Claude-maker Anthropic has hired Amir Salek, founder of Google’s custom chip programme
Anthropic, the company behind the Claude AI platform, has appointed Amir Salek to its compute engineering team as it accelerates plans to design purpose-built AI processors. Salek will report directly to James Bradbury and is expected to play a central role in shaping the company’s in-house silicon strategy.
Seasoned chip leader joins Anthropic
Salek is widely recognized for launching and scaling Google’s custom silicon efforts, where he helped deliver the first seven generations of Tensor Processing Units (TPUs) to the company’s data centers. His background spans decades of leadership in advanced semiconductor design and operations, with earlier roles at Nvidia and, more recently, senior leadership at private equity firm Cerberus Capital Management.
Why Anthropic is moving into custom silicon
Anthropic currently sources compute from multiple providers—including Nvidia, Google, and Amazon—but is laying the groundwork for an internal silicon division. Building proprietary chips is intended to give the company greater control over performance, efficiency, and cost, while helping mitigate the impact of industry-wide chip constraints. Tailored architectures can also be optimized for the demanding training and inference needs of frontier AI models, potentially improving throughput and lowering total cost of ownership.
Part of a broader industry shift
The move reflects a sector-wide trend as leading AI companies seek deeper control over their compute stacks. Rival OpenAI has developed a custom processor, known as Jalapeno, in collaboration with Broadcom, with initial production expected to ramp later this year. Across the industry, vertical integration—spanning model research, software, hardware design, and data center operations—has become a strategic priority to ensure predictable access to compute and to unlock model-specific performance gains.
Balancing in-house development with third-party capacity
While investing in internal chip design, Anthropic is simultaneously expanding its access to external compute. The company has arranged significant silicon purchases, including an agreement with UK-based startup Fractile valued at roughly $250 million for initial units. It has also secured large-scale data center capacity through partnerships with providers such as Riot Platforms and Volta Infra Holdings. These arrangements aim to guarantee the server resources necessary to support rapid model training cycles, growing user demand, and the rollout of new AI capabilities.
What Salek’s hire signals
Bringing Salek on board underscores Anthropic’s intent to control more of its computational foundation—combining custom hardware with robust third-party capacity to meet near-term needs. With continued pressure on supply chains and escalating demands from increasingly capable models, the company’s dual-track approach positions it to innovate on chip design while maintaining the flexibility and scale required to operate at the frontier of AI.
As the in-house silicon program takes shape, expect Anthropic to refine its hardware-software co-design strategies, deepen collaboration across its research and infrastructure teams, and pursue performance targets aligned with its safety and reliability objectives for Claude and future model families.
