Earlier this week, Cathie Wood highlighted the intensifying competition for Nvidia Corp as ARK Investment Management forecasts that custom AI chips could command more than one-third of the compute market by the end of the decade.
Frank Downing, ARK’s director of research for next-generation internet, said in a post on X that the firm expects “over a third of the compute market will be custom silicon by 2030.”
He defined custom as non-GPU chips — effectively alternatives to products from Nvidia and Advanced Micro Devices, Inc. — though he noted that industry lines are “blurring.”
non-GPU in this case
I would say non-Nvidia/AMD but now that Nvidia effectively bought Groq lines are bluring
Don’t Miss:
“Everyone knows Google’s TPU, but Amazon is the sleeping giant that is waking up,” Downing wrote.
A chart shared by Downing shows traditional servers rapidly losing share to accelerated computing, with application-specific integrated circuits, or ASICs, gaining ground alongside GPUs through 2030.
We predict that over a third of the compute market will be custom silicon by 2030. This announcement is another step in that direction.
Everyone knows Google’s TPU, but Amazon is the sleeping giant that is waking up.
Have a blog coming soon that dives deeper into this. https://t.co/aR7QgFFcf7 pic.twitter.com/wuv6DwwJFP
Trending: Disney Was Built on Character IP — This Pre-IPO Company Is Using the Same Playbook
Wood amplified the message, sharing the post and adding: “Competition for Nvidia.”
Competition for Nvidia. https://t.co/siQ7FHsFta
The comments follow a multi-year partnership between Amazon.com Inc. and OpenAI.
Amazon committed up to $50 billion to the ChatGPT maker and expanded an existing compute agreement by $100 billion over eight years.
A key element of the deal centers on OpenAI’s use of Amazon’s custom Trainium chips, including next-generation versions expected in 2027. OpenAI will consume roughly 2 gigawatts of Trainium capacity, underscoring the scale of the commitment.
See Also: 1.5 Million Users Are Already Working Inside This AI Platform — Investors Can Still Get In
Meanwhile, Alphabet Inc.’s Google continues positioning its Tensor Processing Units as an alternative to Nvidia’s GPUs, with reports that Meta Platforms, Inc. has agreed to lease TPUs for advanced AI development.