Anthropic Hires Google Chip Veteran, Eyes Custom Silicon

Three developments in 48 hours sketch a company accelerating vertically: Anthropic is hiring to build its own chips, tapping outside debt to fund infrastructure, and quietly giving enterprise customers more control over their data.

Key points

• Anthropic hired Amir Salek, the founder of Google’s chip program, signaling a serious push toward in-house semiconductor design.
• The move aims to improve model performance, reduce costs, and cut dependence on external AI chip suppliers.
• Broadcom is assembling more than $60 billion in debt to secure chips and compute capacity for Anthropic and other AI firms.
• Anthropic separately updated its enterprise data retention policy, allowing customers to store 30-day interaction logs on their own cloud infrastructure.
• The three moves together reflect a company trying to control more of its cost structure and go-to-market story ahead of a likely IPO.

What does hiring a Google chip founder actually signal?

Amir Salek did not run a minor program at Google. He founded the chip effort that eventually produced Google’s Tensor Processing Units, hardware that became central to how Google trains and serves its own AI models. Anthropic bringing him in is not a research hire. It is an organizational bet that custom silicon will be a meaningful part of the company’s future architecture.

The strategic logic is familiar. Relying entirely on Nvidia or other third-party suppliers means paying market rates and sitting in the same queue as every other hyperscaler and AI lab. Companies that own their silicon supply chain, at least partially, gain pricing leverage and can tune chips specifically to their model architectures. Apple, Google, and Amazon have all followed this path. Anthropic appears to be starting that journey now.

The timing matters for investors watching the IPO trajectory. Custom chip programs are expensive and take years to yield production hardware. Anthropic is signaling a long-term capital commitment, which adds to an already substantial infrastructure spend visible in the $1.3 billion Texas data center loan reported earlier this week.

Is the Broadcom debt deal a feature or a warning sign?

Broadcom assembling over $60 billion in debt to fund chip and compute procurement for Anthropic and others is a striking number, and Bloomberg framed it explicitly in terms of “mounting financing strains and circular-funding risks” in AI infrastructure.

The circular risk is worth spelling out. AI labs raise equity and debt to buy compute. Chip and infrastructure suppliers take on their own debt to expand capacity for those labs. If revenue growth at the lab level slows or misses projections, the debt sitting across the entire supply chain becomes harder to service simultaneously. Anthropic is not directly responsible for Broadcom’s balance sheet, but it is named as one of the anchor relationships justifying that capital raise. That linkage will draw scrutiny from any serious IPO underwriter or public market investor.

Wall Street’s appetite for absorbing this much AI-related debt at once is also, per Bloomberg’s own framing, under strain. That is worth monitoring as Anthropic works toward a potential credit line and public offering.

What does the enterprise data policy change actually do?

The update is narrower than a headline rewrite of Anthropic’s data practices, but it is commercially meaningful. Enterprise customers using Anthropic’s most capable Claude models are now permitted to retain 30-day interaction logs on their own cloud infrastructure rather than having that data sit with Anthropic.

For regulated industries, legal, financial services, healthcare, this kind of customer-controlled data residency is often a procurement requirement, not a preference. Anthropic reducing its own data exposure while giving enterprises more security control removes a meaningful objection in those sales cycles. It also positions Claude more competitively against offerings from Microsoft and Google that can lean on existing enterprise cloud relationships to make similar guarantees.

Together with the chip hire and the infrastructure financing picture, the policy change fits a company that is methodically closing the gaps between its current product and what large enterprise and government customers need before they sign long-term contracts.