Ultra 4 stands on its own in Apple's smartwatch lineup, offering one option: a squircle 49mm screen in a thicker, much more rugged design that's ready for the road — and truly any road. It still comes in Black or Natural Titanium and looks pretty much the same as previous iterations on your wrist. Still, the one-size-fits-all device is an issue for smaller wrists.

I've been testing the silver Natural Titanium version, and it's been a welcome change after rocking the Black Ultra 3 for well over a year. I like the contrast of the silver rim around the 49mm Retina display, and especially how the ‘Ultra Orange’ Action Button on the left side pops against the silver case.

The Action Button remains a nice addition for easily running a customized Siri Shortcut or quickly starting a workout, and the 49mm Always-On Retina OLED LTPO3 display still looks great in pretty much any condition, whether that's under fluorescent lights, in the gym, or outside on a very sunny day.

The Action Button sits alongside the speaker and microphone on the left side, while the rugged Digital Crown (with its orange ring) and the main side button live on the right. The top and bottom of the Apple Watch are still where the bands connect, and there's no change to the connector — fingers crossed Apple never changes that, as it’s exemplary.

The main design changes, aside from the new bands, are on the rear. Apple is rolling out a new Health Sensing System to the Ultra 4 and Series 12, which looks a little like a complex sundial on the back of the case. It features a new wider layout for electrodes and diodes to cover a larger surface area on the wrist, which lets the Apple Watch take more accurate health measurements — particularly when tracking heart rate.

This is also where the magnetic Apple Watch charger attaches, and the box includes an updated charging puck. It still connects to a wall adapter or another power source via USB-C, with improvements to charging speed. And while the Series 12 is still a stunner and the Ultra 3 could run laps around plenty of other watches, there's a larger battery inside the Ultra 4. I'll share more on this in the Performance section, but battery life is seriously impressive, surpassing its 50 listed hours.

Aside from the Health Sensing System on the back, though, much of the design here is Apple sticking with what has made the Ultra successful since its launch. It's an ultra-rugged, road-ready Apple Watch that's likely a necessity for endurance athletes, but its distinctive build also appeals to those who simply want something different from the standard Apple Watch.

• Design score: 5/5

Apple Watch Ultra 4: Features

(Image credit: Future/Jacob Krol)

• Readiness score is excellent
• Siri AI features to come
• Improved vitals app

With the design mostly unchanged, much of what's new with the Apple Watch Ultra 4 comes from watchOS 27 and its health and activity features. I'm starting with the latter, as it's truly been the highlight of my testing.

I've worn an Apple Watch since it first launched in the spring of 2015. My original aluminum model arrived on launch day, and I was admittedly hooked on closing my rings. Now, after more than a month with the Ultra 4 — alongside the Apple Watch Series 12 — my new favorite feature is Readiness Score. It received a big round of applause at the launch event, and for good reason: it fills a gap in Apple's health feature set compared with the competition.

Many other wearables, including smartwatches and smart rings, offer ways to contextualize your data and help you understand what it means. Most notably, the Oura Ring provides a daily score that gives you an idea of what you're ready to take on. It's not medical advice, of course, but it can help you make sense of the data it collects. However, that feature sits behind a paywall, as do comparable offerings from Whoop, although Google Health, Garmin, and others provide it for free after you purchase a tracker.

(Image credit: Future/Jacob Krol)

Apple's take on a Readiness score is sometimes less elaborate, but it's free if you have a compatible new Apple Watch and enough data for it to work. Based on your sleep score and previous day's activity, you'll get a score from 0 to 10 tailored to you, alongside a matching phrase. It's presented as a curved line or half-moon graphic. Sometimes, I was greeted with a blue 7, marked "Ready," with the explanation: "You had a good night's sleep last night, and all your other metrics are typical. It looks like you're rested and ready for today."

Jacob Krol/FutureJacob Krol/Future

Below is a smaller breakdown of Fitness, Vitals, and Sleep Score. These can change throughout the day depending on your activity rings and whether you log workouts. When you open the Health or Fitness app on your connected iPhone, there's more data to explore, too.

It's become my favorite feature of watchOS 27 because of how it integrates with the Activity Rings experience, while building on Vitals. Rather than opening another app each morning and scrolling through a series of statistics, you're presented with a single score that feels actionable. Compared with other health wearables, it's also nice that Apple isn't locking it behind a monthly subscription. You do need a compatible recent Apple Watch, so there's an upfront cost, but you won't have to keep paying to use the feature.

(Image credit: Future/Jacob Krol)

watchOS 27 also brings Siri AI to the wrist, provided your Apple Watch is paired with a compatible iPhone. While some more extensive requests can take an extra beat, the experience has been smooth, whether I raise my wrist to talk or simply say "Hey Siri." You can also ask for relevant information about your activity directly from your wrist.

The new Siri Modular watch face certainly spotlights this feature, placing Siri Suggestions below the time and above other complications. At times, it might suggest identifying a song with Shazam, remind you of your upcoming bedtime, provide the weather forecast, or offer other contextually relevant information.

Thanks to the new Health Sensing System, which supports activity and heart-rate tracking, blood oxygen measurements, hypertension notifications, and sleep tracking, Apple says heart-rate measurements are more accurate than ever. The Ultra 4 also introduces heart-rate variability (HRV) to the experience. You can add a complication that updates your heart rate in real time, as well as one for your step count.

• Features score: 4.5/5

Apple Watch Ultra 4: Performance

(Image credit: Future/Jacob Krol)

• Battery life performs as described
• Accurate against HRMs and competitor wearables
• Not yet able to test Audio Intelligence features

Apple's new S11 chip is doing a lot of the heavy lifting in the Apple Watch Ultra 4, and it's a noticeable part of the experience. Apps open quickly, navigating watchOS 27 feels incredibly responsive, and everything runs smoothly. Suffice to say, it’s what you expect from a $799 smartwatch in 2026, and while it doesn’t run laps around the Ultra 3, it has its moments where it has a bit more pep in its step.

The S11 also features a Secure Enclave and a separate always-on processor, helping balance performance with efficiency. That last part matters even more with the Ultra 4's physically larger battery.

Apple promises up to 50 hours of battery life with normal use, up to 84 hours in Low Power Mode, and up to 45 hours in Max Extended Workout Mode. In my testing, I've easily managed two and a half days on a charge with the Always-On display enabled and sleep tracking running overnight. That's seriously impressive for a fully featured smartwatch, even if some competing watches can last longer. The Ultra 4 won't win every battery-life comparison, but getting multiple days of use without sacrificing those everyday features is a big deal.

When you do eventually run low, faster charging makes it easier to top up. I've been able to get about 30% battery back in 15 to 20 minutes, which is handy if you forgot to charge overnight or want to give the watch extra juice before heading out. It doesn't eliminate the need to charge, of course, but combined with that extended runtime, it makes battery anxiety much less of a concern.

The S11 chip and new Health Sensing System also help power features like Readiness Score, which has quickly become one of my favorite additions to watchOS 27. But the chip is also laying the groundwork for Apple's forthcoming Siri Rewind and Live Recap features.

I haven't been able to test either just yet, and while I still expect them to arrive in the coming months, Apple hasn't confirmed exactly when they'll ship. If those features are the main reason you're considering the Ultra 4 — or the Apple Watch Series 12, for that matter — I'd wait until Apple provides a firm release date and we've had the chance to put them through their paces.

There are other smaller improvements that make the Ultra 4 nicer to use, too. Shazam and Sound Recognition are handy additions to have on your wrist, letting you identify music or recognize certain sounds without needing to reach for your iPhone. Shazam still struggles with some live music, though, so don't expect it to identify every song you hear at a concert.

(Image credit: Future/Jacob Krol)

Then there's the pedometer. It's a little wild that it's taken this long, but you can finally add a complication that updates your step count in real time right on your watch face. It's a simple quality-of-life improvement, but one that makes the Ultra 4 more useful at a glance.

In my testing across indoor cycling, outdoor walks, yoga, and indoor walks, the overall workout statistics and heart-rate readings have been in line with those from other devices, including the Oura Ring 5, Pixel Watch 5, Fitbit Air, and Apple Watch Series 12. Against an electrical heart rate monitor during a 5km run, the Apple Watch Ultra 4’s average heart rate reading was just 2 bpm away from the monitor’s. That’s well within the margin for error and just 1 bpm away from the Garmin Fenix 9 Pro’s readings during the same run.

Taken together, these aren't necessarily the sorts of changes that will make an Ultra 3 owner immediately rush out to upgrade. Instead, the S11 chip delivers a smoother, more efficient experience, while the larger battery makes it easier to take advantage of everything the watch has to offer. And with two and a half days of real-world battery life, that's a combination that's hard to beat.

• Performance score: 5/5

Should you buy the Apple Watch Ultra 4?Buy it if..

You need long battery life

The Apple Watch Ultra 4 can last for days, performing for up to two and a half days of real-world use in my testing.

You want Apple’s latest health features

The Readiness Score and Health Sensing System are only available on the current gen watches.

Don’t buy it if..

You don’t need Apple’s AI features

If you don't want the Audio Intelligence features or the Readiness Score, the experience is pretty close to Ultra 3.

Also consider

Apple Watch Series 12

Apple's cheaper, less powerful wearable.

Garmin Fenix 9 Pro

The Apple Watch Ultra 4's biggest competitor in the premium adventure watch space.

Read our full Garmin Fenix 9 Pro review

How we tested

We tested the Apple Watch Ultra 4 by sleeping, working out, and living with the watch during a testing period comprising multiple weeks, staff members, and a review unit. We compared the watch's readings against competitor wearables, including an electrical heart rate monitor, drained the battery, and used as many available features as possible.

First reviewed October 2026.

• Craft is Canva's first publicly available AI image generation model
• It was trained on existing templates so it understands layers
• Developers can access the model via an API

Canva has launched its first-ever publicly available AI image generation model, and it's already proving to be hugely successful.

Looking to take on the likes of industry giants such as OpenAI and Google, 'Craft', its image generating model, debuted at number four on DesignArena's leaderboard, falling short of three leaders from OpenAI but coming in better than a number of other GPT, Grok and Gemini models – including Nano Banana 2.1.

With such high performance, Canva is proudly boasting that Canva Research, its internal research division, is now the second-highest performing AI lab in terms of image generation models behind OpenAI, which is a big deal.

Canva Craft

The launch follows significant investment from the company into its own internal lab, which has tripled in size over the past year.

But besides placing fourth overall for image generation, Canva Craft actually came in first place of 87 models for people and portraits, first of 85 in landscapes and nature, and second for abstract designs and patterns.

The company explained that Craft workers by planning an image's composition before generating a result, which means it decides where individual elements should appear, how composition should be structured, where words need to appear and how the entire thing can come together in a cohesive, readable way. Something that's especially important for the type of work Craft is likely to conduct, particularly around generating marketing materials and social media graphics.

The model's success likely stems from training – it was trained on millions of human-made templates in Canva's library, so it understands layout, visual hierarchy and typography unlike models that were trained on flat images.

Craft came about because Canva is a firm believer that developing its own in-house models gives it greater control over the economics of generative AI, rather than having to rely on third parties, and we're already seeing how unpredictable agentic AI costing is impacting enterprise-grade users.

But Craft isn't being locked into the Canva ecosystem, because the company is also making it available via an API to let developers integrate the model into other platforms.

As for the next steps, Canva already promises it's working on future generations of Craft with faster and more affordable options on the cards.

• Audi is the first manufacturer to introduce Spotify’s ground-up app for cars
• New platform takes advantage of passenger screens
• A more comprehensive interface brings with it more features

Audi has announced that it is the first automaker to introduce Spotify’s dedicated in-car application, which has been co-developed by the music streaming service and the Volkswagen Group’s Cariad automotive software company.

Yes, a dedicated Spotify app already exists for cars running Android Automotive, such as those from Renault, Volvo and more, but that is a template-based solution, which uses standardized layouts and controls. It also keeps a tight lid on the myriad Spotify features that have been introduced over the past few years.

Jonathan Tarlton, Automotive Business Development Lead at Spotify, told The Verge that the company was “drawing all the pixels” during the development stages, going on to state that it owns the user experience and that it is “100 percent Spotify”.

Unlike the limited Google Automotive OS app, as well as the smartphone projections experienced by both Android and Apple users in Android Auto and CarPlay, the software debuting on Audi models uses a layout that feels more like the desktop experience, making the most of massive infotainment real estate.

When passenger screens are specified, for example, front-seat passengers can watch Spotify music videos while the vehicle is moving, they can also see real-time lyrics and scan QR codes to join listening parties or create ad-hoc road-trip playlists.

The large central display in Audi’s current line-up can do all of those things, but for safety reasons, this functionality is blocked as soon as the wheels start rolling.

Analysis: A slicker Spotify experience

(Image credit: Audi/Spotify)

Audi says that this is the first time that music videos and real-time lyrics, as well as popular features like Spotify Jam and Spotify DJ have been featured on a native app, but doesn’t mention if or when it will be arriving on other models within the Volkswagen Group or further afield.

Judging from the imagery Audi has provided, the native app looks a lot cleaner than anything currently found on Android Automotive OS, with a layout that closely mimics the set-up found on laptops and desktops.

I don’t now about you, but I often get frustrated with the limited capabilities of the Apple CarPlay projection, as well as the clunky music search functionality. I often end up listening to the same old playlists on repeat because of this.

In contrast, Audi says that its AI-powered voice assistant works seamlessly to control the music while driving, with the “Hey Audi” command used to switch playlists or search for individual tracks.

Spotify users must have an account and use a smartphone to sign in, but Audi says the new Spotify app can be downloaded from the Audi Application Store free of charge.

• Critical PHP2Shell flaws also impact internet-exposed NAS devices and PHP-based Docker workloads
• Major vendors are still releasing patches, leaving some Synology, QNAP, and other systems exposed
• Researchers urge immediate updates, removal of public exposure, and VPN-only remote access

Websites and web servers are apparently not the only products affected by the php2shell vulnerability - Network Attached Storage (NAS) devices are at risk, too - and to make matters worse, some of the most popular manufacturers have not yet released a patch, despite the flaws being actively abused in the wild.

Research by Apple Insider noted that the flaw also affects software maintained by NAS manufacturers, as well as applications running independently in Docker containers, with NAS devices that are reachable through the public internet also at risk.

What's particularly concerning is that some of the biggest NAS manufacturers have said to be working on a fix, but the work is yet to be done. Synology, for example, released a new security advisory saying PHP versions 7.3 - 8.4, Mail Station, and Synology Calendar are all affected and adding that it released updated PHP packages to address the flaws. Those running Synology NAS products should head over to the company’s Package Center and download archive and look for updated PHP 8.2 and PHP 8.4 packages.

Twofold risk

The company also stressed that there is a difference between installed PHP packages and PHP components within DiskStation manager, and added that its built-in PHP isn’t exposed to the public internet, therefore most likely not at risk.

Another major NAS manufacturer, QNAP, confirmed a wide range of affected products, including QTS 5.2.11 and older, QTS 6.1.9, QuTS hero and QuTScloud, and Note Station 3. It said it was currently working on a fix, but did not give a timeline on when the patch might become available. So far, there is no evidence of in-the-wild attacks, but the company does recommend removing the devices from the open internet, disabling unnecessary port forwarding, UPnP, and remote-access services.

Other manufacturers are also currently working on a fix, including TerraMaster, Ugreen, and Asustor, Apple Insider found. “NAS owners should check their manufacturer's security notices and available updates for their specific models,” the publication said.

For NAS owners, the risk is twofold, since PHP is found in both the operating system and Docker containers.

Those running PHP-based services in Docker should check if their container images are running affected versions of PHP, and update them as soon as humanly possible.

There is also the problem of running older devices that have reached end-of-life and are, as such, no longer supported with patches. These, security experts agree, should be removed from public access. Unsolicited inbound connections should be blocked, and the device should be kept accessible only locally. Remote access should be achieved through a secure VPN, instead of forwarding NAS management ports.

P2PShell?

PHP2Shell is a name for a pair of vulnerabilities - CVE-2026-109618 and CVE-2026-109619 - that were recently discovered in the PHP programming language. They are considered critical and allow malicious actors to take over websites, steal data, deploy malware, and more. However, the story that has so far revolved around internet sites and web apps is now expanding into NAS devices, too.

PHP2Shell is shaping up to be the biggest security incident of the year, and possibly longer, arguably rivaled only by the Spectre/Meltdown flaws from eight years ago.

The pair of vulnerabilities - both critical in severity and both allowing for remote code execution - affects the vast majority of websites and web apps in existence today.

Knowledge of the flaws leaked four days before the patches were released, according to Amazon, and the number of exploitation attempts went from a few hundred 10 hours before publication, to millions today.

However, most of these attempts seem to have been unsuccessful. Analysis is ongoing as to why that is the case, with some researchers thinking code execution needs to get past ASLR, which is a difficult thing to do remotely. However, breaking in only seems to be a matter of time, which is why the security community urges manufacturers to release a fix as soon as possible, and owners to apply them without delay.