According to media reports, OpenAI plans to buy Google’s TPU chips. Oracle has announced that starting in 2027, it will offer the same chips to customers through Oracle Cloud, making it the first hyperscaler outside Google to do so. For the market, a real alternative to Nvidia has appeared. For users, it raises hopes of cheaper computing. The big question is whether those savings will trickle down to what users pay or stay inside the companies.
Hardware for Rivals
OpenAI has already rented TPUs through Google Cloud, and now the conversation has shifted to buying them directly. Oracle will deploy TPUs in its own data centers and let customers use them. In doing so, Google is expanding sales of its own chips, which it mostly used in-house until now.
The paradox is that Google could make money from OpenAI even as ChatGPT eats into Google Search and competes with Gemini for users. In other words, Google used to defend its search against outside AI models, and now it sells infrastructure to the companies gunning for it.
A Second Major Player
Until now, Nvidia was almost the only place to buy AI hardware. Now another major competitor is emerging, and it will change the terms of negotiations.
That doesn’t mean OpenAI and others will walk away from Nvidia. More likely, big AI companies will gain more flexibility to split their computing workloads across different platforms. The fiercer the rivalry, the stronger the case that hardware and cloud pricing will be revisited.
When Will Prices Fall, if Ever?
Google touts the efficiency of its new TPUs, and the company keeps bragging about how much cheaper it’s getting to run AI. Yet we still pay the same as a year ago, and we keep hitting usage limits. So when will those publicized savings actually reach users?
Cost cuts at these companies don’t guarantee price cuts. The savings could go into model development or higher profits. With competition only starting to take shape, it’s too early to talk about prices dropping any time soon.
Alphabet will release its third-quarter earnings on the evening of October 28. We’ll see how much Google earns from its infrastructure ambitions and whether it can turn the billions it has poured into AI into real money — if, of course, it discloses TPU economics.
