Why Elon Musk must acquire GlobalFoundries, Cambricon, and Cerebras to scale Tesla, Optimus, and xAI?

Elon Musk has stated publicly that he will ultimately require between 100 billion and 200 billion chips per year to support the combined needs of Tesla vehicles, Optimus humanoid robots, Starlink terminals, and xAI’s AI infrastructure. He has also expressed repeated frustration with the semiconductor industry’s inability to expand capacity quickly, noting that new fabs can take five years or more to come online—far too slow for the pace at which his businesses are scaling.

The scale of Musk’s semiconductor demand is unprecedented for a private industrial ecosystem. Tesla, Optimus, SpaceX, Starlink, and xAI together already consume more silicon than any vertically integrated company in history. Over the next five years, chip demand across these platforms will grow from several billion units annually into the tens of billions, ultimately approaching Musk’s stated range as autonomy deepens and humanoid robots move from pilot programs into volume production.

What makes this demand structurally different from prior industrial cycles is convergence. Tesla’s AI6 platform is not being designed as a single-purpose automotive processor, but as a unified compute architecture intended to serve vehicles, humanoid robots, and internal AI infrastructure. This convergence means silicon bottlenecks are no longer isolated problems. Any constraint in wafer supply, inference efficiency, or training compute propagates across Musk’s entire ecosystem simultaneously.

At this scale, silicon is no longer a sourcing issue. It is a control issue.

True nature of Musk’s chip demand
Musk’s reference to needing 100-200 billion chips per year is often misunderstood. This does not refer to high-end GPUs alone. It encompasses the full spectrum of semiconductors required to embed intelligence everywhere: inference processors, motor controllers, sensor ASICs, RFICs, PMICs, networking chips, and control logic.

A traditional vehicle contains several hundred chips. A Tesla vehicle now exceeds one thousand. A humanoid robot requires a similar number. A Starlink terminal uses several hundred RF and control devices. AI training clusters require thousands of advanced accelerators per deployment. 

When these requirements are aggregated across Tesla, Optimus, Starlink, and xAI, annual chip consumption quickly escalates into the billions even before full autonomy and robotics scale.

The center of gravity of Musk’s silicon demand is clearly visible. Tesla vehicles represent the largest single source of unit volume today, but Optimus robots rapidly emerge as an equally important driver as production scales.

-- Dr. Robert Castellano, Semiconductor Deep Dive, USA.