The roster notice was short, but the message carried. Clive Chan, a senior engineer who helped launch OpenAI’s effort to design its own chips, has left the company and this week joined Anthropic, according to people familiar with the matter. Chan was one of the first hires on OpenAI’s silicon project, a figure described in company circles as employee No. 2 of the initiative.
Chan’s background spans the two most demanding hardware environments in AI. At Tesla, he worked on the Autopilot deep-learning infrastructure team, handling GPU optimization and cluster scheduling for the systems that train and run the company’s self-driving networks. It was work that combined the painstaking craft of squeezing performance from thousands of chips with the operational pressure of keeping them running in production. He joined OpenAI in early 2024 to help build the custom chip program that the company has since made central to its long-term cost strategy.
The move, reported by seven media outlets in an aggregation by Readhub, is small in numbers but large in symbolism. It lands at a moment when the competition among AI labs has moved from models and products to the machines underneath both.
OpenAI’s chip ambitions are among the most closely watched engineering projects in the industry. The company has been working with Broadcom and Taiwan Semiconductor Manufacturing Co. on custom accelerators, part of a broader push to reduce its dependence on Nvidia and to lower the cost of running models at scale. A senior hardware leader leaving that effort, whatever the reason, raises questions about momentum at a project with billions of dollars of expected spending attached.
For Anthropic, the hire fills an obvious gap. The company has built its models largely on rented computing power, drawing capacity from Amazon and Google, its two largest investors. Bringing in an engineer who has built and run GPU fleets at Tesla and OpenAI signals that Anthropic wants more control over its hardware destiny rather than a permanent role as a tenant of other companies’ clouds.
The defection also reveals the shape of the AI talent market. The scramble for engineers has moved from model researchers to the people who keep thousands of accelerators healthy, schedule training runs and optimize the flow of data between chips. These are the skills that determine whether a lab’s compute bill is a rounding error or a crisis, and they are among the hardest skills to hire. Compensation packages in this corner of the market have climbed steeply, and the poaching shows no sign of slowing.
Analysts said the recruiting cuts in both directions. Anthropic has spent the past year hiring aggressively from OpenAI, and OpenAI has returned the favor; the difference now is that the battleground has extended to hardware. A person familiar with Anthropic’s plans said the company is building out a systems engineering group to match its research ambitions, though the group’s head count remains small.
The broader signal is strategic. Anthropic has long argued that its edge lies in model quality and safety engineering. But the center of gravity in the industry is shifting toward infrastructure, and the companies that control chips, clusters, power and cooling set the terms for everyone else. Chan’s arrival suggests Anthropic’s leadership has drawn the same conclusion and is prepared to spend to act on it.
For OpenAI, the loss lands in an unusually busy season. The company is preparing for a reported initial public offering, is in discussions with the White House over a possible government equity stake, and is committing ever larger sums to the data centers and chips it needs to stay ahead. Hardware talent is a small line item in that picture, but it is not a trivial one.
Chan’s time at Tesla included exposure to Dojo, the company’s attempt to build its own training supercomputer, an effort that made Tesla one of the few companies outside the big chipmakers to design large-scale AI hardware. Engineers with that mix of experience, custom silicon plus fleet operations, are rare, and their value has climbed as the industry’s infrastructure bills have ballooned.
Compensation in this corner of the market has followed the demand. Recruiters say senior hardware engineers at frontier labs now command packages that rival those of top model researchers, and the bidding has pulled in engineers from chipmakers, cloud companies and the autonomous-driving industry alike. For Anthropic, hiring Chan is also a signal to its own investors that it intends to spend on infrastructure, not just on models.
Chan joins Anthropic as the company files confidentially for its own IPO, a deal that will test whether investors accept a frontier lab whose compute is rented from its own investors. His job, if his record is any guide, will be to help change that math from the inside.
For the engineers who remain at OpenAI, the departure is a familiar feature of a hyper-competitive market: talented people move, and the work continues. The chip race between the two labs will not be decided by any single hire. But the movement of people like Chan, the kind of engineer who makes large systems work, is the clearest evidence that the next phase of the AI competition will be fought in silicon, not just in research papers.
