Nvidia acquired Groq, the LPU inference chip startup, and launched its own Vera CPUs this week — completing a vertical integration play that extends well beyond American borders.1
Broadcom is simultaneously building custom AI processors for Alphabet and OpenAI, with Arm joining the OpenAI collaboration.1 Three of the world's most powerful AI operators are now co-designing the silicon that runs their models.
The consolidation concentrates AI compute authority in a small cluster of US firms with dedicated manufacturing through TSMC in Taiwan. Alphabet's TPU program, developed with Broadcom over years, is the template. What has changed is scale and simultaneity: multiple hyperscalers are locking in custom silicon relationships at the same moment.
For AI programs outside this circle — European sovereign AI initiatives, Gulf-state cloud buildouts, Indian infrastructure ambitions — the strategic gap is widening. Custom chips tuned to specific model architectures deliver better performance-per-watt than commodity GPUs. At billions of inference queries daily, that efficiency advantage compounds into significant cost separation.
Tiger Global has increased stakes in Nvidia, Broadcom, and TSMC concurrently.1 The investment pattern signals where infrastructure value accrues: not in GPU purchasing, but in custom silicon partnerships and dedicated manufacturing capacity.
Taiwan's position is central and exposed. TSMC manufactures the chips undergirding this consolidation. Geopolitical risk around the Taiwan Strait is therefore also AI infrastructure risk — a calculation European and Asian governments are actively pricing into national technology strategies.
China's push for chip independence, accelerated by US export controls, represents the only parallel vertical integration effort at comparable ambition. Chinese fabs remain multiple generations behind TSMC in leading-edge capacity, limiting near-term competitive parity.
Nvidia has an estimated 18-24 month window before custom silicon deployments create structural pressure on its commodity GPU business.1 The Groq acquisition and Vera CPU launch are direct responses to that clock.
The race to own AI's silicon stack is no longer a US domestic story. It is the central infrastructure competition shaping AI capacity worldwide.
