As the US and China both push quantum computing, sensing, and networking as strategic priorities, the real question isn't who's ahead but ahead of what, since a quantum advantage is only meaningful relative to the classical system it's compared against. Quantum radar is the clearest case: the underlying physics of quantum illumination is real and has shown genuine detection gains in controlled lab conditions, but a laboratory advantage doesn't automatically translate into a long-range operational radar advantage once range, clutter, atmospheric loss, and receiver complexity enter the picture. The same trap applies to quantum machine learning and quantum computing broadly, where a system can look superior on paper simply because it was benchmarked against a weak or poorly optimized classical competitor rather than the best available alternative.
Apple is laying off over 200 employees across its Vision Pro, Siri, and software engineering divisions as it refocuses engineering priorities on artificial intelligence and upcoming hardware form factors.
Zipline has been rapidly scaling its autonomous drone delivery infrastructure, and this Uber partnership signifies a massive leap in commercial drone logistics for everyday consumers.
Anthropic has hired former Google custom silicon executive Amir Salek to join its compute team as the frontier artificial intelligence lab begins laying the groundwork for an in-house semiconductor initiative.
This post introduces low-level optimization techniques to compute the day of the week from an integer day count faster than standard compiler-generated modulus operations.
Etched raised $700 million at a $21 billion valuation, led by Jane Street, just a month after closing a $10.3 billion round and eight months removed from a $5 billion valuation in December. The jump is backed by real deployment: Jane Street tested and bought Etched's hardware before investing, and has now put its first “frontier inference cluster” rack into production in its own data center. Etched's pitch centers on two custom-built components for AI inference, a low-voltage prefill chip and a new memory/interconnect design for the decode stage, and the company has walked back its original plan to etch specific models into silicon in favor of systems that run any frontier model.
Google is reportedly collaborating with AMD to develop a 10th-generation TPU variant featuring on-package CPU cores and advanced 3D packaging. While Google maintains its long-standing partnership with Broadcom for conventional AI accelerators, this hybrid design is expected to leverage AMD's x86 CPU IP and SoIC packaging expertise to integrate general-purpose compute directly alongside TPU matrix engines. The architectural shift targets complex reinforcement learning, reasoning, and agentic workloads that demand low-latency execution and higher CPU-to-accelerator compute ratios than standard LLM pipelines.
Cerebras Systems has announced its next-generation wafer-scale AI supercomputing system, claiming substantial speed and throughput advantages over Nvidia's flagship accelerators. Built around its massive single-wafer processor architecture, the new machine eliminates inter-chip interconnect bottlenecks to deliver ultra-low latency and higher tokens-per-second performance for large language models. The rollout sharpens competition in datacenter AI infrastructure as Cerebras pitches its specialized wafer-scale silicon as a higher-performance, power-efficient alternative to traditional multi-GPU clusters for frontier model training and real-time inference.
Nvidia has reserved advanced production capacity on TSMC's 1.6nm (A16) manufacturing node for its next-generation "Feynman" architecture, slated for production in the second half of 2028. Incorporating nanosheet transistors alongside backside power delivery networks, TSMC's A16 process offers substantial improvements in logic density, power efficiency, and signal routing. By securing early foundry capacity, Nvidia locked in its long-term hardware roadmap.
Sponsored
Runware is building its own infrastructure to make AI compute cheaper and more abundant. Sonic Inference Pods pack up to 1200 GPUs into modular 1MW data centers, deployed in weeks rather than years and purpose-built to squeeze more performance from every GPU. Run any AI workload on them through Runware Serverless [8].
As liquid cooling becomes standard for high-power AI accelerators, heat now travels through a whole chain - cold plate, manifold, hoses, distribution unit, and facility heat rejection - rather than staying inside the server, meaning problems anywhere along that path can quietly eat into a chip's thermal margin before temperature alarms ever trigger. Component temperature alone only confirms a chip is currently within limits - it doesn't reveal how hard the control system is working to hold that temperature or how much margin is left, since pumps and valves are designed to compensate for developing issues invisibly.
This article details how AI datacenters are evaluating the transition to 800VDC architectures of power distribution. Moving away from current 48V distribution streamlines the conversion path from medium-voltage grid AC to sub-1V silicon, cutting conversion heat losses and reducing heavy copper cabling requirements. Realizing this grid-to-gate efficiency relies on silicon carbide (SiC) and gallium nitride (GaN) power semiconductors, solid-state transformers, and upgraded DC-rated switchgear to manage high-voltage safety and rapid load switching across AI clusters. While these technologies are rapidly maturing, it still may take years to change datacenter designs that form the bedrock of all computing stacks.
US AI investment has become dangerously circular. Suppliers are financing customers who spend that capital back on the suppliers' own hardware, making it hard to separate real end-user demand from financial engineering. China's push into overseas industrial AI hardware, robotics, data-center equipment, and automated production systems has a cleaner signal, since a robot shipped abroad and installed on a factory floor can be checked against customs data and production volumes rather than internal financing loops.
The reported FCC ban on Chinese optical transceivers shows that the US-China tech war is expanding past chips and into the smallest links of AI's physical plumbing. Washington's concern centers on data theft, service disruption, and long-term dependence, since these modules sit physically embedded throughout the network and could theoretically carry malware. A sudden ban risks backfiring, as US suppliers can't match Chinese scale, leading hyperscalers like Amazon and Microsoft to face real supply disruption if forced to substitute quickly.
AI chip startup Groq has reached a $3.5 billion valuation in a new funding round following its IP acquisition by Nvidia.
Autonomous trucking company Kodiak Robotics has officially commenced road testing on public California highways following state regulatory clearance.
Nokia is shutting down its research and development operations in China and cutting approximately 1,600 jobs as part of a broader corporate restructuring and cost-reduction program.
Etched, a high-flying AI chip startup, is aggressively poaching senior hardware engineers and system architects directly from Nvidia.
Nvidia's SEC filing revealed it holds nearly 123 million SpaceX shares, worth roughly $21 billion at the end of June (now closer to $17 billion after SpaceX's post-IPO share slide), a stake that originated through Nvidia's investment in xAI before Musk merged it into SpaceX. It comes right after Musk confirmed SpaceX will build its data centers exclusively on Nvidia hardware, citing the Vera Rubin architecture as "the best AI computer," locking in another major customer to Nvidia's silicon while Nvidia simultaneously holds equity in that same customer.
The bear case on AI infrastructure spending has always been simple: with Nvidia shipping a new architecture roughly every year, GPUs should lose economic value within two or three years, undermining the long-dated debt financing the whole buildout leans on. CoreWeave's Q2 earnings call complicated that story, disclosing a renewed contract for Nvidia's 2020-era A100 GPUs running all the way through 2029, at pricing CEO Mike Intrator said matches “full freight” from years ago.
Ukrainian military intelligence says Russia's S-71 “Monochrome” cruise missile, which offers reduced observability and autonomous targeting, uses an Nvidia Jetson Orin NX module, a consumer-grade system-on-module normally sold for robotics, drones, and edge AI research rather than military hardware. Nvidia confirmed the Orin NX isn't officially sold in Russia or export-controlled, and said it can't track products after resale but will act on confirmed export-control violations. In the missile, the chip likely powers a local electro-optical perception system, recognizing images in real time to assist terminal guidance, echoing an earlier Ukrainian claim that Russia used the same modules in Shahed drones.
Semiconductor equipment OEMs are increasingly adopting build-to-print contract manufacturing to scale production and meet soaring global chip demand without heavy capital expenditures. As expanding AI fabs drive severe equipment backlogs, toolmakers partner with electronics contract manufacturers to produce sub-assemblies from existing engineering drawings. This distributed model provides immediate access to cleanrooms and precision machining, cutting equipment lead times, preserving OEM capital, and improving supply chain resilience across regional fab expansions.
A 19-company industry coalition led by Lightmatter has formally launched the Open Silicon Photonics for AI Systems initiative as an official Open Compute Project workstream. Alongside a 300-page foundational architecture white paper, the initiative aims to establish an open, vendor-neutral blueprint for integrating co-packaged optics into datacenter infrastructure. Targeting OCP Modular Hardware System and Open Rack v3 standards, the framework scales clusters from 72 to over 1,024 nodes while overcoming copper reach and bandwidth bottlenecks.
The semiconductor and datacenter industries are actively debating the necessity of 1-megawatt server racks driven by continuous, power-intensive AI workloads. While hyperscalers explore ±400V DC power distribution, advanced liquid cooling, and 3D-IC packaging to support ultra-dense compute, critics argue this scaling vector is unsustainable. Instead of cramming accelerators into single dense enclosures to preserve short copper reach, alternative approaches advocate scaling horizontally across racks using optical interconnects and specialized silicon.
Intel VP Robert Hallock confirmed the company plans to keep Raptor Lake CPUs in production and stabilize supply of its LGA 1700-compatible chips “for years to come,” following a spike in demand for the four-year-old Alder Lake and Raptor Lake lineups. It's a direct response to the DDR5 price crunch: Raptor Lake supports both DDR4 and DDR5, letting buyers upgrade without paying today's inflated DDR5 prices, and rumors point to a possible third refresh, tentatively “Raptor Lake Next,” arriving alongside the DDR5-exclusive Nova Lake in early 2027.
Semiconductor Engineering reports that Co-Packaged Optics cannot scale to profitable high-volume manufacturing without industry-wide testing standardization. As datacenters adopt optical interconnects for AI workloads, test cells face severe bottlenecks from proprietary fiber connectors, non-standardized laser alignment schemes, and fragmented data formats.
Nvidia has scaled back its planned financial guarantee for OpenAI's 10-gigawatt Ohio data center from $250 billion to under $120 billion, covering only the project's first phase, after investors raised concerns about the chipmaker's risk exposure.
United States utility-scale battery storage projects are stalling as aging electrical grids struggle to accommodate rapid interconnection demands.
The US Commerce Department has warned Apple against sourcing memory chips from Chinese suppliers CXMT and Yangtze Memory Technologies to alleviate ongoing supply shortages.
The Missile Defense Agency is transferring mature, in-sustainment programs like THAAD and Aegis/SM-3 to the Army and Navy to free up bandwidth for newer priorities feeding into Golden Dome.
AMD has launched its Ryzen AI X100 processor series, directly challenging Nvidia's dominant GPU-centric embedded robotics platforms. Engineered for real-time edge processing and autonomous robotics, by integrating heterogeneous compute units directly with low-latency NPU execution engines, AMD aims to reduce system latency, lower thermal output, and displace Nvidia Jetson hardware across industrial automation, drone, and humanoid robotics deployments.
Google introduced its Pixel 11 smartphone lineup alongside the $1,900 Pixel 11 Pro Fold, emphasizing deep software intelligence. Powered by the new Tensor G6 processor, the devices feature incremental hardware updates, improved camera sensors, and enhanced battery efficiency. Rather than focusing only on physical changes, Google is prioritizing its mobile strategy around deeply integrated Gemini AI features, bringing real-time multimodal processing, proactive context awareness, and system-level agentic tools directly to flagship and foldable form factors.
Samsung Foundry has updated its advanced semiconductor roadmap, pushing its 1.4nm process node rollout to 2029. Samsung plans to introduce High-NA EUV scanner technology to enable 1nm-class and sub-1nm nodes starting in 2030 and beyond. This timeline shift allows the foundry to focus on optimizing current 2nm Gate-All-Around nodes while refining ultra-advanced packaging and power delivery networks for future hyperscale clients.
High Bandwidth Memory is rapidly becoming the ultimate proving ground for 3D semiconductor assembly and advanced packaging yields. As memory stacks scale vertically with more DRAM dies, manufacturers face challenges regarding thermal dissipation, substrate warping, and micro-bump defects. Mastering novel testing methodologies and direct Cu-Cu hybrid bonding on HBM production lines is crucial for semiconductor foundries. These assembly innovations establish foundational yield management standards for next-generation 3D IC architectures and complex heterogeneous AI chiplets by learning from HBM experiences.
Lockheed Martin Skunk Works and the US Air Force Test Pilot School flew the X-62 VISTA, a variable-stability test aircraft, through eight flights in which an onboard AI agent completed 27 autonomous air intercepts against a live T-38 target. The aircraft carried Lockheed's Legion Pod, an infrared search and track sensor that fed live targeting data directly to the AI, which then independently worked out the target's position and flight path and maneuvered the aircraft into intercept position, closing the full sensor-to-action loop in real time rather than in simulation. It builds on Lockheed's Supermassive AI framework, developed and flight-validated in roughly three months, with the next phase integrating RTX's PhantomStrike radar and expanded sensor and compute capacity onto the X-62 for further mission-systems testing.
Attackers commonly hijack sensitive operations like control-flow transfers by corrupting the data pointers that feed into them, and existing defenses either only cover part of the attack surface or carry overheads too heavy for constrained systems like IoT devices, kernels, and bare-metal firmware. ROLoad-PMP tackles this with new RISC-V instructions that can only load data from read-only memory regions tagged with specific keys, guaranteeing the integrity of whatever a pointer points to even if the pointer itself gets corrupted.
Chipstrat discusses with AMD CVP Madhu Rangarajan why agentic AI deployment models are driving renewed demand for server CPUs alongside dedicated GPU accelerators. As multi-step agentic workflows increasingly handle tool calls, code sandboxing, API orchestration, and context retrieval, host processors bear significant computational burdens. AMD is strategically positioning its high-core EPYC CPUs to capitalize on these x86 execution requirements.
Over two days in late July, NHTSA and the DOT rolled out the biggest recalibration of federal autonomous vehicle oversight in years, headlined by the first-ever commercial exemption letting Zoox manufacture and deploy robotaxis that don't comply with eight specific Federal Motor Vehicle Safety Standards, covering things like windshield wipers, mirrors, and crash protection built around the assumption of a human driver. Rather than a blanket pass, the exemption comes with an oversight condition that lets NHTSA issue evolving "Operational Authorizations" as Zoox's self-driving system matures, a notable shift from rigid one-size-fits-all compliance toward more adaptive regulation for fast-moving hardware.
AI chipmaker Cerebras Systems reported lumpiness across its core hardware business despite raising its full-year financial targets.
Cisco reported fourth-quarter results and a full-year sales forecast that topped Wall Street expectations, driven by record enterprise demand for artificial intelligence infrastructure.
Nvidia is reportedly testing lower memory configurations for its upcoming Rubin Ultra AI accelerators in response to persistent global High Bandwidth Memory deficits.
This post analyzes Beijing's campaign aimed at creating an autonomous, sanctions-resistant quantum technology ecosystem.
Nvidia announced financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, aiming to raise over $500 billion in outside capital for AI "factories," the data centers that train and run AI models, with Jensen Huang pitching this compute as a durable, reusable "investable asset class." The core problem is that these are non-binding MOUs, not signed deals, and the entire pitch hinges on residual value, what the hardware is actually worth if the paying customer walks away, since Huang's own evidence for durability (rising H100 rental rates) proves current earnings, not resale value.
Chinese AI accelerator firm Moore Threads, founded by former NVIDIA executives, filed for a Hong Kong listing to raise capital for global R&D. As Nvidia faces strict US export bans, Moore Threads is positioning itself alongside Huawei and Cambricon to capture China's domestic AI datacenter hardware market.
Agility Robotics says Digit v5 will work alongside humans without traditional safety fencing, using upgraded perception to detect people and cycloidal actuators to react while maintaining balance and payload control. Better battery life and fast charging target 20-hour operating days, while an ISO flange will support standard industrial end-effectors and eventually autonomous tool swapping. Early customer deployments are planned for December.
Sandia National Laboratories and Elemental Nuclear Energy are building supercritical carbon dioxide (sCO2) Brayton Cycle generators, which use CO2 held in a state with properties of both a liquid and a gas as the working fluid instead of steam, letting the closed-loop system convert heat to power more compactly than a conventional steam turbine. The first target is a 1-megawatt system running on natural gas and waste heat aimed at small modular data centers and remote military sites, with commercial deliveries planned for 2028.
Dyna-2 is a video-diffusion world-action model pre-trained on more than 1 million hours of egocentric human-manipulation video, with no robot data included during pre-training. Dyna reports that prediction performance improved monotonically across a 1,000-to-1-million-hour training ladder, including zero-shot evaluation on 39 robot tasks from embodiments excluded from training. After limited robot post-training, normalized physical performance across 14 tasks rose from 20% to 53% as human-video pre-training scaled, with the largest model leading on nine tasks. Roughly 10 minutes of robot demonstrations were sufficient to post-train bottle-cap opening with two five-fingered robot hands.
Researchers propose a hybrid computing framework that moves beyond traditional peak TOPS/W efficiency metrics. The paper demonstrates that combining digital orchestration with analogue, photonic, and neuromorphic accelerators delivers a scalable path toward energy-efficient AI systems. Rather than forcing all workloads onto pure digital logic, the proposed system dynamically routes tasks so physical substrates earn an expanding role only where they offer a measurable, system-level efficiency advantage while digital software manages noise, integration, and fallbacks.
Leading Chinese artificial intelligence firms continue to rely on Nvidia GPUs for training frontier AI models, delaying a full transition to domestic silicon alternatives. Despite Beijing's aggressive drive for semiconductor self-reliance and heavy investments in local accelerator developers like Huawei, technical hurdles persist. Domestic AI chipmakers face ongoing challenges with software ecosystem maturity, inter-chip interconnect bandwidth, and cluster stability during massive-scale training runs.
Sodium-ion batteries made from abundant salt reserves are emerging as a viable commercial alternative to lithium-ion energy storage systems. While sodium-ion cells possess lower energy density than lithium, their low raw material costs and rapid charging capabilities make them ideal for stationary grid storage and short-range urban electric vehicles worldwide.