The Information said Anthropic signed at least 14.8GW of new compute capacity over the past 11 months, with the resources set to come online over the next several years. Including 1GW to 2GW the company had already locked in before October last year, Anthropic’s total contracted compute footprint is estimated at about 16GW. Based on currently disclosed agreements, the potential maximum cost could reach $517 billion, with most of that spending expected to fall over the next decade. The expansion was described as a response to stronger-than-expected demand for Claude, particularly from Claude Code and Cowork. Amazon could supply as much as 5GW under a new agreement, while Google and Broadcom add another 5GW, and Microsoft and NVIDIA account for about 1GW. Anthropic also leased the full capacity of SpaceX Colossus 1, securing more than 220,000 NVIDIA GPUs, including H100, H200, and GB200. Even at that scale, Anthropic’s compute plan may still trail OpenAI’s stated goal of reaching 30GW by 2030, with OpenAI projecting about $750 billion in cumulative compute spending by then.

According to a report from The Information, Anthropic has signed new compute capacity agreements totaling at least 14.8GW over the past 11 months, with potential costs reaching up to $517 billion. Combined with 1-2GW locked in before October last year, the company's total committed compute capacity now stands at approximately 16GW. This expansion is primarily driven by demand for its Claude AI model, with the growth of Claude Code and Cowork products exceeding expectations. Key partners include Amazon (up to 5GW), Google and Broadcom (5GW combined), Microsoft and NVIDIA (about 1GW total). Anthropic also leased the entire compute capacity of SpaceX's Colossus 1 data center, securing over 220,000 NVIDIA GPUs including H100, H200, and GB200 models. Many contracts extend beyond 2030. For comparison, OpenAI aims to reach 30GW by 2030 with cumulative spending of around $750 billion. The $517 billion figure represents the maximum potential cost under current contracts, with some capacity possibly being deferred or unused, and some contracts including early termination clauses.

ChatGPT, Claude, Grok and some Codex users were hit by service issues on the morning of Sept. 3, but there is still no single official explanation linking all three incidents. The clearest cross-company statement came from SpaceX, which said Grok problems followed an outage at its Memphis compute center and added that it also wanted to apologize to impacted compute partners. That matters because Anthropic is paying SpaceX $1.25 billion a month to use Colossus 1 in Memphis, according to the report. Colossus 1 is described as having more than 220,000 NVIDIA GPUs, including H100, H200 and GB200 units. The report says the contract runs through May 2029, implying about $15 billion a year in payments. OpenAI’s timeline was different. It said ChatGPT and Codex were unavailable from 7:43 a.m. because of an internal routing error and recovered at 8:17 a.m., roughly 80 minutes after the earlier issues at Anthropic and xAI began. Status pages from Microsoft Azure, AWS and Google Cloud reportedly showed no matching incidents that day, while Cloudflare said it saw no major service disruption.

NVIDIA announced the RTX Spark device powered by the N1X chip, shipping in October, featuring up to 128GB unified memory and Blackwell GPU delivering petaflop-level performance. The company also released the open-source NVIDIA PAIR tool, which pools idle computing power across home devices to accelerate AI agent tasks, cutting completion time by over half.

AI computing firm Nscale announced a multi-year strategic partnership with Figure, expanding the deal to over $6 billion. The agreement represents the potential deployment of up to 100,000 NVIDIA Vera Rubin platform GPUs for AI training and inference.

Oppenheimer has raised its price target for SpaceX (SPCX) from $250 to $280, maintaining an 'Outperform' rating. Analyst Tim Horan highlighted SpaceX's vertically integrated AI platform, access to NVIDIA GPU resources, and rapid infrastructure expansion. He noted strong recent progress in Grok and Cursor, with AI usage exceeding expectations. The revised long-term revenue outlook is based on higher AI compute pricing and potential returns from the next-generation NVIDIA Rubin chip.

NVIDIA reported FY2027 Q2 results that stood out even by its own recent standards: revenue reached $96.22 billion, up 106% year over year, while data center revenue climbed to $89.02 billion, up 117%. Adjusted EPS came in at $2.22, above the $2.09 consensus, and the company guided Q3 revenue to $108 billion, plus or minus 2%, ahead of the roughly $104 billion analysts had expected. Shares rose about 4% in after-hours trading to around $218. The report showed how concentrated NVIDIA’s business has become around AI infrastructure. Data center now accounts for about 93% of total revenue, with hyperscalers still the largest buyers. At the same time, management highlighted faster growth from AI cloud, industrial and enterprise customers. One of the most closely watched figures was supply commitments, which jumped from $119 billion at the end of Q1 to $279 billion at the end of Q2, largely tied to memory procurement for the Vera Rubin platform. Still, the main debate after the release was not revenue growth but gross margin. NVIDIA held Q2 gross margin at 75%, guided Q3 to 74%, and said Q4 could fall to 71% to 72%, citing memory costs that rose more than expected. That shift, at NVIDIA’s scale, can move quarterly gross profit by more than $10 billion from one modeling assumption to another, which is why margin has become the center of the market’s attention.

NVIDIA has introduced NVHBM, a new custom high-bandwidth memory technology, and folded it into its NVLink Fusion ecosystem as AI infrastructure demands shift beyond raw compute toward tighter coordination across processors, memory, interconnects, and cooling. AWS chip unit Annapurna Labs is the first partner to adopt the technology and said its next-generation in-house Trainium4 chip will support both NVLink Fusion and NVHBM. The announcement came alongside a separate commercial commitment: Amazon confirmed plans to buy up to 2 million next-generation high-end GPUs from NVIDIA between 2027 and 2028, covering Blackwell Ultra, Rubin, and Rubin Ultra. According to ABMedia, the arrangement reflects a changing relationship between hyperscalers and NVIDIA, where internal chip development and external GPU procurement now sit side by side rather than in direct opposition. ABMedia said NVHBM moves the memory controller into the base die of a 3D HBM stack, freeing die space on the XPU for more compute cores or cache. The report cited gains of up to 30% more memory bandwidth, about 15% lower HBM power consumption, and up to 25% more chip area released for compute. NVIDIA is also pushing the design toward a standardized specification across suppliers including SK hynix, Samsung, and Micron.