The Next Trillion-Dollar Tech Company May Be an AI Chipmaker

I always remember a certain custom from Nvidia’s yearly GTC conference in San Jose. The audience reacts as though Jensen Huang has uncovered some ancient artifact when he enters the stage wearing a leather jacket and holds up a chip the size of a dinner plate. That scene would have been a curiosity five years ago. It’s the most significant event on the tech calendar today. In October of last year, Nvidia became the first company to reach $5 trillion in market capitalization. And whether it can be overthrown is, for some reason, the most fascinating topic of discussion in the industry at the moment.

The peculiar twist is that Nvidia may not resemble the next trillion-dollar tech company at all. It could be a silently employed custom silicon designer for the hyperscalers. It could be a licensor of chip architecture that doesn’t produce anything. It may even be a business you’ve never heard of before, one of the lesser-known brands making headlines through a quiet collaboration with Anthropic or a SPAC deal. The GPU era seems to have peaked—not in a collapse sense, but rather in the way an industry develops with the introduction of new, more specialized tools.

In this case, Google has been the silent leader. Originally designed for internal use with Google Gemini, its Tensor Processing Unit, or TPU, has since emerged as the preferred chip for businesses seeking to leave Nvidia’s ecosystem. Claude’s manufacturer, Anthropic, is working to deploy a gigawatt of TPU computing power this year. For some time now, OpenAI has made it clear that it wants its own silicon, built with Broadcom’s assistance and based on its own models. It’s worth sitting with that final detail. These custom AI chips are actually being co-designed by Broadcom for Alphabet, OpenAI, and other companies. Although it isn’t as well-known as Nvidia, with a market capitalization of almost $2 trillion, it is already a member of a group that most people were unaware existed.

Then there’s Arm Holdings, which designs the underlying chip architecture used in most of the world’s smartphones and, increasingly, AI data centers. Arm licenses designs rather than manufacturing chips. Arm powers Nvidia’s Grace chip. Likewise, Google’s Axion, Microsoft’s Cobalt, and AWS Graviton. Arm reported $1.14 billion in revenue during its most recent fiscal quarter, a 34% increase from the previous year. $620 million was made from royalties alone. Given its current valuation of about $140 billion, it would require a roughly seven-fold increase to reach trillion-dollar status. That would have sounded ridiculous ten years ago. With Neoverse CPU deployments now surpassing one billion units, it seems more like math than aspiration.

Watching this unfold, it’s hard not to think about how the 2000s played out with Cisco and the networking boom. Everyone believed that the old-fashioned monarch would retain the throne. Broadcom, Arm, Qualcomm, even smaller players like AMD and Forge Nano — they’re all building around the idea that general-purpose AI compute is going to fragment. A chip made for every task is not what OpenAI wants. It seeks one that is power-efficient, tuned to its models, and designed for inference. The economics tilt is at that point.

Energy costs, geopolitics, Chinese competition, and the longevity of existing AI architectures are some of the unknown factors that will determine whether any of these businesses truly surpass a trillion dollars. However, Nvidia’s $5 trillion valuation alone is unlikely to be the last chapter. The next one is probably being built by someone else in a quiet factory somewhere, and most of us won’t find out until the news breaks.