The AI chip race heats up: why Anthropic, OpenAI and Big Tech are designing their own silicon

The race to build the most powerful artificial intelligence is increasingly being fought not only in software but in silicon. According to reports, a growing number of the biggest names in the industry are moving to design their own chips, seeking more control over the hardware that powers their models at a time when demand for computing power has never been higher and cutting edge processors remain both scarce and expensive.

Anthropic joins the race

One of the most notable recent entrants is Anthropic, the company behind the Claude family of models. According to reports, Anthropic has been building an in house chip design team and intends to keep hiring engineers capable of co designing hardware and software together, a move that signals its ambition to shape the very processors on which its models run rather than relying entirely on outside suppliers.

The company has been careful to frame the effort as an addition rather than a break. According to reports, Anthropic is maintaining partnerships with Nvidia, AMD, Amazon Web Services and Google Cloud as part of what it describes as a multi chip strategy, stressing that its own design work is not a replacement for those partners but a way to add flexibility across its computing options.

A crowded field

Anthropic is far from alone in this pursuit, and the trend now spans much of the industry. According to reports, OpenAI recently unveiled its first custom AI chip developed alongside the semiconductor firm Broadcom, marking another major step in the movement of leading AI developers toward creating their own dedicated hardware.

Several of the largest technology companies have been travelling this road for some time. According to reports, Meta continues developing its MTIA accelerators, Amazon already designs its Trainium and Inferentia chips, and Google has spent years building its Tensor Processing Units, giving each of them a degree of independence in how they run their most demanding workloads.

Why build your own chips

The motivation behind this wave of custom silicon is rooted in the sheer scale of AI's appetite for computing. According to reports, the driving logic is that a single supplier, no matter how dominant, simply cannot satisfy all of the unprecedented demand generated by modern AI, which pushes companies to look for additional sources of the processing power they need.

The cost of relying on the most advanced off the shelf hardware helps explain the appeal. According to reports, high end Nvidia Vera Rubin graphics processors carry a price of around 55,000 dollars per unit, while a fully populated rack of 72 such chips can cost between roughly 7.8 million and 9.1 million dollars before networking is even taken into account.

The cost of going custom

Designing a chip in house, however, is far from a cheap alternative. According to reports, developing a custom AI processor can cost roughly 500 million dollars before manufacturing even begins, a figure that underlines why only the best funded companies in the sector are in a position to seriously pursue their own silicon ambitions.

That enormous upfront investment has to be weighed against the potential long term benefits. According to reports, the appeal of custom chips lies in greater control, tailored performance and more predictable access to hardware over time, advantages that the largest players calculate are worth the heavy initial spending given how central computing has become to their businesses.

The scale of those figures also helps explain who can realistically take part. According to reports, with design costs running into the hundreds of millions of dollars and top tier hardware priced in the tens of thousands per unit, the ability to build custom silicon is effectively limited to a small group of exceptionally well resourced companies, reinforcing the concentration of power among the sector's biggest names.

Not the end of Nvidia

Despite all the activity, the shift is not being framed as a move to sideline the current market leader. According to reports, companies pursuing their own designs are doing so to complement rather than entirely replace Nvidia, whose processors remain central to the industry, with custom chips serving as an extra option rather than a wholesale substitute.

What it means for AI

Taken together, these moves point to a future in which building leading AI increasingly means controlling the hardware underneath it. According to the available reports, the spread of custom chip projects across Anthropic, OpenAI and the biggest technology firms suggests that the contest for AI supremacy will be decided as much by who can secure and design the right silicon as by who writes the cleverest algorithms.