AMD's Madhu Rangarajan, corporate VP of Compute and Enterprise AI, says the company never set out to build its Zen 6 EPYC 9006 "Venice" chips around agentic AI, a not-so-subtle dig at NVIDIA and Arm, which both pitch their latest server CPUs as made for that exact job. Rangarajan still called EPYC 9006 the best CPU for agentic workloads during a media session covered by Tom's Hardware. Chief Zen architect Mike Clark said much the same in a separate talk, admitting AMD didn't see agentic computing coming, though he added that the company always expected the CPU to keep a role in AI software. Here's what Rangarajan said:

We think that the EPYC 9006 CPU is the best CPU for the agentic era, and spoiler alert, we didn't magically know four years ago that agentic was going to be a thing and purpose-built something for agentic. The reality is an agentic workflow is a diverse set of tasks… and in order to execute those tasks well, you need compute that is able to handle a diverse set of workloads with a full portfolio.

Anyone wondering what is filling up their GPU's VRAM can now try gputaskman, a free, open-source tool from Microsoft VP Scott Hanselman. It runs in a terminal window with an htop-style layout and lists every process using the GPU with its memory footprint, engine activity, and loaded runtime, such as CUDA or Vulkan. A panel up top tracks utilization, clocks, temperature, and power. Hanselman shared the project on X last week and asked for pull requests.

Task Manager has shown GPU usage since 2017, and it will happily tell you that 15 GB of video memory is in use. It just won't say what is using it. gputaskman goes after that gap with a heavy focus on local AI. Its model inspector can identify the LLM a process has loaded and report its quantization, weights, and estimated KV cache cost, while Ollama and LM Studio users also get context size and proper model names. Hanselman calls it early, experimental software, and the AI breakdowns are estimates for now.

Otto, the company building the operating system for the agentic era, today introduces Otto One, the first purpose-built agent computer: a compact, screenless computer designed for an AI agent to live and work on. Running on Otto OS, its proprietary software layer, Otto One lets people deploy, direct and monitor autonomous agents from a phone or desktop while the device works continuously in the background, even when the user closes their laptop or walks away.

AI agents are finally capable of taking on a persistent role in people's lives. The real opportunity is not a better inbox assistant; it is an agent that can keep a household organized, manage a budget and taxes, coordinate care and appointments, and keep a business moving. To do that well, it needs usable context and the authority to act across the financial, personal and work systems that make up a life.

Intel and Altera today announced a collaboration to develop a flexible, real-time sensor bridge for a range of Edge & Physical AI industries including healthcare, aerospace and defense, and robotics applications. By combining Intel compute, connectivity and software with Altera FPGAs, the companies aim to help original equipment manufacturers (OEMs) address the growing complexity of connecting, processing and acting on real-world data across increasingly sensor-rich edge environments.

Edge and Physical AI systems can require data from different sensors and interfaces to be brought together while meeting requirements for timing, performance, power and cost. Some applications also require very high bandwidth or flexible form factors in which input/output (I/O) is located away from the compute board. Sensor data must then be streamed rapidly into memory for the host Intel processor to use across edge, robotics and other workloads.

Gigabyte today announced the launch of AI TOP AERO, a new AI desktop powered by the NVIDIA RTX Spark platform. Positioned as a Personal AI Studio, AI TOP AERO combines an NVIDIA Blackwell architecture GPU, up to 128 GB of unified system memory, and Gigabyte's cooling and compact chassis design. The new desktop expands Gigabyte's AI PC lineup, bringing local AI development, content creation and gaming to personal workspaces.

NVIDIA RTX Spark unites NVIDIA AI, RTX graphics, and unified memory in one slim, powerful Windows PC. Designed to run larger AI models locally--without cloud dependency--and move between creative work, development, and immersive gaming without missing a beat. Connect apps, tools, and workflows into one seamless experience. Private by default. RTX Spark brings NVIDIA's full-stack AI platform, ready for the AI workloads of today and whatever comes next. Delivered on the familiar, trusted Windows platform.

There is a new player in the operating systems arena from KORA Systems Inc. Called KORA, this operating system is the first bottom-up OS designed specifically for AI PCs, enabling agentic workflows. It is one of the first concepts where a system is so heavily customized that AI workflows are native to it. The company developed this customized operating system on top of NixOS, using it as a foundational layer.

The customization includes the development of what are called neural workspaces and file systems, which replace traditional folders and desktop apps. While these traditional elements are still present, the workflow on KORA OS is organized around intent, context, and memory. Each workspace is divided based on user intent, and the OS manages the files accordingly. Security is robust, as KORA implements each workspace within a secured sandbox for applications and agents, with separate permissions. The memory and context of the workflow remain on the user's desktop, never leaving the machine. Features like semantic search and email are natively integrated, allowing you to search across your entire OS or email to find what you need.

The power of frontier AI models is increasing, and the gaming realm has already experienced what AI can do in terms of mods, changes, bug fixes, and many other aspects when a powerful model is put to work. Late last night, Anthropic released its most powerful frontier model, Fable 5.2, which brings the best intelligence from the company's offerings into an API. As a demonstration of this model's capabilities, we are seeing a relatively complex first-person shooter developed from scratch today. This includes models, sounds, and textures, all created by Fable 5.2. It took about eight hours, 445 million input tokens, and about 2.7 million output tokens, totaling around $550 if done through an API. The result is a game called "Saltline," which can be played entirely in a browser through Claude Artifact, a platform used to showcase creations.

Below you can find the "Saltline" game, where the objective is a simple mission to cut three relays while being constantly attacked by an army of robot dogs and drones. After completing the objective, the main character reaches a boat, and the mission is complete. While this is a very simple task, the fact that Fable 5.2 accomplished all of this from a prompt is an amazing achievement. It can deploy thousands of agents to get one thing done, and with better descriptions, it can perform tasks much closer to what you have imagined.

Gaming PCs sitting idle could one day earn their owners credits, or even cash, by training AI models, at least if a newly published Microsoft patent application ever becomes a product. Titled "Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization," the filing US-20260349820-A1 from Microsoft Technology Licensing was submitted in April 2025 and made public on November 5. It describes chopping large training jobs into small micro-tasks sized to what each device can handle, then sending them to phones, laptops, and gaming PCs whenever they have spare capacity. Owners would have to explicitly opt in, after which the system could track the work done and apply credits to their account, redeemable for free software, access to the trained model, or money.

To stay out of the way, the system watches GPU and CPU load, memory use, temperature and power draw, and checks less often when the GPU is busy or hot. Apparently, tasks can be paused or stopped when limits are crossed, and results get cross-checked across devices to catch faulty or tampered work. The filing doesn't give a payout rate or list supported hardware, and Microsoft hasn't announced a Windows or Xbox service, so it's impossible to say whether the rewards would cover the extra electricity. Of course, this is only an application for now, and plenty of those never turn into products.

ASRock Rack is promising up to 4 TB of DDR5 on a single socket with its DMRD16-2L+, one of the first motherboards built for Intel's upcoming Xeon 7 "Diamond Rapids" processors. The board was shown at the OCP Global Summit 2026 and has since been detailed by ServeTheHome. It has 16 DIMM slots in a one-DIMM-per-channel layout, matching the 16-channel memory setup Intel detailed at HOT CHIPS. Regular RDIMMs top out at 128 GB per module and 8000 MT/s, while MRDIMMs run at up to 12,800 MT/s, with taller 2U MRDIMMs reaching 256 GB each. Populate all 16 slots with those, and you hit 4 TB, along with a theoretical bandwidth of roughly 1.6 TB/s.

Expansion is unusual, with no standard PCIe slots and no OCP NIC 3.0 connector. All I/O goes through up to 16 MCIO x8 connectors, each rated for PCIe 6.0 or CXL 3.0, for 128 lanes in total. The rest is fairly basic: one PCIe 5.0 x4 M.2 slot, two Intel I350 Gigabit Ethernet ports, an ASPEED AST2700 BMC with a dedicated management port, and VGA output. ASRock Rack fits it all on a 12 x 13-inch EEB PCB, about the size of a large E-ATX board, so it could squeeze into some bigger workstation cases. Pricing and availability have not been shared yet, but that is hardly surprising, since Intel has the Xeon 7 family slated for a 2027 debut on the new LGA9324 socket. The board is aimed at servers, though anyone planning a high-memory workstation for local AI will probably keep an eye on it.

Microsoft has filed a patent application for an AI system that can change what someone says in a live video and redraw their lips to match. The filing, US 2026/0341309 A1, was submitted in April 2025, published on October 29, 2026, and first spotted by patent tracker Patentlyze. In short, an AI agent scans the live audio and picks words to swap or remove based on rules and a viewer's profile, such as their age, languages, or location. A generative model then replaces those words in both the audio and video, reworks the frames around each edit and plays the result with a delay of roughly one to ten seconds. It could run locally on AR glasses, phones, tablets or cameras.

A separate claim covers live translation with matching lip movements, which is easily the most practical use. However, the filing also lists filler words, stutters, mispronounced or offensive words, and colloquial terms like "y'all" or "bite the bullet" as things to replace, and the new words can even come from an internet search. It mentions lectures, award shows, game shows, and live sports as examples. The original feed may be stored as "ground truth," but the main claim says nothing about telling viewers what was changed. Of course, Microsoft files plenty of patents that never become products, and the examiner has already rejected the claims as filed, which is routine at this stage. Sony has been busy with AI patents as well, ranging from frame generation to an AI that can play for distracted players.

Google has started rolling out the November update for Googlebook, and the biggest change is how much memory the platform's on-device AI takes up. On the Acer Googlebook 14, ASUS Googlebook 14, and Lenovo Googlebook 15, Google's release notes say it has improved memory usage for on-device AI to make multitasking more stable on 16 GB devices. Apparently, this comes down to updated memory management in the NPU driver, which reportedly frees up between 3 GB and 6 GB of system RAM and lets the OS claw back memory when Chrome needs more of it. The Help Center notes don't put a number on it, though, and the 3 GB to 6 GB figure comes from Google's announcement on Reddit.

The update, build CL3B.260622.286.R2, carries the November 1 security patch and around 30 fixes in total. Among them are keyboard and trackpad input dropping out on convertibles, black screens with DRM-protected video on external monitors, external USB microphones not working, Zoom crashes with background effects, and bootloops when an update gets interrupted. HP and Dell owners also get a fix for Bluetooth audio lip-sync issues, and the Linux Terminal gets several reliability fixes. All three models getting the memory fix ship with 16 GB of RAM. Freeing up to 6 GB on those machines is a big deal, since that's over a third of the total memory. The update is rolling out gradually, and owners can check for it under Settings > System > Software updates > System update.

Police in Oakland, California, are investigating the theft of six pallets of NVIDIA hardware worth $11.4 million that were bound for Fort Worth, Texas, the Bay Area News Group reports. A truck carrying a food transportation company's logo picked up the cargo in Oakland on August 29 as planned, but the truck and its drivers then went off the grid heading south. According to authorities, the truck made several multi-hour stops before police lost track of it in San Bernardino County. Apparently, nobody realized the cargo was gone until early September, and Oakland police say the theft was reported to them on September 13. No arrests have been announced.

What exactly was on those pallets hasn't been disclosed, so we don't know if it was AI accelerators, complete systems, or GeForce graphics cards. Speaking to ABC7, UC Berkeley professor Olaf Groth said NVIDIA chips can sell for roughly double their US price abroad, with a $550,000 compute system going for $1 million or more. He added that the heists are getting bigger and more complex, which points to increasingly professional crews.

A little over a year ago, PIF and Silver Lake were announced as two parts of EA's new ownership trifecta, following a $55 billion buyout agreement, and it has now been revealed in a new Reuters report that PIF and Silver Lake have met with Unity Software to initiate talks to buy the game engine developer. Similarly to the EA buyout, insiders claim that the Unity buyout would also be a leveraged buyout agreement, although there have been no claims about a potential buyout amount.

As was the case leading up to the EA buyout, Silver Lake has already invested in Unity, with an 8% share in the company. Following the recent EA buyout, the gaming giant's new owners have made several changes at the company, including a restructuring that led to significant layoffs and the closure of BioWare, one of the most recognizable game studios in gaming history. Earlier in the year, EA also reportedly accelerated its investment in and use of AI in order to cut costs. Unity itself has announced a similar increased adoption of AI with its new Unity Spark AI systems.

AMD CEO Lisa Su now calls AMD a "data center first" company, two days after AMD reported a 40% drop in gaming revenue. Speaking to Yahoo Finance's Brian Sozzi on November 5, Su said AMD's data center business should grow 50% sequentially from Q3 to Q4, and suggested it could more than double next year. Here's how she put it:

Dr. Lisa Su, AMD CEOWe are a data center first company, that is our profile, and you know, people need more computing to power all these applications.

Data Center brought in a record $8.5 billion in Q3, up 97%, and now makes up 64% of AMD's $13.4 billion total, up from 47% a year ago. Gaming fell to $773 million, or under 6% of revenue. AMD blames it mostly on lower semi-custom sales, meaning the console chips it makes for Sony and Microsoft, along with weaker graphics demand. Gaming was flat against Q2, though, and last year's Q3 was boosted by a 181% console-driven jump to $1.3 billion. Su used the same "data center first" line back in 2024, when data center was just over half of revenue.

Beelink today announced the Kickstarter launch of the ME Pro, a modular NAS available in two-bay and four-bay configurations. Its swappable motherboard design lets users upgrade performance or add drive bays as their needs change. The ME Pro uses cooling and vibration damping to help reduce operating noise and comes with Fygo OS preinstalled for file sync, backup, and remote access.

Modular Hardware for Flexible Upgrades and Quiet Operation

• Swappable motherboard modules. A slide-out design supports compatible Intel and AMD motherboard modules. The same module can move between compatible two-bay and four-bay enclosures in either direction, so users can keep their existing motherboard when expanding storage.
• Compact, one-piece metal chassis. The enclosure takes up about 50% less space than NAS systems with the same number of drive bays, making it easier to fit on a desk or in a living room or studio.
• Cooling and vibration control. Thermally conductive silicone and aluminium brackets draw heat away from the drives. A vapor chamber and internal blower distribute and dissipate heat, while a tightly secured assembly helps reduce vibration noise. The ME Pro is designed for round-the-clock operation, with noise levels as low as 34 dB.

Valve now requires developers to launch their game through the Steam client and run it for at least five minutes before submitting it for review, as part of several new additions to Steam's pre-release checklist. It doesn't have to be the developer playing, since anyone's playtime counts, as long as it was logged within the past seven days, or since the last build upload if that's older. Apparently, Valve keeps getting builds that were never run through Steam at all, which is how games end up with broken launch options or missing depots.

The changes are meant to cut down Steam's review backlog. Valve says store page and build submissions have jumped over the last few months, and it's quite obvious why that would be. For Valve, build reviews that usually take three to five days are now taking up to 12 business days, largely because games that fail basic checks have to be reviewed again. Other new checks include at least five screenshots tagged as gameplay, a test save written to Steam Cloud for games that support it, a list of required redistributables like DirectX or Visual C++, and complete achievement setup.

Leaker Haze2K1 says on X that Intel might put the Arc B580 on the end-of-life (EOL) list. Apparently, Intel wants to prioritize workstation and AI inference cards like the Arc Pro B50, which are reportedly selling really well and carry higher margins. Intel hasn't confirmed anything, and Haze2K1 only says the card "might" go EOL, so this is only a rumor for now. However, the logic checks out. The Arc B580 launched in December 2024 at $249 and beat the RTX 4060 in TechPowerUp's review. The Arc Pro B50 and B60 use the same BMG-G21 silicon, and the B60 pairs the same 20 Xe cores with 24 GB of memory instead of 12 GB. Intel estimated the B60 at around $500 at launch, roughly double the B580's price. Clearly, Intel stands to gain more monetary benefit from the workstation cards.

As for the consumer GPUs, their supply already looks quite thin. The B580 is reportedly out of stock at most retailers, and prices have also climbed well above MSRP. The leaker doesn't mention the Arc B570, so we don't know if Intel has similar plans for that card as well. Intel's larger BMG-G31 chip also showed up in March in the workstation-only Arc Pro B70 and B65, so the shift toward pro cards seems to be underway already. If the B580 does go away, gamers will lose one of the more affordable 12 GB cards, even though the Arc B580 was selling above its MSRP. That said, I wouldn't expect cheaper Arc cards anytime soon

Marvell Technology, Inc. (NASDAQ: MRVL) and Lumentum (NASDAQ: LITE) today announced that they have entered into a definitive agreement for Marvell to acquire Lumentum in an all-stock transaction. Under the terms of the agreement, Lumentum stockholders will receive a fixed exchange ratio of 4.500 shares of Marvell common stock for each share of Lumentum common stock they own. The transaction has an implied enterprise value of $115 billion for Lumentum, representing approximately $1,132 per share of Lumentum common stock and a premium of approximately 23.9% based on each company's unaffected closing share price as of October 28, 2026, the last trading day before market speculation regarding a potential transaction began. Post-closing, current Marvell stockholders will own approximately 67% of the combined company on a fully diluted basis, while Lumentum stockholders will own approximately 33%.

The AI infrastructure industry is at an inflection point. Rapidly growing bandwidth requirements and the physical limitations of copper-based connectivity are accelerating the transition toward optical technologies. As AI infrastructure scales to increasingly complex architectures, particularly scale-up and scale-in, the industry is shifting from discrete, modular components toward highly integrated optical solutions that require much closer co-design of electrical and optical technologies to deliver greater bandwidth, performance and energy efficiency.

Qualcomm has officially validated its High Bandwidth Compute (HBC) concept, which the company showcased this summer. Instead of using traditional HBM and compute dies, Qualcomm has developed a technology that stacks multiple LPDDR layers and embeds a base die that serves as a compute engine for AI inference. In Qualcomm's San Diego lab, the first-generation HBC has achieved silicon validation with full read-write and near-memory computation capabilities. HBC utilizes LPDDR memory, stacking it across multiple layers in a 3D vertical space and interconnecting them with through-silicon vias (TSVs). This approach provides better energy efficiency than traditional HBM, which stacks regular DDR in vertical layers, as the low-power DDR (LPDDR) chips consume less power while delivering similar bandwidth and capacity.

Qualcomm has successfully stacked multiple LPDDR layers on top of each other, with the base of the HBC being a compute die that performs near-memory computation and offloads processing to near-memory dies. This technique is similar to what HBM4 memory currently does, with the base die being a logic die to better integrate compute solutions for packet tracing and data preparation for entering and exiting HBM. However, Qualcomm aims to achieve higher efficiency and more optimized data movement, with a 6x increase in bandwidth per watt compared to the current HBM specification. The goal is to have compute directly under these memory stacks so that data travels only a few microns instead of a few centimeters.

Two days after arguing that the DRAM in home PCs would be put to better use in data centers, OpenAI CEO Sam Altman sat down with around 50 Samsung presidents on Saturday, October 31. The group went well past the chip business, with leaders from Samsung's finance and biotech arms joining the electronics side at the company's training center in Yongin. Samsung presented it as part of a leadership seminar series for its company-wide AI push, and the talks apparently covered where AI is headed, the markets it could open up, and how it changes the way large companies make decisions. According to Samsung, Altman told the room that adopting AI alone doesn't give anyone an edge, and that the payoff comes from using it to move faster and build better products.

Nothing came out of it in terms of a new supply contract, at least publicly, and Samsung's own write-up stuck to AI strategy. Still, it's hard to look at the visit separately from the memory shortage. Samsung signed a letter of intent with OpenAI back in October 2025 to act as a memory partner for Stargate, with OpenAI's demand projected at up to 900,000 DRAM wafers per month. Separately, on October 31, SK Group chairman Chey Tae-won also reportedly met Altman in Seoul, with SK hynix and SK Telecom executives joining, according to industry sources. Last month, AMD's Lisa Su spent hours with Samsung's chip chief chasing HBM supply, and amid all these meetings, Micron's CEO expects the memory crunch to get worse in 2027 and 2028. It seems like every major buyer is knocking on the same Korean doors.

As usual, generative AI appears to be splitting the gaming industry in half. While Valve recently acknowledged the issue of AI-generated clones on Steam—although declined to actually do anything about it—Roblox just published a shareholder letter in which it boasts that "AI-enabled game creation, automation in code generation and AI supported asset creation are leading to an explosion of content" since it launched its AI assisted game builder tools.

The letter adds that over 80% of its top 100 games by hours played in the month of September were AI assisted, while the raw number of games created with its Build generative AI platform topped 70,000 since Q2. Usage of Roblox's AI tools for game development more than tripled among top creators. It also notes that in Q3, two games "created almost entirely with AI" managed to break into the top 20 Roblox game ranking by concurrent player count.

Since Valve launched the Steam Deck, there has been no question that its user-friendly UI has been one of its biggest selling points, with its openness and publisher-agnostic approach coming in a close second. A new Decky Loader plugin makes use of this openness to implement both the PlayStation 5 and Xbox Series X|S home screen user interface layouts on the Steam Deck, effectively replacing the stock Steam UI. The projects are available for free on GitHub (as decky-ps5-dashboard and decky-xbox-dashboard, both from the same user. The project lead, kromossomom-debug on GitHub, is a self-described AI user, notably not self-identifying as a developer, and Claude is openly listed as a contributor in the projects, so it's safe to assume they're mostly vibe-coded.

Decky PS5 Dashboard features the rounded UI elements from the PlayStation UI, with square game tiles and a screenshot from the highlighted game as the dashboard background, and it includes info cards under the highlighted game, with detailed HowLongToBeat data shown in the card alongside user play-time, review scores, achievements, and game install size. It also implements the PS5's Media tab, accessible with L1/R1, and it implements the PS5 color scheme across the Steam UI menus.

AI game cloning, reverse-engineering, and decompiling have all become increasingly popular of late, with a cottage industry of gaming projects popping up almost overnight, including everything from controversial Fortnite and Fallout clones to fan-made console ports. Now, it seems as though Anthropic may have had enough of the bad press and potential pressure from gaming industry giants, with some users reporting (via Tom's Hardware) that Claude will no longer recreate game mechanics or existing games.

One X user flagged a new rule in Claude that could potentially be the source of the restriction, which states "Do build original games, games in a genre, and games from the user's own designs and assets. Don't build a recreation of a specific existing game, even if it is renamed or called original." While the intention with the new rule may have been to prevent copyright infringement and potential legal implications for Anthropic, it seems as though it has resulted in a flurry of new issues on the Claude GitHub, with reports including claims that Claude is refusing to clone a developer's own work from earlier in a project for a similar mechanic in a latter part of the game.

Apple is once again the world's most valuable company, taking the top spot back from NVIDIA after Friday's trading session. Apple shares slipped 1.2% on October 30, leaving it with a market cap of $5.03 trillion, while NVIDIA dropped 4.2% to close at $4.93 trillion. This is the first time Apple has been ahead since it briefly overtook NVIDIA in late July. For context, NVIDIA became the first company to cross $5 trillion almost exactly a year ago, in October 2025. Friday's sell-off came after a run of tech earnings that disappointed investors, with Intel and Amazon following earlier reports from Alphabet and Meta. Intel's stock fell roughly 29% on Friday after a weak Q4 outlook, which CNBC called its worst day since the 1970s.

It looks like the bigger concern is whether the huge spending on AI data centers will pay off, and NVIDIA, as the main supplier of AI GPUs, is directly tied to those budgets. The outlays have grown large enough that Alphabet and Amazon have seen their free cash flow turn negative, pushing them toward debt markets for funding. Apple, meanwhile, has largely stayed out of the AI infrastructure race and kept its capital spending in check. Its shares are up 27% this year, more than twice NVIDIA's gain, and its free cash flow for fiscal 2026 (ended September 30) is projected to climb 38% to $142 billion. The gap between the two is only about $100 billion, though, so the ranking could easily flip back with one good day for NVIDIA.

OpenAI CEO Sam Altman thinks the DRAM sitting inside personal computers would be better used in data centers. He made the comments on the Cheeky Pint podcast, hosted by Stripe co-founder John Collison, in an episode released on October 29. The two were discussing the memory supply chain when the topic turned to local inference, meaning running AI models on your own hardware instead of the cloud. Here's what Altman said, at roughly the 29-minute mark:

Sam Altman, OpenAI CEOA lot of people want local inference. And somehow, they assume that the problem is solved now. Because, we are going to have open-source models. We're going to be able to run them on our computers. And so we're not going to need all the stuff in the data center. But from a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer. And, being used 5% of the time. You want maximum utilization in the cloud if you're trying to deal with this very supply constrained environment. And somehow everyone's... many people's intuition is the opposite of that.