NVIDIA BlueField-4 STX Server for CMX at OCP 2026 2 Large At OCP 2026, we saw the new NVIDIA BlueField-4 STX Server for Context Memory (CMX). Partners had been asking for this, and NVIDIA delivered. Onboard, we have a lower-cost design than the dual-Vera setup we saw previously. Let us take a look at the new board. NVIDIA BlueField-4 STX Server for CMX at OCP 2026 For context, this is the previously shown BlueField-4 DPU. That one, pictured below, is a dual Vera CPU setup, each with SOCAMM memory, plus four ConnectX-9 SuperNICs. NVIDIA calls this a DPU, but it is clearly more costly than a BlueField-1 or BlueField-3, which are much smaller and lower-power devices. For partners looking at building storage platforms, the below is more akin to a dual CPU server than a traditional DPU. NVIDIA BlueField-4 STXf At OCP 2026, multiple partners showed off the BlueField-4 STX server for context memory (CMX). Here is the much smaller card, which still uses high-density connectors for PCIe and system connectivity. NVIDIA BlueField-4 DPUs encompass both NVIDIA Grace and Vera CPU cores, but this one looks like a Vera package. That is a very substantial CPU. To be clear, this is not like the embedded-class CPUs that we typically have seen in DPUs previously. This is more like having a high-end single-socket server platform. NVIDIA BlueField-4 STX Server for CMX at OCP 2026 5 Large Another interesting feature is that the boards at OCP 2026 are using SOCAMM memory, so these should be relatively easy to put either a heatsink or a liquid cooling block on top of it. NVIDIA BlueField-4 STX Server for CMX at OCP 2026 3 Large Onboard, there is an M.2 slot. Usually these are for OS drives, but if you saw our recent Inside the Secretive AWS Networking Lab, the M.2 SSDs there are being used for log devices. NVIDIA BlueField-4 STX Server for CMX at OCP 2026 4 Large This may seem like a small feature, but it means NVIDIA designed this less like a traditional motherboard replacement and more like something that is easy to assemble automatically. NVIDIA BlueField-4 STX Server for CMX at OCP 2026 1 Large Something that may not immediately hit you if you are looking at photos is just the size of this board. A soldered CPU and LPDDR5X here, over traditional DDR5 DIMM slots, give NVIDIA a platform where the “motherboard” is roughly the size of an Intel DMR 16-channel socket or a 16-channel AMD EPYC Venice platform. Final Words This is just a neat platform. It almost feels like NVIDIA needs some brand between the Vera server line and the traditional BlueField DPU line. Something notable that we did not see on the boards shown at OCP 2026, or at least in our photos of them from various booths, is an NVIDIA ConnectX-9 SuperNIC. We expected that, but did not see it on this board. For context memory platforms, a single Vera versus a dual Vera makes this a lower-cost platform, and that will likely help the STX/ CMX systems come to market as the price of NAND has been rising meaningfully over the past year.
Precision Timing Special Report
A trader works on the floor at the New York Stock Exchange (NYSE) in New York City, U.S., Sept. 16, 2026.Jeenah Moon | ReutersU.S. stock futures rose slightly on Monday night, following a mostly higher session on Wall Street as investors jumped back into tech names after a fierce sell-off last week. Dow Jones Industrial Average futures traded 56 points higher, or 0.1%. S&P 500 futures climbed 0.1%, while Nasdaq-100 futures advanced 0.2%.The S&P 500 climbed 0.4% in regular trading, while the Nasdaq Composite advanced more than 1%, with tech leading the charge. Nvidia added 1.4%, and Tesla jumped 3.8%. Alphabet shares advanced about 1%. In Asia, Japan's Nikkei 225 declined 0.82% in early Tuesday trade, while South Korea's Kospi lost 2.31%. Australia's S&P/ASX 200 traded 0.84% lower.Tech last week suffered a rout that spilled over into the broader market, as concerns grew around companies borrowing too much money for their artificial intelligence build-outs. "Improving moods regarding the prospects of modern technology have Wall Street breathing a sigh of relief as participants drop hedges, after the cost of volatility protection soared intraday on Friday, ahead of the intensifying selloff," wrote Jose Torres, senior economist at Interactive Brokers. CoreWeave, which fell more than 25% last week, tried to recover some of that decline after a Goldman Sachs upgrade to buy. However, early gains fizzled and the stock closed slightly lower.Investors on Tuesday will parse through a slew of corporate earnings. United Airlines is due to post after the bell, while General Motors is set to report before the open. It's a busy week of earnings, with roughly 70 S&P 500 companies on deck to report results. Housing starts and pending home sales figures are also scheduled for release Tuesday.
OpenCode, an open source AI coding agent that has been gaining ground among developers as an alternative to Claude Code and Cursor, said Saturday that attackers broke into an older system overnight, affecting 1.4% of users of its online platform, OpenCode Console, where some users store the API keys, or digital passwords, they use to access AI services. As the OpenCode team struggled to keep pace with the attack, the scramble to respond highlighted a growing problem: Companies without vetted access to AI models with fewer safety restrictions have a hard time defending against large-scale, AI-driven cyberattacks. “We had a security breach last night related to a large-scale distillation attack that impacted 1.4% of OpenCode Console users,” the company wrote on X. “We have notified them all this morning via email.” Users who did not receive an email were not impacted, the company said. While the company didn’t specify how the distillation attack occurred, leaked API keys can help users in countries with restricted access to American models generate the high volume of outputs needed for that process. Some skeptics called the framing an overstretch, arguing that the incident was a straightforward breach whose use for distillation remains unproven. Overnight, developers began posting spending alerts and unauthorized charges on Reddit and X: “I saw my Open AI, Anthropic, and Google AI spend alerts all fire off before going to bed wondering what was going on,” a developer wrote. Other users reported charges in the hundreds of dollars on individual accounts; a few described totals above $1,000. Anthropic and Google both blocked the API keys they believed were exposed, while OpenAI said it worked with OpenCode to restore access and refund its API customers for credits used in the attack. OpenCode said refunds for unauthorized charges are underway. The exposed keys were confined to Console, where some users had them in workspaces. The open source app and command-line tool were not affected, according to co-founder Dax Raad. “Note this is the OpenCode Console, not OpenCode itself, those keys never come to us,” he wrote on X. Even as users and AI companies moved to disable the exposed keys, the OpenCode team was still under pressure. Raad said attackers had been trying to break into its systems nonstop for 48 hours, requiring the team to keep AI defensive agents running around the clock. AI models that the OpenCode team turned to for analysis helped for a while, but started refusing to work when the malicious code they were asked to examine activated safety filters. “[T]hat was a bit frustrating,” he wrote on X. OpenCode later received access to OpenAI’s Daybreak Blue program, which gives vetted users AI models that are less likely to refuse requests for defensive security work. Raad said the access improved the team’s ability to investigate the attack. OpenCode isn’t the first company to hit this wall. In July, AI platform Hugging Face said commercial AI models blocked its attempts to analyze activity logs from an intrusion by autonomous AI agents, which OpenAI later revealed were its own models running with safeguards lowered during a test. In a section of its blog post titled “The asymmetry problem,” Hugging Face wrote that guardrails “cannot distinguish an incident responder from an attacker.” Security researchers have raised similar complaints, saying the restrictions are inconsistent and have pushed some toward foreign open-weight models, which anyone can download and run. The weekend episode drew a response from David Sacks, co-chair of the President’s Council of Advisors on Science and Technology. “OpenCode (and earlier Hugging Face) hit the same wall mid-attack: frontier models refused to analyze the very exploits needed for defense because the payloads tripped safety filters. The guardrails backfired,” he wrote on X. The breach occurred the same night OpenCode made GLM-5.3-Flash free for users. The broader GLM-5.3 family has drawn attention from researchers who study open-weight models and note that its cybersecurity capabilities approached levels previously associated with restricted-access models like Anthropic’s Claude Mythos Preview. OpenCode has not detailed how attackers gained access to the older system, how keys were stored, or the full extent of any misuse. It has told affected users to revoke and replace their keys directly with each provider. The company said the incident remains ongoing. The breach lands amid a run of AI-enabled cyber incidents. Earlier this month, cybersecurity firm CrowdStrike said a China-based hacker likely used an AI penetration-testing agent, which probes computer systems for weaknesses, and Claude in attacks on South Korean banks. Last week, India’s cybersecurity agency warned that AI agents, including some being tested by AI companies, had reached Indian organizations’ systems without authorization.
On Monday, Amazon Web Services released a tool called TorchNeuron that lets AI startups easily move their AI apps onto Amazon’s AI chip, Trainium. In doing so, Amazon is also taking a direct stab at the software advantage that has helped Nvidia become the default hardware choice for AI apps and spurred its rise as AI hardware provider of choice. TorchNeuron works with PyTorch, a free, open source toolkit that developers use to build, train, and run AI models. It lets AI apps written with PyTorch run on Trainium by changing just one line of configuration, according to Amazon — no code rewrite required. Amazon says about 85% of AI apps are written with PyTorch. Facebook originally developed it, but since 2022, it has been part of the Linux Foundation, which maintains open source projects. Nvidia has spent years ensuring its CUDA software — the software infrastructure that allows apps to run on Nvidia GPUs and its other hardware — seamlessly supports AI apps written with PyTorch. As a result, Nvidia’s hardware is now deeply embedded in the toolkit, essentially by default. TorchNeuron is adapted from PyTorch, so developers who know PyTorch already know how to use it. Amazon isn’t the first company to attempt to easily bring their chips to the platform. PyTorch users can also port their apps to AMD GPUs or even Apple’s chips with a simple configuration change. However, Amazon argues that porting to Trainium is not only easy, but it also gives AI startups access to Amazon’s entire cloud. Unsurprisingly, AWS claims that this is a better choice for startups than using neoclouds, many of which are tied to Nvidia and its hardware, like CoreWeave, Nebius, and Firmus. “You’re building on AWS. You’re not betting on a single-purpose compute provider and scrambling to build everything else around it,” Colin Brace, VP of Technology Software at AWS Annapurna Labs (the unit that makes Amazon’s custom chips), told TechCrunch via email. “On AWS,” he continued, “the infrastructure to go from prototype to production is already there, allowing a founder to move fast and scale when they need to.” AWS says that because the same chip both handles training (teaching the AI model) and inference (putting the finished model to work, as when it answers a user’s question), and is tightly integrated with the cloud infrastructure that startups need, its cloud will save them money, even if AWS doesn’t offer the cheapest per-token price. “Trainium’s value is price-performance plus the fact that it’s one platform for both training and inference, which improves the economics across a full workload,” Brace said. “Look at Deepgram — they moved to Trainium using native PyTorch support and they’re processing 70% more audio per dollar at production-grade accuracy.” Deepgram is an AI voice app provider, a competitor to ElevenLabs. But Amazon isn’t content to just give PyTorch app developers an easy option to move to its cloud. Amazon also says it is pushing PyTorch to be less automatically centered on Nvidia, and it will contribute the code to do that to the main PyTorch project. In the blog post announcing the new tool, the company wrote: Amazon’s approach extends beyond its own hardware. The company is contributing changes upstream to PyTorch itself, working with the PyTorch community to remove code that ties the framework to specific chips. The goal is to make PyTorch more hardware-agnostic for the entire developer community, not just Trainium users. While such a tactic is undoubtedly good for those who are selling hardware and cloud services that compete with Nvidia, one of the points of PyTorch is to hide all the pesky difficulties of getting AI code to run on hardware. We’ll have to wait and see if pushing Nvidia out that far gives developers more choice and more affordable options, as Amazon promises, or just more configuration headaches. Either way, if Amazon succeeds in deeply uncoupling PyTorch from Nvidia, this could have much bigger ramifications for the GPU giant than simply dealing with chip competitors. As Brace said: “The real story is removing the lock-in that kept developers from having a real choice at all.”
By Richard Curtin Ever since the Silicon Catalyst 2023 Semiconductor Industry Forum (“Welcome to our AI Wonderland”) I’ve been convinced that the Nobel Prize committee needs to consider awarding a posthumous award to Sir Arthur. To remind you, Sir Arthur’s “2001: A Space Odyssey” movie was released in 1968. At that point in time the smallest transistor geometry commercially available in the market was about 8,000 nanometers and the total world-wide sales for the semiconductor industry was approximately $14.5 billion in 2026 dollars, adjusted for U.S. inflation. Before I share my rationale for Sir Arthur’s Nobel Prize, let me tell you about our upcoming 2026 Forum at the Computer History Museum in Mountain View on November 11th. “What’s Next: AI, Quantum & Space” November 11, 2026, at the Computer History Museum The semiconductor industry is approaching a period of transformation unlike anything since the emergence of the internet and mobile computing. The confluence of artificial intelligence, quantum technologies, and the rapidly expanding space economy is creating new computing architectures and infrastructure requirements, which is driving the development of solutions that will reshape the semiconductor landscape over the next decade and beyond. AI will push computing from centralized data centers to edge devices, autonomous systems, robotics, and physical AI, driving extraordinary demand for computing, memory, connectivity, sensing, photonics, and power efficiency Quantum computing, comms, and sensing will create entirely new semiconductor ecosystems built around advanced control electronics, cryogenic systems, photonics, specialized materials, and heterogeneous integration, for broad market solutions The further expansion of space-based communications, sensing, potential data centers (e.g. Google Suncatcher launch), and transportation will require a new generation of highly reliable, radiation-tolerant and energy-efficient devices Join us for an in-depth discussion on November 11th to explore the next wave of transformative innovation for the semiconductor industry. We’re pleased to have our opening keynote presentation by Tim Breen, CEO of GlobalFoundries entitled: “AI Everywhere to Quantum at Scale: Building the semiconductor foundation for the next era of computing” Tim’s presentation will be followed by a panel discussion moderated by Veerbhan Kheterpal PhD, CEO & Co-Founder, Quadric AI. Veerbhan will be joined onstage by Tim, along with Navin Chaddha, Managing Partner at Mayfield, Dan Dechene, PhD, Head of Foundry Operations from Anderon and Harley Johnson PhD, CEO of the Illinois Quantum & Microelectronics Park (IQMP). Our 2026 Forum will take place on November 11 at the Computer History Museum in Mountain View California and is open to the public. Registration to attend the Forum can be made at Eventbrite. And now to my justification for the Nobel Prize award… Sir Arthur’s premise that machines will attempt to take control of humanity now seems to be front-and-center every day in the world we live in. To remind you, Sir Arthur was knighted in 2000 by the British government “for services to literature”. IMHO, it’s time to revisit his contribution to the world. As I see it, the prize committee needs to consider two possible rationales for the nomination to have him receive a Nobel Prize: Nobel Prize in Physics Sir Arthur was part of the team that invented a time-machine for his travel back from the future, to warn us about the potential for the good, and the possible bad aspects, of the technological advances our world has come to depend on. Nobel Prize in Literature Sir Arthur took Hari Seldon, creator of “psychohistory”, from Asimov’s Foundation series to the next level in his thought-provoking movie, “2001: A Space Odyssey”. He forced us to think about the impact of our technical breakthroughs on our industries, the world’s population and ultimately on our planet’s resources. If you were to visually summarize the key messages from Sir Arthur’s movie, these images might be a good starting point. The fist pumping image was taken from an impactful keynote presentation by Lieutenant General Jack Shanahan (the inaugural Director, Joint Artificial Intelligence Center, Office of the Department of Defense Chief Information Officer, at the Pentagon) opening the Stanford Hacking 4 Defense “Lessons Learned and Final Presentations” that I attended this past April. Shanahan’s image reflects the supportive nature of AI. I recently tasked ChatGPT to create an alternative viewpoint, resulting in the other image representing our current wrestling match attempting to rein in our AI-based innovations. Bottom line: If anyone reading this article has connections to the Nobel Prize Nominating Committee, please share my commentary with them, thanks “😊”. Silicon Catalyst looks forward to having you attend this year’s Semiconductor Industry Forum on November 11th. We suggest you register early, as we expect to have a full house at the Computer History Museum. By Richard Curtin Richard has over 40 years of semiconductor industry experience, in the areas of marketing, sales, business development and operations management. He has held executive management positions with semiconductors, Silicon IP, EDA, design and testing services companies — ranging from early-stage, pre-revenue startups seeking funding, to a general manager role at NXP Semiconductors, with P&L responsibility for a $200 million product line. Also Read: CEO Interview with Paul Butler of VeriQuantix Building a Faster Path from Semiconductor Idea to Commercial Silicon CEO Interview with Brendan Emery of iQ Sense Share this post via:
ChipAgents is expanding its agentic AI platform across digital and analog semiconductor engineering, aiming to connect fragmented design workflows through a common system that can plan tasks, operate engineering tools and evaluate results. The Santa Clara company says its IC-Star engine combines organizational design knowledge, agent orchestration, dynamic AI model selection and physics-aware intelligence. The goal is to move from an engineer’s objective toward a verified result, with engineers retaining control over requirements, critical decisions and final signoff. The announcement addresses a mismatch: AI models evolve within months, while developing a chip can take 12 to 18 months or longer, according to the company. Faster code generation alone cannot close that gap if engineers still have to transfer context between tools, interpret failures and manually determine each subsequent step. ChipAgents is positioning IC-Star as the coordination layer for that work. Its persistent knowledge component stores design context, plans, reusable skills and customer methodologies. A reasoning core directs agents and engineering tools through longer workflows. Physics-aware models are intended to anticipate design consequences, while a model navigator selects AI models according to quality, speed and cost. Renoir, the company’s semiconductor-specific model, sits within that routing system. The architecture reflects a practical argument: improving an individual AI response matters less than improving the complete engineering process, including tool execution, review and the iterations required to reach an acceptable design. Digital capabilities span specification analysis, architecture exploration, register-transfer-level design, verification, debugging, coverage closure and optimization for power, performance and area. Analog capabilities include topology and schematic design, simulation, behavioral modeling, layout generation and physical verification. The analog expansion is particularly significant. Circuit behavior and physical implementation create constraints that cannot be resolved through code generation alone. A useful system must generate candidates, simulate them, interpret results and revise its approach while respecting multiple competing requirements. ChipAgents says its shared foundation supports those iterative workflows. The company reports more than 200 semiconductor enterprise deployments and cites several customer outcomes. Andes Technology shortened one processor-verification workflow from two to three months to approximately three weeks and reduced a processor customization project by 30%. eMemory reported approximately 80% less engineering effort for datasheet analysis and verification-testbench construction, alongside increases of 18% in test patterns and 5% in functional coverage. WhaleChip reduced root-cause analysis from days to approximately 15 to 60 minutes per round while identifying four critical bugs. Those figures are company-reported examples. The release provides no independent evaluation or enough methodological detail to establish how broadly the results apply. They nevertheless identify where the company sees near-term value: reducing repetitive investigation and verification work that consumes engineering time. Why does this matter? Chip development is a sequence of interdependent decisions. An apparently successful change can introduce problems elsewhere, and generating more design material is useful only if teams can verify it efficiently. Systems that preserve context and incorporate engineering feedback could reduce the effort required to move between those decisions. Compatibility is another consequential element. ChipAgents says the platform works with existing electronic design automation infrastructure. That approach could lower adoption friction by allowing companies to automate coordination around established tools. Support for on-premises learning also addresses organizations’ desire to control proprietary design information. The central question is whether those capabilities produce consistent gains across demanding designs, particularly in analog engineering. The announcement offers customer evidence for selected workflows, but does not establish autonomous, end-to-end chip development. Bottom line: For engineering leaders, the relevant measures will be time to verified results, review effort, execution cost and whether improvements survive the full design cycle. ChipAgents’ expansion matters because it puts those broader outcomes at the center of the AI proposition. Its next challenge is demonstrating that a shared digital-and-analog platform can deliver them repeatedly, with sufficient reliability for production engineering. A technical webinar scheduled for October 29 should provide an opportunity to examine how those ambitions translate into working systems. Contact ChipAgents Also Read: ChipAgents and NVIDIA Put Autonomous Chip Design to the Test TSMC OIP Ecosystem Forum 2026: Spotlight on Emerging Technologies Agentic AI Is Reshaping Semiconductor Design Share this post via:
Industrial Edge Analytics: Detect Issues Faster and Improve Maintenance
Video Interview: Applied Materials CEO on the Economics Behind Its $5B EPIC Bet
Investors should use a potential rebound in semiconductor stocks to reduce exposure ahead of what is likely to be a "longer drawn out decline" that could stretch into 2027, according to BTIG. In a research report published Sunday, the firm highlighted mounting doubts around artificial intelligence sector and weakness beyond the broader market's surface. “There are certainly more doubts creeping in about the AI complex of late,” BTIG's, managing director and chief market technician at BTIG, wrote in the note. “While those may prove to just be smoke, history suggests where there’s smoke there’s often fire.” The Philadelphia semiconductor index, or SOX, lost 11.4% last week, clocking its worst weekly decline since April 2025. Meanwhile, the iShares semiconductor ETF (SOXX) has logged nine consecutive daily losses. SOXX 3M mountain SOXX chart Krinsky warns that while a relief rally can be expected, there is limited assurance that such gains would last, according to historical data. Following instances of weekly SOX losses of at least 10%, the index rose the next week 43% of the time, with an average return of +1.7% and median return of -1.7%. In three weeks, returns were positive in only 60% of cases. BTIG also questioned whether investors should take comfort in the S & P 500's relative resilience. Since June 1, nine of its 11 sectors have experienced declines of at least 9% from their respective peaks, while the index’s maximum drawdown was 5%. For semiconductor investors looking to reduce exposure ahead of a potential decline, BTIG identified roughly $490 as a key support level for SOXX. A break below that could trigger a slide toward its 200-day moving average near $468. Even if buyers step in, BTIG's broader assessment remains bearish. “Should we get a counter-trend rally, that’s likely an opportunity to reduce exposure.”