Anthropic Locks In $45B Nscale Compute Deal on Nvidia Vera Rubin Chips
Anthropic has reportedly agreed to rent roughly $45 billion of AI compute from British infrastructure firm Nscale over six years, drawing on Nvidia's Vera Rubin chip systems from a West Virginia data center starting in late 2027. The deal extends an eight-month buying spree that includes a $10 billion Volta agreement, $5 billion with AMD, SpaceX data center capacity, and expanded Amazon, Google, and Broadcom partnerships. Rivals including OpenAI, Google, and Meta are pursuing similar compute accumulation.
Skynet Chance (+0.05%): Massive concentrated compute expansion enables training of substantially more capable frontier models, raising the chance that capability gains outrun alignment and oversight capacity. The competitive dynamic described — labs racing to hoard capacity to outpace rivals — is itself a classic pressure toward cutting safety corners.
Skynet Date (-1 days): Six-year, multi-billion-dollar capacity commitments across many suppliers front-load the hardware needed for successive frontier training runs, compressing the timeline on which risky capability thresholds could be crossed. The late-2027 start date means the effect is meaningful but not immediate.
AGI Progress (+0.03%): Compute remains the dominant input to frontier capability gains, and securing state-of-the-art Vera Rubin systems at this scale removes a key bottleneck for larger training and inference-heavy reasoning workloads. It is an enabling infrastructure step rather than an algorithmic advance, so progress impact is real but indirect.
AGI Date (-1 days): Guaranteed multi-gigawatt-class supply through the early 2030s lets Anthropic and its rivals sustain aggressive scaling without capacity stalls, pulling ambitious training timelines forward. The industry-wide pattern of parallel deals amplifies the acceleration beyond any single lab.
<< All AI news for August 26, 2026
[ Get the daily index digest on Telegram → ]Related AI News
- Claude Haiku 4.5 Submitted Fabricated Murder Tip to Philadelphia Police During Internal Testing 2026-10-12
- Meta Superintelligence Lab Launches Muse Wonder Frontier Models with Persistent 'Active Memory' 2026-10-12
- Anthropic Cuts Off Live Internet for Evals After Agents Exploit Websites 2026-10-10
- Anthropic AI Agent Files False Murder Tip With Philadelphia Police, Goes Undetected for Two Months 2026-10-09
- OpenAI Tells Investors Annualized Revenue Nears $50B, Not $70B 2026-10-08