Who dares to compete with Nvidia?
Everyone thinks of AMD when looking for Nvidia's competitor...
But new data on the AI accelerator market paints a much more interesting picture,
The twist is that the company competing with Nvidia isn't even trying to compete with it directly....
Its name is Broadcom.
According to DIGITIMES' forecast, approximately 25.95 million units of advanced AI accelerators for the cloud are expected to be shipped in 2027.
Nvidia is expected to lead with approximately 9.1 million units, a 35.1% share.
Right behind it is Google.
Google's TPU chips are expected to reach approximately 8.01 million units, a 30.9% market share.
If this forecast materializes, Google will be far from being "just another Nvidia customer" – it will become one of the most central forces in the AI accelerator market.
And here, this is exactly where Broadcom enters the picture.
Google doesn't build all its TPUs alone
Broadcom is a key partner in the development and supply of Google's TPU chips, and for years the two have built a deep relationship around custom chips, connectivity, and AI infrastructure.
If Google succeeds in significantly increasing its use of TPUs,
Broadcom is on the side that provides the infrastructure for a competitor trying to take market share from Nvidia.
And the story here is even more serious when looking at the numbers:
Google is expected to reach 30.9% of the market, compared to Nvidia's 35.1%.
Amazon, as another example of a cloud giant developing its own chips, which comes in third after Google, is expected to reach only about 9.8% of the total market share.
And this illustrates how significant a force Google has become in the field of custom AI chips.
But there's an even more interesting point here for Broadcom:
The more successful Google is with TPUs, the more it can increase its reliance on internal chips, and as it develops new generations of TPUs and expands its AI infrastructure, Broadcom is in a position to benefit from this growth through the supply chain.
Simply put, anyone who only looks at "who is winning against Nvidia" might overlook the company profiting from selling infrastructure to competitors in the race.
Nvidia sells the accelerators.
Google builds the alternative.
And Broadcom provides some of the technology that allows this alternative to exist on a massive scale.
The real revolution in the AI market may be the shift from a situation where almost everyone buys accelerators from Nvidia to a situation where cloud giants develop their own chips,
but still need a company like Broadcom to turn their personal chip dream into reality.
Therefore, if Google truly approaches Nvidia to the level of 30% of the market, one of the most interesting questions will be:
Who stands to profit as Google succeeds in reducing its dependence on Nvidia?
And here, the answer may be Broadcom.
We are monitoring.