Anthropic and AI infrastructure provider Nscale are reportedly on the verge of announcing a second compute deal worth 45 billion dollars, according to Bloomberg reporting relayed by TechCrunch on October 19. The talks would come a little over a month after the two companies announced a first 45 billion dollar compute deal. Bloomberg's sources, cited by TechCrunch, cautioned that the talks are still underway and the details could change, and TechCrunch said it had reached out to both companies for comment. Treat this as a reported deal in negotiation, not a signed one, but the direction of travel is unmistakable.
The reported deal would be the latest in a long string of giant compute commitments Anthropic has struck over the past year, and it lands while both companies are pursuing public listings. Anthropic is still expected to go public this year. Nscale is also seeking an IPO, although Bloomberg reported on Monday that the company is facing investor pushback over a proposed valuation of 30 billion dollars. TechCrunch noted that the sheer scale of AI compute spending has spooked a segment of investors who worry the industry is spending too much for too little profit. Anthropic, so far, has pressed ahead anyway, on the stated belief that securing capacity now is what wins later.
The same day brought a different kind of capacity bet. Tenstorrent, the inference chip startup led by veteran microprocessor engineer Jim Keller, raised 1.5 billion dollars at a 6.2 billion dollar valuation, TechCrunch reported, more than double the 2.6 billion dollar valuation it fetched in late 2024. The round was led by Trace Capital and Innovation X, with Cadence Ventures, Eclipse, Fidelity, SanDisk, and T. Rowe Price also participating. Keller told Bloomberg the company chose to stay independent and raise instead, five months after reported acquisition interest from Intel and Qualcomm at valuations up to 5 billion dollars, after showing its Galaxy servers could run DeepSeek models. The new capital is aimed at chips tailored to serving open-weight models and video generation.
Policy tightened on the other side of the stack the same day. Reuters reported on October 19 that the U.S. Department of Homeland Security is ordering the removal of DeepSeek products, and third-party services built using the Chinese company's models, from U.S. civilian agency IT systems, in a notice published Monday. DHS is also banning new acquisitions of those products and services. The department said DeepSeek models pose multiple threat vectors and raised concern that their use by government employees or vendors could result in sensitive data being transferred to China. For any vendor selling into government, model provenance is now a formal procurement question, not a talking point.
For people building with AI, these three stories are really one story about who controls supply. Frontier capacity is being locked up in reported nine- and ten-figure commitments before it exists, which means access, pricing, and rate limits for everyone downstream will be set by deals most builders never see. At the same time, Tenstorrent's raise is a reminder that the open-weight ecosystem is attracting serious hardware money aimed squarely at inference, the part of the bill builders actually pay every day. And the DHS order shows compliance teams will increasingly ask where a model came from, not just what it scores. Plan for all three: diversify providers where you can, keep an open-weight fallback runnable, and know the provenance of every model in your stack before a customer or an agency asks.
