Google's In-House AI Chip Strategy Could Be a Bigger Threat to Nvidia Than Investors Think. Here's Why.
Google's in-house Tensor Processing Units (TPUs) pose a significant threat to Nvidia's AI chip dominance. By offering cost-efficient custom processors and expanding its 'neocloud' rental model via a Blackstone joint venture, Google aims to capture market share. This shift, mirrored by other tech giants like Amazon and Microsoft, could erode Nvidia's 86% market share and compress its high profit margins as hyperscalers prioritize internal efficiency over third-party GPUs.
Alphabet (GOOGL 0.50%) (GOOG 0.29%) has been developing its own Tensor Processing Units (TPUs) for years. But it wasn't until recently that the company started to see these processors not just as a side project but as a real alternative to Nvidia's (NVDA +3.90%) graphics processors.
The shift could be consequential for Nvidia, as Google focuses more on using its own processors and renting them to other AI companies.
Here's why Google's TPUs could be a bigger threat to Nvidia than investors might think.
Image source: Getty Images.
Custom processors are really good at AI compute
It used to be that graphics processors were the hands-down winners for all things artificial intelligence.
But what AI companies have found recently is that designing their own custom processors can be a great way to achieve fast and efficient AI computing.
For example, Google's TPUs can handle AI workloads at an estimated total cost savings of up 30% compared to using chips made by other hyperscalers. That's because the custom processors can be designed specifically for how its Gemini AI model processes information.
Alphabet is spending up to $190 billion in capital expenditures this year, and management has said, "Next year, we expect it to significantly increase compared to 2026." Drastically reducing AI compute costs could help Google eventually run its AI data centers far more efficiently, and make its massive AI investments eventually worth the high cost.
And Google isn't the only one doing this. Many tech companies are looking more to custom processors to make their AI models more efficient and reduce costs. Space Exploration Technologies (SPCX 4.51%) is building what some are calling a "sovereign AI" in which SpaceX owns everything from the chip design and manufacturing to the AI model itself. And others, like Amazon and Microsoft, are designing their own AI processors as well.
All of which means that Nvidia could lose its dominance in the AI chip design market. It's not inevitable, of course, but as AI investments have skyrocketed, tech giants are trying to figure out how to make these investments pay off.
NASDAQ: GOOGL
Key Data Points
Why Nvidia can't take this lightly
Google announced a few months ago that it's starting a joint venture with Blackstone to deploy 500 megawatts of its own TPU capacity by 2027 and added that it has "plans to scale significantly over time."
That level of capacity shows Google is serious about using its TPUs as more than just a pet project.
What's more, it plans to rent some of that capacity to other tech companies. This system is called a neocloud business model, in which a tech company uses its own processors and data center and rents some of its capacity out to others. The rapidly expanding neocloud market could take 20% of the AI cloud market by 2030.
If more tech companies pivot to renting out Google's TPUs, or using their own processors, it will not only hurt Nvidia's market share in the AI data center space -- currently around 86% -- but it could also bring Nvidia's margins down.
Nvidia enjoys an enviable gross profit margin of about 74%, but with more competition looming from Google's TPUs, it might not be that long before Nvidia can't command the same pricing power it once did. And that could be one of the biggest threats to Nvidia's dominance in a long time.
