AI PCs were the whole story at Microsoft’s Windows and Surface event in San Francisco on Wednesday 7 October 2026. Satya Nadella and Nvidia’s Jensen Huang shared a stage to open pre-orders for the Surface Laptop Ultra, a 15-inch machine built on Nvidia’s RTX Spark chip, from $2,599. Microsoft also showed AI models that previously needed the cloud running on the laptop itself, and released Microsoft Execution Containers, a sandbox that stops AI agents touching data they have not been given.
Reuters framed the day as a direct challenge to Apple, which spent September pitching its new Mac mini and Mac Studio to companies as a cheaper alternative to renting data centres. Both companies are now selling AI PCs on the same idea: run more AI on a machine you have already paid for, and stop paying the cloud for every token. Microsoft even launched a trade-in offer aimed squarely at MacBook Pro owners.
This follow-up to our preview of the Surface event covers what was announced, how the new AI PCs compare with Apple’s machines on price and performance, why security got more stage time than speed, and what UK businesses should do next. For the chip itself, see our earlier explainer on the RTX Spark superchip.
Table of contents
- What Microsoft Announced for AI PCs on 7 October
- Why Microsoft Is Aiming Its AI PCs Straight at Apple
- Microsoft’s Performance Claims for AI PCs Against the Mac
- The Price Problem Facing AI PCs
- Hybrid Intelligence: How AI PCs Split Work With the Cloud
- Security Is the Real Selling Point for AI PCs
- Copilot+ Lives On as AI PCs Split Into Tiers
- Nvidia’s Stake: AI PCs as a Way Into Intel and AMD’s Market
- How Our Surface Event Preview Scored
- What AI PCs Mean for UK Businesses
- AI PCs FAQ
- References and Further Reading
What Microsoft Announced for AI PCs on 7 October
Nadella called the event “the birth of the next generation of Windows agentic platforms”, according to Barron’s. Behind the slogan were four concrete things: new hardware, local models, a security layer for agents, and a new Copilot that works on the PC as well as in the cloud. The table collects the main announcements.
| Announcement | What it is | When |
|---|---|---|
| Surface Laptop Ultra | 15-inch laptop on Nvidia RTX Spark, up to 128GB unified memory | Pre-order now, from $2,599; ships 16 October |
| Surface RTX Spark Dev Box | Small desktop for developers, 128GB, same chip | Pre-order now, $5,999; US only, ships November |
| Microsoft Execution Containers (MXC) | Policy-based sandbox for AI agents | Generally available on Windows 11 |
| Local models | MAI Code 1.1 Flash, a new Nemotron, DeepSeek V4 Flash | Rolling out on RTX Spark PCs |
| Copilot hybrid intelligence | Local files, local actions and local models for Copilot | Copilot+ PCs, “in the coming months” |
| GitHub HydraFusion on Windows | Routes coding tasks between local and cloud models | Experimental preview later in October |
| DGX Station for Windows | Deskside supercomputer on Nvidia GB300 | Later this year |
| Muse for Windows | Meta’s personal AI agent as a native Windows app | “Coming soon” |
The new AI PCs
The Surface Laptop Ultra pairs an Nvidia Grace CPU with up to 20 cores and a Blackwell GPU with up to 6,144 cores on one chip, with up to 128GB of memory shared between them. Microsoft says it can run AI models of more than 120 billion parameters locally and reach up to one petaflop of AI performance, a figure its footnote says relies on Nvidia’s sparsity feature. Five other PC makers opened pre-orders for RTX Spark AI PCs on the same day.
The software behind the AI PCs
The bigger change for AI PCs is in Windows. Pavan Davuluri, who runs Windows and devices, described a platform where agents “run locally when it makes sense, reach the cloud when they need to”. Microsoft calls this hybrid intelligence, and it now sits under Copilot, GitHub’s coding tools and the new agent sandbox.
The guests
Huang did not just appear as a supplier. He told the audience “we have reinvented the computer as we know it”, and said the new containers were essential for agents. Microsoft also confirmed that Meta’s Muse agent is coming to Windows, which gives it a consumer-facing name to sit alongside the developer tools.
Why Microsoft Is Aiming Its AI PCs Straight at Apple
Reuters reporter Stephen Nellis, who covered both companies’ launches, put the rivalry plainly. Microsoft’s move is “a bet that some work that currently happens in its costly Azure cloud computing data centers can shift to high-powered Windows machines”, where customers pay for the hardware. And it is “a market opportunity rival Apple is also chasing with new Mac computers”.
Apple moved first in September
Apple’s new Mac mini and Mac Studio went on sale on 22 September. The Mac mini starts at $899 with the M6 chip, and the Mac Studio starts at $2,499 with the M5 Max or $5,499 with the M5 Ultra, according to Apple. A Mac Studio with 512GB of memory follows in late October. Reuters reported that fully specified Mac Studios “can cost nearly $20,000”.
Apple’s pitch to companies was unusually direct. “There’s no cost per token. You’re just using the machine again and again,” Apple hardware chief Johny Srouji told Reuters. At its launch, Apple showed four Mac Studios linked over Thunderbolt running a trillion-parameter model from a single wall socket. Our Mac Studio M5 Ultra review covers that machine in detail.
Microsoft is defending home ground
Apple is the challenger in offices, not Microsoft. Reuters quoted IDC’s Linn Huang putting Apple at about 4.6% of the enterprise desktop market against 91.3% for Windows. The chart shows the split, with the remainder worked out from those two figures.
Enterprise desktop market share, IDC figures quoted by Reuters (September 2026)
That gap explains the tone. Microsoft does not need to win offices from Apple; it needs to stop Apple’s AI argument from taking the most valuable buyers, the developers and AI teams who set what the rest of a company uses. Those are exactly the people the new AI PCs are priced for.
The trade-in offer
Microsoft made the target explicit. Its Surface blog carries a section titled “Break up with your MacBook”, offering up to $1,000 cash back when a buyer trades in an eligible MacBook Pro for a Surface Laptop Ultra. The small print limits it to the online Microsoft Store in the United States and Canada, between 7 October and 23 November 2026, with trade-ins assessed by Teladvance.
Apple’s growing PC share
Apple is also growing while Windows makers shrink. IDC’s preliminary data for the second quarter of 2026, as reported by Tech Times, showed every major Windows PC maker shipping fewer machines than a year earlier while Apple grew 10.1%, helped by its low-cost MacBook Neo.
| Vendor | Q2 2026 units | Share | Change on a year |
|---|---|---|---|
| Lenovo | 16.6m | 24.4% | -2.1% |
| HP Inc. | 13.0m | 19.1% | -9.0% |
| Dell | 9.3m | 13.6% | -5.0% |
| Apple | 6.7m | 9.9% | +10.1% |
| Asus | 5.0m | 7.4% | +0.2% |
| All PCs | 68.2m | 100% | -4.9% |
Those are total PC shipments, not AI PCs, and the MacBook Neo is a cheap laptop rather than an AI workstation. But they show why Microsoft is in a hurry: Apple now ships roughly one PC in ten, and it has a simple story about AI that does not depend on dozens of hardware partners.
Microsoft's Performance Claims for AI PCs Against the Mac
Microsoft published three benchmark claims comparing RTX Spark AI PCs with a 16-inch MacBook Pro using Apple’s M5 Pro chip and 64GB of memory. The chart shows them as published.
Claimed speed-up of RTX Spark Windows PCs over a 64GB M5 Pro MacBook Pro (vendor testing, as a share of the largest claim)
Read the footnotes
The footnotes matter more than the headline numbers. All three tests used pre-production AI PCs with 64GB of memory, ran in September 2026, and were carried out by Microsoft or Nvidia rather than an independent lab. The token test used llama.cpp with a 27-billion-parameter Qwen model and a fixed 8,192-token prompt; the image and video tests used ComfyUI with Nvidia’s own NVFP4 number format.
What the numbers do and do not show
Time to first token measures how quickly a model starts answering a long prompt, which favours a big GPU. It says nothing about how fast the rest of the answer arrives, battery life, or the M5 Max Macs at the top of Apple’s range. Microsoft’s separate display claim, that the screen reaches up to 25% brighter peak HDR than an M5 Pro MacBook Pro, is also measured against Apple’s published specifications rather than a side-by-side test.
The leak that came before
Before launch, a TechPowerUp forum member tested a pre-release Surface Laptop Ultra with 24GB of memory and found it trailing Apple’s M5 Pro in several general benchmarks, as we reported in our preview. That was unfinished hardware and software, but it is a reason to wait for independent reviews before treating the vendor numbers for these AI PCs as settled.
The Price Problem Facing AI PCs
The awkward fact of the day was price. Reuters said one of the key challenges facing Microsoft and Nvidia is “the steep pricing of these devices after a memory-chip crunch drove up costs”. CNET listed the Surface Laptop Ultra configurations as pre-orders opened; its figures are rounded to the nearest $100 except where Microsoft gave an exact price.
| Chip | Memory and storage | US price |
|---|---|---|
| 18-core CPU, 5,120-core GPU | 24GB, 512GB | $2,599 |
| 18-core CPU, 5,120-core GPU | 32GB, 1TB | About $3,300 |
| 20-core CPU, 6,144-core GPU | 32GB | About $3,600 |
| 20-core CPU, 6,144-core GPU | 48GB | About $4,000 |
| 20-core CPU, 6,144-core GPU | 64GB | About $4,300 |
| 20-core CPU, 6,144-core GPU | 128GB, 1TB | $5,899 (listed out of stock at launch) |
How the prices compare with Apple
Reuters set those prices against Apple’s. The entry Surface costs more than Apple’s $1,999 base MacBook Pro but comes with 24GB of memory against 16GB. At the top, a MacBook Pro with a 40-core GPU, 128GB and 2TB of storage costs $6,700, against $5,899 for the 128GB Surface with 1TB. Reuters stressed the machines “are not directly comparable”.
| Machine | Memory | US price |
|---|---|---|
| MacBook Pro, base model | 16GB | $1,999 |
| Surface Laptop Ultra, base model | 24GB | $2,599 |
| Surface Laptop Ultra, top model | 128GB | $5,899 |
| MacBook Pro, 40-core GPU, 2TB | 128GB | $6,700 |
| Mac Studio, M5 Ultra (starting price) | Base configuration | $5,499 |
| Surface RTX Spark Dev Box | 128GB | $5,999 |
| Nvidia DGX Spark, new 64GB model (partners) | 64GB | About $4,999 |
| Nvidia DGX Spark, 128GB | 128GB | $6,950 |
Price per gigabyte of memory
For AI PCs, memory decides which models a machine can run at all, so dollars per gigabyte is a fair rough guide. The chart divides each price in the table by its memory. It ignores storage, screens and CPU differences, so treat it as a sense check rather than a verdict.
US price divided by unified memory, dollars per GB (as a share of the highest figure)
On that rough measure the top Surface works out cheapest per gigabyte of the machines in the table, and the entry models are where the money buys least memory. But the 128GB model was out of stock on day one, and the AI PCs most buyers can afford have 24GB to 64GB, which caps the size of model they can run.
The memory crunch behind the prices
Memory is the reason AI PCs cost so much. Tech Times, citing SigmaIntel, reported that LPDDR5X laptop memory rose 89% in price in the second quarter of 2026 alone, as chipmakers moved capacity to the high-bandwidth memory that AI data centres buy. Nvidia has raised its 128GB DGX Spark to $6,950, nearly 75% above its launch price, The Register reported on 2 October.
An analyst’s warning
Moor Insights & Strategy analyst Anshel Sag summed it up for Reuters. Two years ago, “the software wasn’t ready, but the hardware was. Now the software is ready and the hardware is too expensive to actually run it locally.” His conclusion: “only the people who have the budget can really afford to run AI locally.” When Microsoft first pitched AI PCs in 2024 as a way to cut cloud costs, most of the machines cost under $2,000.
Hybrid Intelligence: How AI PCs Split Work With the Cloud
Hybrid intelligence is Microsoft’s answer to the cost problem: do the cheap, private or repetitive work on the PC, and send only the hardest tasks to the cloud. Microsoft’s own description is that customers’ needs “are outpacing what their cloud budgets can support”. Several pieces were announced to make that work on AI PCs.
Microsoft’s coding model, on the laptop
MAI Code 1.1 Flash, Microsoft’s own coding model first shown at Build, now runs locally on AI PCs. It has 137 billion parameters in total but uses only 6.8 billion for each step. Microsoft says it compressed the model to 3-bit precision, cutting its size by nearly 80% while keeping coding quality and a 256,000-token context window. This is the “AI coding model that can run directly on personal computers” in the Reuters report.
Open models: Nemotron and DeepSeek
Microsoft is also bringing an upcoming Nvidia Nemotron model of more than 70 billion parameters, compressed to use just over 20GB of memory, and DeepSeek V4 Flash, a 284-billion-parameter model. Davuluri told Reuters a version of DeepSeek’s V4 can run on machines with at least 60GB of memory and outperform OpenAI’s GPT-5 on some coding and reasoning tasks. Our DeepSeek V4 guide covers the model family.
Routing work between laptop and cloud
GitHub’s HydraFusion, which already picks the right cloud model for each coding task, will be able to use models running on the PC. It arrives in the GitHub Copilot app, the Copilot command-line tool and Visual Studio Code as an experimental preview later in October. Microsoft is also adding llama.cpp support to Windows ML, so developers can try new open models quickly.
Copilot gets local context, actions and models
On Copilot+ PCs, Copilot will be able to read relevant files and recent activity with permission, take actions such as organising files or running diagnostics, and use local models. Copilot chief Jacob Andreou told Reuters: “Copilot will still use the cloud for the hardest tasks, but for times when cost or privacy matter more, it can delegate down to local models that run directly on your computer.” Microsoft says these features begin rolling out “in the coming months”.
Why the cost argument matters
The pitch for AI PCs only works if local hardware is cheaper over its life than the cloud tokens it replaces. That is the same argument Apple made in September, and it is the reason Nadella talks about “unmetered intelligence”. Our earlier coverage of the new Copilot with Code and Autopilot explains the cloud side of that bill.
Security Is the Real Selling Point for AI PCs
The first ten minutes of the keynote were almost all about agent safety, CNET reported. That reflects a difficult year for cybersecurity: AI agents from the biggest labs have broken into commercial and government systems, including the attack on Hugging Face that preceded Nvidia’s $13 billion purchase of the company. “We needed to make the desktop the most secure place for agents to execute,” Nadella said.
How Microsoft Execution Containers work
MXC, now generally available on Windows 11, puts an agent inside a boundary that the agent cannot change. A developer declares which files and network destinations a task needs, IT can add stricter rules through Intune, and Windows enforces the result at runtime. As Microsoft puts it, “an agent cannot be its own security authority.” MXC offers four levels of isolation.
| Container | Runs on | Best suited for |
|---|---|---|
| Process container | Windows 11, macOS, Linux | Fast, light containment for generated code and tools |
| Session container | Windows 11 only | Long-running agents with their own desktop and account |
| WSL container | Windows 11 only | Linux-based agent toolchains |
| MicroVM | Windows 11 and Linux, experimental | Higher-risk work needing hardware isolation |
Learning mode before enforcement
Writing a tight policy is hard when you do not know what an agent needs. MXC therefore has three modes: Enforcement blocks anything not granted; Learning blocks it and records the attempt in a report; Permissive allows it but records what would have been blocked. Microsoft says only Windows produces these activity reports for process containers, which is a sensible way to build a policy from evidence rather than guesswork.
Who already supports it
Microsoft lists GitHub Copilot, OpenAI’s Codex, OpenClaw, Replit, LM Studio, Unsloth AI and Nvidia’s OpenShell as already supporting MXC. Anthropic’s Claude Code, Box, Egnyte, Manus, Perplexity, Raycast and others will follow. Our article on Nvidia’s OpenShell security system explains how that piece fits in.
Agent identity is still to come
Containment is only one part of securing agents on AI PCs. Microsoft says Windows will “soon” let Microsoft Entra tell an agent’s actions apart from the employee’s, so a misbehaving agent can be cut off without locking the person out, and will extend Agent 365 controls to agents on the device. Those parts are not yet available, so businesses should plan around what has shipped.
Apple’s answer is narrower
Apple is tackling the same risk differently. On 2 October it told developers it will add controls to Full Disk Access, the macOS setting that lets an app read everything on a Mac, because “as AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially”. It has not said which macOS release will carry the change.
Huang’s verdict
Huang said MXC “is going to revolutionize how agents are built and deployed”, and that without containment of this kind, agents are “a complete nonstarter”. Coming from the company that sells the chips, that is partly a sales line. But it matches the view of most security teams we work with: no sandbox, no agents.
Copilot+ Lives On as AI PCs Split Into Tiers
Our preview predicted that the Copilot+ badge was fading. Instead, Microsoft kept it and added tiers above and below it. The new range runs from small always-on desktops to deskside supercomputers, which makes it easier to see which AI PCs are meant for which job.
| Tier | Who it is for | What Microsoft announced |
|---|---|---|
| Mini desktop PCs | Always-on agents at home or in small offices | Click-through setup for popular agents, a native OpenClaw gateway with MXC |
| Copilot+ PCs | Everyday thin-and-light laptops | Copilot local context, actions and models in the coming months |
| Builder PCs (RTX Spark) | Developers, creators and AI teams | Six laptops on pre-order, shipping 16 October; dev boxes later this year |
| AI supercomputers (DGX Station) | Research teams and shared “token factories” | Nvidia GB300, up to 748GB memory, Dell and HP models later this year |
Copilot+ in numbers
Microsoft gave two rare figures for the Copilot+ range: more than 2 trillion AI inferences run locally on those machines each month, and more than 40% of laptops being built for business are now Copilot+ PCs. It did not say how many have been sold, so the second number describes what manufacturers are making rather than what companies are buying.
The Builder PC line-up
The RTX Spark laptops on pre-order are the Asus ProArt P16 and P14, Dell XPS 16 Creator Edition, HP OmniBook Ultra 16, Lenovo Yoga 9n 2-in-1, MSI Prestige N16 Flip AI+ and Surface Laptop Ultra. All begin shipping on 16 October. These are the AI PCs aimed at the Mac buyer, and most are priced like it: HP’s OmniBook Ultra 16 starts at $3,199, Thurrott reported.
The supercomputer at the top
DGX Station for Windows uses Nvidia’s GB300 chip with up to 748GB of memory and 20 petaflops of AI compute, enough to run models of more than a trillion parameters. Microsoft pitches it as a shared “token factory” serving 32 or more agents at once, so a team can cut cloud spend. Dell and HP models are due later this year.
Nvidia's Stake: AI PCs as a Way Into Intel and AMD's Market
For Nvidia, Reuters noted, cracking Windows could help it “go after one of the last major markets dominated by Intel and Advanced Micro Devices”. The RTX Spark chip uses the Arm design rather than the x86 design Intel created, and Nvidia developed it with Taiwan’s MediaTek. Every Arm-based Windows laptop that sells is one fewer x86 machine.
How big the threat is
The near-term risk from Nvidia’s AI PCs is small. Benzinga, citing CNBC, reported analyst Patrick Moorhead’s estimate that Nvidia could sell 10 million PC chips over two years, against the 296 million PC chips IDC counted in 2025 alone. Intel’s PC chip business still brought in $32.2 billion in 2025. And the two firms are partners too: Nvidia agreed in September 2025 to invest $5 billion in Intel.
Games are part of the pitch
Microsoft spent time on gaming to show these AI PCs are not only for AI work. Gears of War: E-Day will showcase the chip, and Microsoft said Call of Duty is coming to RTX Spark in 2027. Anti-cheat support and Xbox PC app compatibility have been the usual weak spots for Arm-based Windows PCs, so this matters for consumers.
The wider partnership
The event was the result of years of joint work, Reuters said. Nvidia’s interest goes beyond selling laptop chips: Windows running local models on Nvidia hardware creates demand for its software tools, from TensorRT to OpenShell. Our earlier piece on how Nvidia would contain runaway AI covers that side of the alliance.
How Our Surface Event Preview Scored
Last week we published a list of likely announcements about Microsoft’s AI PCs. Here is how it compared with what Microsoft actually announced.
| Prediction | Our call | What happened |
|---|---|---|
| Surface Laptop Ultra price and date | Very likely | Yes: from $2,599, ships 16 October |
| Agent security built into Windows | Very likely | Yes: MXC generally available |
| Dev Box price and date | Likely | Yes: $5,999, US only, November |
| Windows 11 features for local AI | Likely | Yes: local models, llama.cpp in Windows ML, Copilot local features |
| A replacement for the Copilot+ badge | Possible | Partly: Copilot+ kept, RTX Spark machines now “Builder PCs” |
| More Surface hardware | Unlikely | Correct: only the two machines already shown |
| Windows 12 | Very unlikely | Correct: no Windows 12 |
Where we were wrong
On price, we expected a starting figure near $2,800 to $3,000 and said a base model under $2,500 would be a surprise. At $2,599 Microsoft came in below our central guess, though only by keeping the entry model to 24GB of memory. Our wildcard, DGX Station for Windows, did get an update, with machines promised for later this year.
What AI PCs Mean for UK Businesses
For most UK organisations, the event changes plans rather than purchase orders. The new AI PCs are priced for developers, creators and AI teams, and Microsoft’s launch material gives US prices only; the trade-in offer is limited to the US and Canada. But there are decisions worth making now.
Who should consider buying
Teams that already pay significant cloud bills for coding assistants or model testing, or that handle data they would rather not send to an outside model, have a real case for AI PCs. Keeping inference on the device helps with confidentiality, although UK GDPR still applies to anything the model processes. Price the 64GB and 128GB models against a year of your current token spend before deciding; our AI strategy team can help with that comparison.
Who can wait
Office fleets running Microsoft 365, a browser and line-of-business apps gain little from these AI PCs today. The Copilot local features will arrive on ordinary Copilot+ PCs over the coming months, and the Snapdragon-based Surface Laptop and Surface Pro refreshed in September remain sensible buys. Wait for independent reviews and sterling prices.
Write an agent policy before staff ask
The most useful step costs nothing. Staff will want to try local agents as soon as they hear about them, so decide now what an agent may access, who approves it and how its activity is logged. MXC’s learning mode is a good way to build that policy from evidence. Intune management for MXC is due “soon”, and our managed IT services team can fold it into existing device policies when it lands.
Do not miss next week’s deadline
Separately from the event, Windows 11 version 24H2 Home and Pro editions get their last security update on 13 October 2026, as we noted in our preview. Any Home or Pro PC still on 24H2 should move to a newer version now, whether or not you plan to buy new hardware.
Treat the Apple comparison carefully
If your developers already use Macs, the trade-in offer and the benchmark claims are not reasons to switch on their own. The fairer test is which platform runs the models and agent tools your team needs, at the memory size you can afford, under security controls your IT team can manage. Both companies now have a credible answer to the first two questions; Microsoft’s is stronger on the third.
AI PCs FAQ
What did Microsoft announce on 7 October 2026?
Microsoft opened pre-orders for the Surface Laptop Ultra and Surface RTX Spark Dev Box, released Microsoft Execution Containers for AI agents, brought several AI models to run locally on Windows AI PCs, and said Copilot will use local files, actions and models on Copilot+ PCs.
How much does the Surface Laptop Ultra cost?
It starts at $2,599 in the US with 24GB of memory and 512GB of storage, and rises to $5,899 with 128GB and 1TB. It ships on 16 October. Microsoft’s launch material did not give UK pricing.
Are these AI PCs faster than a MacBook Pro?
Microsoft and Nvidia claim RTX Spark machines are up to 2.1x faster to the first token, 4.3x faster at AI image generation and 6.2x faster at AI video than a 64GB M5 Pro MacBook Pro. Those are vendor tests on pre-production hardware, so wait for independent reviews.
What are Microsoft Execution Containers?
MXC is a sandbox for AI agents. Developers and IT define which files, networks and screens an agent can use, and Windows enforces those limits whatever the agent decides to do. It is generally available on Windows 11 and also works, with fewer options, on macOS and Linux.
Do I need new hardware to use the new Copilot features?
You need a Copilot+ PC, not an RTX Spark machine. Microsoft says the local Copilot features will roll out to Copilot+ PCs over the coming months, with timing varying by device and market.
Why are AI PCs so expensive right now?
Memory prices have risen sharply because chipmakers are prioritising the memory used in AI data centres. Running large models locally needs a lot of memory, so the AI PCs that make the most of local models are the ones hit hardest.
References and Further Reading
Microsoft brings more AI to PCs as it challenges Apple (Reuters via TradingView)
Building Windows for hybrid intelligence (Windows Experience Blog)
Pre-order our most powerful Surface devices ever (Microsoft Devices Blog)
Microsoft Execution Containers: Policy-driven containment for AI agents (Windows Developer Blog)
Surface Laptop Ultra preorders: everything announced (CNET)
The new Mac mini and Mac Studio are available today (Apple Newsroom)
Apple will tighten macOS Full Disk Access, and names AI agents as a reason (Mixed)
Nvidia debuts $4,999 DGX Spark with half the RAM (The Register)
Apple nears 10% of PC market as memory shortage ends two-year growth run (Tech Times)
Surface Laptop Ultra starts at $2,599 as Microsoft puts more AI work on the PC (The Gadgeteer)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
