Anthropic Hires OpenAI's Second-Ever Chip Engineer as Both Companies Prepare for IPOs

In his public announcement, Chan praised his former team's capabilities even as he departed. "The density of hardware talent on that team is extraordinary, and I don't think there's a better chip design team anywhere," he wrote of OpenAI's custom chip group ^[1].

But he signaled a desire for a new challenge. "I haven't been able to shake the pull to climb a new mountain from the bottom again," Chan said, adding that he was "deeply impressed" with Anthropic's "talent, values, and ambition" and "already energized by the pace and intensity" of his first days at the company ^[1].

Chan's LinkedIn title at Anthropic reads "perplexity per picojoule," suggesting his work will focus on optimizing AI model performance relative to energy consumption — a key metric for inference efficiency that directly affects operating costs at scale ^[2].

Reuters reported in April 2026 that Anthropic was evaluating the possibility of designing its own AI chips but had not yet committed to a specific design or assembled a dedicated team ^[3]. Industry sources estimated that developing an advanced AI chip would cost roughly $500 million in engineering and validation ^[3].

Chan's hire suggests Anthropic is moving beyond the evaluation phase. His experience building custom silicon at both Tesla and OpenAI — two of the most ambitious chip programs outside of traditional semiconductor companies — makes him a natural anchor for any new effort.

Anthropic currently runs its Claude models on a mix of Google TPUs, Amazon Trainium chips, and Nvidia GPUs. The company has a long-term deal with Google and Broadcom for approximately 3.5 gigawatts of TPU-based compute starting in 2027, according to a Broadcom SEC filing ^[3]. Developing proprietary chips would give Anthropic the option to reduce dependence on any single vendor while optimizing hardware specifically for its model architectures.

OpenAI's partnership with Broadcom to develop custom AI accelerators — internally referred to as an "XPU" program — is significantly more advanced than Anthropic's chip exploration ^[3]. The collaboration targets 10 gigawatts of new data center capacity, with hardware deployment expected to begin in the second half of 2026 ^[2].

Losing an early hire from that program is notable but unlikely to derail the effort. Chan himself described the OpenAI chip team as the best in the industry ^[1]. OpenAI has attracted semiconductor talent from across the industry and has Broadcom's extensive chip-design infrastructure behind the partnership.

Broadcom, the publicly traded design partner behind custom AI chips for OpenAI, Google, and at least two other undisclosed hyperscale customers, trades at a market capitalization of roughly $1.83 trillion ^[6].

The talent move underscores the intensifying competition between the two AI labs as they prepare to become public companies. Anthropic confidentially filed its draft registration statement with the SEC on June 1, with most analysts expecting a listing in Q4 2026, possibly as early as October ^[4]. The company's annualized revenue run rate has reportedly grown from $9 billion at the end of 2025 to approximately $44 billion by late May 2026 ^[4].

OpenAI, which has not yet filed publicly, is widely expected to target a listing around September 2026, based on Wall Street Journal reporting from January ^[5]. Its March 2026 Series C round at an $852 billion post-money valuation drew investors including Amazon, Microsoft, Nvidia, SoftBank, and Andreessen Horowitz ^[5].

For both companies, demonstrating a credible path to improved margins will be essential to justifying near-trillion-dollar valuations. Custom chip programs are one of the most direct levers: purpose-built silicon for inference workloads can deliver significant cost-per-query reductions compared to general-purpose GPUs, directly improving unit economics at scale.

Whether Chan's arrival marks the beginning of a formal Anthropic chip program or a more incremental optimization effort remains to be seen. The company declined to comment on Reuters' earlier reporting about its chip exploration ^[3].

What is clear is that the talent war between the two leading AI labs has expanded from researchers and engineers into the specialized world of custom semiconductor design — a domain where experienced hires are exceptionally difficult to find and each one carries outsized strategic value.