The Information reported that Meta Platforms META is in discussions to buy Google’s custom Tensor Processing Unit, or TPU, chips for artificial intelligence. Shares of AI chip suppliers Nvidia NVDA and AMD AMD were down as much as 4% and 7%, respectively, on Nov. 25.

Why it matters: Google’s TPUs appear to be gaining traction outside of Google’s internal use cases. We’ve long expected hyperscalers like Meta to develop custom AI processors (ASICs), but it’s possible that Meta could diminish its chip development and turn to Google for ASICs instead.

  • Meta is a large customer of Nvidia and an important emerging partner with AMD, the latter of which also aspires to gain AI processor share. These reports raise the risk, but not the certainty, that TPUs displace AI chips from these other two suppliers.
  • However, we’ve long believed in an AI world where Nvidia, AMD, and ASICs all have a place. These discussions don’t change this dynamic, in our view. That said, the relative market share of chips from all three sources is uncertain and will be based on future innovation and execution.

The bottom line: We maintain our fair value estimates of $240 for wide-moat Nvidia and $270 for narrow-moat AMD. We retain our Very High Uncertainty Ratings for both firms because AI dynamics are changing rapidly and based on opaque negotiations among tech leaders.

  • For Nvidia, we expect it to inevitably lose market share to ASICs like TPUs in certain use cases. However, the firm’s dominance in AI software and networking around these chips should keep Nvidia as the AI vendor of choice and the industry leader.
  • The reports are perhaps more worrisome for AMD, which needs to prove its AI capabilities to hyperscalers. That said, we view these reports are more likely a shift in Meta’s ASIC strategy to TPUs, rather than a signal that the Meta-AMD partnership is doomed.

Get Morningstar insights in your inbox