Amir Salek — the man who built Google’s TPU from scratch and ran it through seven generations — joins Anthropic’s compute team. This is a hire, not a chip. But the direction it points is unambiguous.
Amir Salek’s TPU program ships its first-generation custom AI accelerator, beginning Google’s decade-long push to cut dependence on Nvidia for inference and training.
2022 — Cerberus
Salek departs Google after delivering the first seven TPU generations, joining Cerberus Capital Management as a senior managing director — a private-equity posting that kept him adjacent to capital-intensive infrastructure.
June 2026 — Broadcom / Apollo / Blackstone SPV
A $35 billion special-purpose vehicle — structured with Broadcom, Apollo, and Blackstone — leases Google TPU capacity to Anthropic at scale, keeping custom silicon off Anthropic’s balance sheet but out of its design control.
Early August 2026 — First Reports
Reporting surfaces that Anthropic has begun assembling an in-house chip-design team — the first indication the lab is treating silicon as an internal capability, not a vendor relationship.
August 21, 2026 — Bloomberg Confirmed
Anthropic hires Salek onto its compute team, reporting to James Bradbury, as Bloomberg reports the lab is “laying the groundwork for its own chips” in a “push into hardware.”
Bloomberg reported on August 21 that Anthropic has hired Amir Salek — the founder of Google’s custom-chip program and its leader through the first seven generations of TPUs, until his departure in 2022 — onto its compute team. Salek will report to James Bradbury. Bloomberg’s framing is deliberate and measured: Anthropic is “laying the groundwork for its own chips” and making a “push into hardware.” That is the full factual perimeter of what is confirmed.
The hedges matter as much as the hire itself, and they should be front-loaded. Anthropic has not announced a chip, a tape-out, a design partner, or a timeline. This is one senior person joining an existing compute team, not a company pivoting overnight into semiconductors. Designing a competitive AI accelerator is a multi-year, capital-heavy undertaking with a meaningful failure rate; the right hire is necessary but nowhere near sufficient. Anthropic also remains multi-sourced today — purchasing compute from Nvidia, Google TPU, and Amazon Trainium — and this hire changes none of that current supply picture. Reading it as Anthropic dropping Nvidia would be the wrong inference.
What the hire does is confirm a direction. It is a continuation of the early-August reporting that Anthropic had begun assembling an in-house chip-design team. And the specific person chosen — not a chip engineer from a startup, but the founder of the most commercially consequential custom AI accelerator program ever built — signals that Anthropic’s compute ambition, whatever its current shape, is being approached with institutional seriousness.
The key insight: Anthropic is not building a chip. It is hiring the person who knows how to build one — and pointing him at the precise problem that limits its ability to scale models: owning none of the silicon it depends on. The hire is a declaration of strategic intent, not an engineering milestone.
The Structural Read
Five analytical frames converge on this hire. They are our framing of the strategic logic — not statements Anthropic made — and they should be read as such.
The model lab reaches down the stack. Anthropic is a frontier-model company. Its core product is intelligence, not silicon. But the stack that produces intelligence runs from software harnesses at the top, through model weights, through training and inference frameworks, down to the accelerators themselves. Anthropic has historically operated at the upper layers, renting the lower ones. Hiring Salek is the clearest signal yet that it intends to extend its control further down — not all the way, not immediately, but directionally. In the Map of AI framework, the question is always where a company sits in the nine-layer stack and which direction it is moving. Anthropic is moving down.
The pincer. On the same morning this news circulated, Nvidia — a hardware company — published results showing its AVO agentic system lifting a model from 30% to 100% on a reasoning benchmark, demonstrating that it is reaching up the stack into the software harness layer. Now Anthropic — a software-and-models company — reaches down into silicon. The two moves rhyme: each incumbent is pushing into the other’s territory because the middle of the stack is where the margin is being competed for. The pairing is thematic, not coordinated — there is no connection between the two events — but the structural pattern they illustrate together is real. See the NVIDIA AVO analysis here.
Compute is the binding constraint. Anthropic already leases custom TPU silicon at enormous scale — the $35 billion Broadcom-Apollo-Blackstone SPV struck in June 2026 structures that capacity without putting it on Anthropic’s balance sheet. That arrangement gives Anthropic scale but not control: it does not own the design, influence the roadmap, or determine the long-run cost per token. Hiring the TPU’s founder is the logical next step in the same strategy: to eventually own what it currently leases. See the Broadcom-Apollo-Anthropic debt structure analysis.
Follow Google’s own playbook.Google built the TPU for one reason: to cut its dependence on Nvidia and reduce the cost of serving models at scale. Amir Salek ran that program. Anthropic hiring him is, quite literally, purchasing Google’s institutional knowledge of how to execute that playbook. The analogy is imperfect — Google’s compute scale dwarfs Anthropic’s, and internal chip programs require sustained capital commitment that Anthropic has not announced — but the strategic logic is the same: vertical integration into silicon as a cost and supply control mechanism. See the Anthropic Q2 2026 revenue and enterprise crossover analysis.
De-risk Nvidia. Owning a chip design means controlling cost, controlling supply, and controlling the roadmap — the three variables that determine a model lab’s ability to scale without permission from a dominant vendor. This is the through-line of the whole compute-sovereignty story. The strategic logic does not require Anthropic to ship a chip in twelve months; it requires only that it begins accumulating the capability to do so eventually. See Beyond NVIDIA’s Moat.
The industry’s biggest players — from both the hardware and software sides of the stack — increasingly believe that a frontier model company and a frontier chip company cannot remain fully separate entities. The margin, the control, and the long-run competitive position live in the integration. Anthropic is not there yet. But this hire is the first structural step toward it, and the direction it points does not change regardless of whether a chip ever ships.
“Anthropic has tapped a Google chip veteran as part of its push into hardware, as the AI lab lays the groundwork for its own chips.”
Three Implications
FOR ANTHROPIC — Compute sovereignty is now a stated priority, not a roadmap item
The hire of Salek, coming weeks after the early-August chip-team reports, establishes that Anthropic’s compute strategy is moving from vendor management toward capability building. That does not mean a chip ships soon — it means the capability to design one is being accumulated internally. The long-run implication is that Anthropic’s cost structure, scaling decisions, and supply security could eventually be determined in-house rather than negotiated with Broadcom, Nvidia, or Amazon.
FOR THE STACK — Vertical integration pressure intensifies on every layer
When the largest model labs begin hiring chip architects, it changes the calculus for every company that currently sits between model and metal — cloud providers, contract chip designers, and inference infrastructure vendors. None of those relationships disappear quickly, but the trajectory of where control flows in the stack shifts. The Broadcom-Apollo SPV is a bridge financing structure, not a permanent arrangement; Salek’s hire suggests Anthropic is already thinking about what comes after it.
FOR NVIDIA — The threat is directional, not immediate
Anthropic is not dropping Nvidia. It buys from Nvidia, Google, and Amazon today, and this hire changes none of that supply. But the pattern — Google built TPU to reduce Nvidia dependence; Anthropic hires Google’s TPU founder — is not ambiguous about the long-term intent. The time horizon for in-house silicon to matter is measured in years, not quarters. What changes now is the signal: another well-capitalized frontier lab has decided that renting compute indefinitely is not the destination.
Business Engineer Framework
The Map of AI Redrawn
The nine-layer Map of AI tracks where companies sit in the stack — and which direction they are moving. Anthropic’s hire of Amir Salek is a textbook downward stack integration move: a model-layer company extending control toward the silicon layer. The Map of AI Redrawn framework explains why that direction is now the central strategic contest in the industry, and what it means for every company currently occupying the middle layers. Read the full analysis alongside the Beyond NVIDIA’s Moat piece to see the full competitive picture.
Anthropic has not built a chip. It has hired the person who knows how — the founder of the most consequential custom AI accelerator program in the industry’s history — and pointed him at its most binding strategic constraint. Whether or not silicon ever ships under an Anthropic logo, the hire tells you something durable about where the company believes leverage lives: not in model weights alone, but in owning the infrastructure that runs them. That is the same conclusion Google reached in 2016. It took Google a decade to make it matter. Anthropic is starting to count.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
