PyTorch Announces the Formation of the Ray Accelerator Special Interest Group (SIG), bringing together hardware vendors, Ray contributors, and application developers to define a vendor-neutral accelerator support roadmap. The group initially focuses on three areas: development and testing infrastructure, hardware-aware scheduling, and networking and transport. Ray currently supports more than ten types of accelerators, including NVIDIA and AMD GPUs as well as TPU, Trainium, and Gaudi.
Marvell Technology announced it will acquire optoelectronic components manufacturer Lumentum for approximately $115 billion (€103 billion), with the transaction to be conducted entirely through a stock swap, whereby Lumentum shareholders will receive 4.5 Marvell shares per share; the deal is expected to close in the first half of 2028 and still requires shareholder and regulatory approval. The two companies said the merger will enable them to address the increasingly complex connectivity needs of next-generation AI infrastructure, and that Marvell's serviceable market will expand to $550 billion. In March of this year, Nvidia had already invested $2 billion in each of the two companies; according to Bloomberg, Nvidia CEO Jensen Huang has expressed support for the merger; after the news was announced, Marvell and Lumentum shares fell nearly 13% and nearly 19%, respectively, on Monday.
two companies said the merger will enable them to address the increasingly complex connectivity needs of next-generation AI infrastructure, and that Marvell's serviceable market will expand to $550 billion. In March of this year, Nvidia had already invested $2 billion in each of the two companies; according to Bloomberg, Nvidia CEO Jensen Huang has expressed support for the merger; after the news was announced, Marvell and Lumentum shares fell nearly 13% and nearly 19%, respectively, on Monday.
In October, Google launched its frontier reasoning model Gemini 4 Argon globally, covering Google AI Studio, the Gemini API, Google Cloud, Search, and the Gemini app. Two weeks later, it announced Gemini 4 Carbon, focused on long-horizon engineering and cybersecurity defense, which is currently undergoing pre-release testing with the U.S. government and select trusted cybersecurity parties through the Fairwind Program. In the same period, it released Genie 4, a real-time interactive world model that supports native audio and natural-language scene control; SynthID Detector opened globally in English and can identify AI content from partners such as OpenAI, NVIDIA, Kakao, and Apple.
ounced Gemini 4 Carbon, focused on long-horizon engineering and cybersecurity defense, which is currently undergoing pre-release testing with the U.S. government and select trusted cybersecurity parties through the Fairwind Program. In the same period, it released Genie 4, a real-time interactive world model that supports native audio and natural-language scene control; SynthID Detector opened globally in English and can identify AI content from partners such as OpenAI, NVIDIA, Kakao, and Apple.
X.PIN relays benchmark data for Tianshu Zhixin's TianGai 300, claiming that its throughput in standard inference tests is 1.05 times that of Nvidia's H200. Under MoE workloads, performance improves by 15% and first-token latency drops by 30%, and when fine-tuning Qwen3-32B, its speed is 1.33 times that of the H100.
Rohan Paul, citing Epoch AI data, points out that the total compute of all AI chips in use grows 3.4× per year; but if the compute of newly purchased chips each year grows only 2.5×, this total will eventually also slow to about 2.5×. The original Epoch AI post states that the compute of AI chips sold each quarter has grown 2.5× per year since the start of 2024, down from roughly 6× in 2023 driven by Nvidia H100.
Bloomberg disclosed that Anthropic accounts for 34.5% of SpaceX's Q3 revenue of $14.575 billion, surpassing the U.S. government to become its largest customer. SpaceX's filing labels this customer only as Customer B, and the related payments are for renting Nvidia GPU compute capacity at its Memphis data center; this compute leasing business also serves Google. Anthropic's S-1 filed on October 26 shows that its arrangement with SpaceX reaches up to $120.5 billion by 2029, while its commitments to Google, Amazon, and Microsoft are minimum values of at least $114.5 billion, $110 billion, and $44.9 billion, respectively.
f $14.575 billion, surpassing the U.S. government to become its largest customer. SpaceX's filing labels this customer only as Customer B, and the related payments are for renting Nvidia GPU compute capacity at its Memphis data center; this compute leasing business also serves Google. Anthropic's S-1 filed on October 26 shows that its arrangement with SpaceX reaches up to $120.5 billion by 2029, while its commitments to Google, Amazon, and Microsoft are minimum values of at least $114.5 billion,
OpenBMB announced the open-sourcing of ForgeKernel; its agent completed 42 iterations in about 22 hours and produced 32 kernel versions, ultimately achieving 1.098x performance on the DeepSeek V4.1 FlashMLA Prefill on an NVIDIA B200, at 472 µs versus the 518 µs baseline. The official FlashMLA implementation remained closed throughout; the agent could only run and time its black-box CLI and could not read source code, documentation, or profiling data. The open-sourced content includes the optimized DeepSeek V4.1 fused prefill kernel, the kernels from the paper covering attention, Gated DeltaNet, Kimi Delta Attention, GEMM, and convolution, as well as the ForgeKernel harness that can be used to optimize one's own kernels; across 27 operators, ForgeKernel matched or outperformed 18 of 22 expert-library kernels and outperformed all 7 kernels published by three kernel-generation systems.
rgeKernel; its agent completed 42 iterations in about 22 hours and produced 32 kernel versions, ultimately achieving 1.098x performance on the DeepSeek V4.1 FlashMLA Prefill on an NVIDIA B200, at 472 µs versus the 518 µs baseline. The official FlashMLA implementation remained closed throughout; the agent could only run and time its black-box CLI and could not read source code, documentation, or profiling data. The open-sourced content includes the optimized DeepSeek V4.1 fused prefill kernel, th
Cohere has completed its Series-E funding, targeting a size of up to CAD 7 billion (about EUR 4.7 billion), at a valuation of about CAD 35.7 billion (about EUR 23.9 billion), led by Schwarz Group, the parent company of Lidl and Kaufland, with participation from existing investors including the Canadian government and Nvidia. Cohere had previously merged with Heidelberg-based Aleph Alpha—in effect, Cohere acquired the latter—and Heidelberg continues to serve as a research and development base. Cohere focuses on Sovereign AI for enterprises and governments, and its technology can run in private clouds, on proprietary infrastructure, or in fully isolated environments. Schwarz Group plans to integrate Cohere models into its own STACKIT cloud.
7 billion (about EUR 23.9 billion), led by Schwarz Group, the parent company of Lidl and Kaufland, with participation from existing investors including the Canadian government and Nvidia. Cohere had previously merged with Heidelberg-based Aleph Alpha—in effect, Cohere acquired the latter—and Heidelberg continues to serve as a research and development base. Cohere focuses on Sovereign AI for enterprises and governments, and its technology can run in private clouds, on proprietary infrastructure,
Eclypsium demonstrated, with NVIDIA's support, that an attacker using a compromised DPU and lax memory isolation can modify runtime model weights during inference, causing the model's answers to change while the model files on disk and hash verification still appear normal. The experiment was conducted in an NVIDIA Grace Hopper environment and depended on conditions such as IOMMU passthrough and lax PCIe isolation, which the researchers explicitly noted are not Linux default settings. The report brings DPUs, GPUs, management controllers and their firmware into the AI security boundary, and notes that model signing, secure boot, runtime isolation and continuous monitoring each answer different questions.
Eclypsium demonstrated, with NVIDIA's support, that an attacker using a compromised DPU and lax memory isolation can modify runtime model weights during inference, causing the model's answers to change while the model files on disk and hash verification still appear normal. The experiment was conducted in an NVIDIA Grace Hopper environment and depended on conditions such as IOMMU passthrough and lax PCIe isolation, which the researchers explicitly noted are not Linux default settings. The report
On October 29, Akamai released beta versions of two products, Akamai Inference and Akamai Agent Sandbox, allowing users to run AI models on Akamai Cloud and securely execute code generated by AI agents. Akamai Inference supports 12 open-weight models, including OpenAI's gpt-oss-120b, Google's Gemma 4, and Nvidia's Nemotron 3 Ultra, and provides an OpenAI-compatible API. Agent Sandbox is based on WebAssembly isolation technology and restricts code access to computing resources, network connections, and system functions.
leased beta versions of two products, Akamai Inference and Akamai Agent Sandbox, allowing users to run AI models on Akamai Cloud and securely execute code generated by AI agents. Akamai Inference supports 12 open-weight models, including OpenAI's gpt-oss-120b, Google's Gemma 4, and Nvidia's Nemotron 3 Ultra, and provides an OpenAI-compatible API. Agent Sandbox is based on WebAssembly isolation technology and restricts code access to computing resources, network connections, and system functions.
LG CNS reported Q3 revenue of 1.6643 trillion won and operating profit of 138.2 billion won, up 9.3% and 15.0% year-on-year respectively, with operating margin rising to 8.3%. Financial IT business revenue grew 21.3% to 408.2 billion won, while AI and cloud business revenue grew 9.3% to 961.5 billion won, accounting for 57.8% of total revenue. In September, the company decided to purchase equipment worth 381.4 billion won, including Nvidia's next-generation Vera Rubin GPUs, to build an AI factory.
CNS reported Q3 revenue of 1.6643 trillion won and operating profit of 138.2 billion won, up 9.3% and 15.0% year-on-year respectively, with operating margin rising to 8.3%. Financial IT business revenue grew 21.3% to 408.2 billion won, while AI and cloud business revenue grew 9.3% to 961.5 billion won, accounting for 57.8% of total revenue. In September, the company decided to purchase equipment worth 381.4 billion won, including Nvidia's next-generation Vera Rubin GPUs, to build an AI factory.
Anthropic's IPO filing shows it holds $131.7 billion in cash and short-term investments, a scale exceeding that of Amazon and Microsoft; among U.S. non-bank companies, only Berkshire Hathaway and Alphabet hold more, and it exceeds the combined total of Nvidia and Apple. Its cash and equivalents rose from $17.4 billion at the end of 2025 to $112.4 billion as of June 30, with interest income of $11 billion in the first half of the year, higher than its operating loss of $693 million. Over the same period, undiscounted obligations for leases not yet commenced rose from $50.2 billion to $321.4 billion, with an additional $265.6 billion in fixed or minimum guaranteed commitments; after June 30, it signed new agreements with approximately $245 billion in minimum payments, and Anthropic said it may settle some leases by issuing convertible bonds rather than cash.
rm investments, a scale exceeding that of Amazon and Microsoft; among U.S. non-bank companies, only Berkshire Hathaway and Alphabet hold more, and it exceeds the combined total of Nvidia and Apple. Its cash and equivalents rose from $17.4 billion at the end of 2025 to $112.4 billion as of June 30, with interest income of $11 billion in the first half of the year, higher than its operating loss of $693 million. Over the same period, undiscounted obligations for leases not yet commenced rose from
The Ant Ling team, in collaboration with the NVIDIA AI team, optimized the inference of the open-weight model Ling-3.0-flash on a single NVIDIA DGX Spark using three techniques. On 1,319 GSM8K problems, decoding throughput increased from 37.81 tok/s to 53.89 tok/s (+42.5%), average latency dropped from 15.21s to 11.49s (-24.5%), and accuracy showed no statistically detectable change.
South Korean inference chip startup FuriosaAI has completed a $620 million funding round at a valuation of $2.9 billion, led by the Korea National Growth Fund with a $280 million investment. The company says its TCP chip, based on the RNGD architecture, consumes less than half the power of Nvidia GPUs and is compatible with PyTorch and vLLM; this round of funding will be used to expand production capacity for its next-generation TCP chips, with plans to produce 20,000 units this year and double that to 40,000 units in 2027, while also jointly developing a third-generation processor with Broadcom.
of $2.9 billion, led by the Korea National Growth Fund with a $280 million investment. The company says its TCP chip, based on the RNGD architecture, consumes less than half the power of Nvidia GPUs and is compatible with PyTorch and vLLM; this round of funding will be used to expand production capacity for its next-generation TCP chips, with plans to produce 20,000 units this year and double that to 40,000 units in 2027, while also jointly developing a third-generation processor with Broadcom.
Mistral AI announced that it is opening the weights of its frontier model Mistral Large 4 (codename Le Chonk), which has 1 trillion parameters with 52 billion active parameters and native multimodal support. ML4 was trained in a European data center on 3,800 Nvidia Grace Blackwell GPUs, improving from 28% to 41% on Terminal-Bench 4.0, tying GLM-5.3 for the best coding-agent score while using an average of 47% fewer tokens per task. Open weights mean enterprises can deploy it on their own hardware and behind their firewalls, and the model offers three reasoning modes: low, medium, and high.
al Large 4 (codename Le Chonk), which has 1 trillion parameters with 52 billion active parameters and native multimodal support. ML4 was trained in a European data center on 3,800 Nvidia Grace Blackwell GPUs, improving from 28% to 41% on Terminal-Bench 4.0, tying GLM-5.3 for the best coding-agent score while using an average of 47% fewer tokens per task. Open weights mean enterprises can deploy it on their own hardware and behind their firewalls, and the model offers three reasoning modes: low,
Mistral AI released the open weights of Mistral Large 4 (codename Le Chonk) on Hugging Face on October 29. The model has 100 billion parameters out of a total parameter scale of the 100-billion class, with about 4.9 billion parameters actually invoked per token, using a MoE architecture with 128 experts, of which 4 are activated each time plus 1 shared expert. Compared with the October 6 preview version, Mistral says it continued reinforcement learning over three weeks, raising the Terminal-Bench 4.0 score from 28% to 41%; the Artificial Analysis coding index is 48%, on par with GLM-5.3 while using an average of 47% fewer tokens per task. API pricing drops to $0.70 per million input tokens and $2.10 per million output tokens; the new Mistral Ecosystem Open-Weight License permits commercial use, but reselling model services requires a separate agreement, and the full version requires 8 Nvidia B200 or B300 GPUs.
continued reinforcement learning over three weeks, raising the Terminal-Bench 4.0 score from 28% to 41%; the Artificial Analysis coding index is 48%, on par with GLM-5.3 while using an average of 47% fewer tokens per task. API pricing drops to $0.70 per million input tokens and $2.10 per million output tokens; the new Mistral Ecosystem Open-Weight License permits commercial use, but reselling model services requires a separate agreement, and the full version requires 8 Nvidia B200 or B300 GPUs.
The PyTorch Foundation has released PyTorch 2.15. All process groups using nccl now default to the rewritten NCCL backend, accompanied by a flight recorder that dumps on-site information when collective communication hangs, and can be reverted via TORCH_DIST_USE_NCCL2=0 or nccl-legacy. This version also adds float8_e4m3fn and _scaled_mm support for Apple Silicon, and the new grouped_mm improves MoE decoding throughput by 1.32–1.34x on M3 Ultra. The default wheel is upgraded to CUDA 13.2, while a CUDA 13.4 wheel is also provided with experimental support for NVIDIA Rubin GPUs; a ROCm 10.0 wheel is newly added, and Python 3.10 and macOS 14 are no longer supported.
-site information when collective communication hangs, and can be reverted via TORCH DIST USE NCCL2=0 or nccl-legacy. This version also adds float8 e4m3fn and scaled mm support for Apple Silicon, and the new grouped mm improves MoE decoding throughput by 1.32–1.34x on M3 Ultra. The default wheel is upgraded to CUDA 13.2, while a CUDA 13.4 wheel is also provided with experimental support for NVIDIA Rubin GPUs; a ROCm 10.0 wheel is newly added, and Python 3.10 and macOS 14 are no longer supported.
Teradyne Robotics and Sereact announced a formal partnership at Nvidia GTC Berlin 2026, running Sereact's physical AI platform Cortex on Universal Robots' collaborative robots for unstructured tasks such as returns processing and picking. The two parties have already delivered dual-arm systems composed of hundreds of UR robots to European customers including Zalando, and are autonomously processing returned goods at Zalando fulfillment centers in Greven, Germany, and Świebodzin, Poland. Cortex is a VLA model trained on Nvidia Isaac Sim and Cosmos, and runs on Nvidia Jetson Thor at deployment.
Teradyne Robotics and Sereact announced a formal partnership at Nvidia GTC Berlin 2026, running Sereact's physical AI platform Cortex on Universal Robots' collaborative robots for unstructured tasks such as returns processing and picking. The two parties have already delivered dual-arm systems composed of hundreds of UR robots to European customers including Zalando, and are autonomously processing returned goods at Zalando fulfillment centers in Greven, Germany, and Świebodzin, Poland. Cortex i
Akamai has released two services, Akamai Inference and Akamai Agent Sandbox: the former provides 12 open-weight models via an OpenAI-compatible API, with parameter counts ranging from 8B to 2.8T, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b, and Google Gemma 4; the latter runs code generated by AI agents in isolation based on WebAssembly, and is released as an open-source component of the Spin framework. Akamai has also expanded its NVIDIA Blackwell Ultra GPU clusters, with each eight-GPU node equipped with 2.3 TB of HBM3e memory.
rvices, Akamai Inference and Akamai Agent Sandbox: the former provides 12 open-weight models via an OpenAI-compatible API, with parameter counts ranging from 8B to 2.8T, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b, and Google Gemma 4; the latter runs code generated by AI agents in isolation based on WebAssembly, and is released as an open-source component of the Spin framework. Akamai has also expanded its NVIDIA Blackwell Ultra GPU clusters, with each eight-GPU node equipped with 2.3
Oracle announced that it will deploy Google TPU on OCI, along with inference chips from OpenAI, Cerebras, SambaNova, and Positron AI, giving customers more computing options beyond Nvidia GPUs. Oracle says it will become the first hyperscale cloud provider other than Google Cloud to offer TPU; TPU and the Cerebras wafer-scale engine are planned to launch in 2027, and Positron Atlas is already available. At the same time, Oracle is expanding its partnership with Nvidia, with the Vera Rubin NVL72 system and Vera CPU now generally available on OCI.
racle announced that it will deploy Google TPU on OCI, along with inference chips from OpenAI, Cerebras, SambaNova, and Positron AI, giving customers more computing options beyond Nvidia GPUs. Oracle says it will become the first hyperscale cloud provider other than Google Cloud to offer TPU; TPU and the Cerebras wafer-scale engine are planned to launch in 2027, and Positron Atlas is already available. At the same time, Oracle is expanding its partnership with Nvidia, with the Vera Rubin NVL72 s
