Every question someone asks ChatGPT has a cost.
To produce responses, pictures, and code, OpenAI requires massive amounts of processing power, and a large portion of that infrastructure has traditionally relied on pricey processors such as Nvidia’s (NVDA).
According to reports, OpenAI is now getting ready to purchase AI processors from Alphabet’s (GOOGL) Google, one of Nvidia’s main up-and-coming competitors.
Because Google’s Tensor Processing Units, or TPUs, are increasingly being promoted as a more affordable solution to run AI workloads, the possible discount is significant to customers.
A $20 ChatGPT Plus membership might not be cheaper, even with lower infrastructure expenses. Still, OpenAI may have greater leeway to maintain lower costs, raise use limits, or offer more powerful models without charging clients more.
OpenAI plans to purchase Google’s AI chips after already beginning to rent TPUs in 2025, Reuters, citing The Information, reported.
This case makes a dispute among some of the world’s largest IT firms seem much more commonplace: How much will it cost you to employ AI?
OpenAI is hunting for cheaper ways to run ChatGPT
The focus is mostly on training a frontier AI model, which requires massive processor clusters. However, training takes place on a regular basis.
Every time a user requests a task from the model, inference, which is the computation process that transforms a user’s request into an AI answer, takes place.
Billions of such interactions may make even small variations in computation cost crucial for a service like ChatGPT. This helps explain OpenAI’s hardware diversification.
Although it is still a major Nvidia client, it has also collaborated with Broadcom to create its own AI processor (AVGO).
By the end of 2026, the Jalapeño chip, specifically designed for inference, should be in use. Reuters reported in June that OpenAI designed the chip partly to improve the economics and availability of the infrastructure behind its AI products.
Google offers OpenAI an additional choice. Its latest TPU 8i is designed specifically for inference and delivers up to 80% better performance per dollar than the previous Ironwood generation, according to Google Cloud.
This does not imply that Google’s processor is 80% less expensive than Nvidia’s. The model, workload, software, contract, and infrastructure around the chip all affect the cost of AI technology.
However, since each new user increases the need for processing power, an AI business seeks to enhance performance per dollar.
Google turned cost control into a potential business
There’s more to the agreement.
Because it didn’t want to completely depend on outside chip providers for its massive processing demands, Google first developed TPUs.
Google may now market its internal cost-control effort to other AI companies.
I recently wrote for TheStreet that Evercore ISI believes Google’s external TPU business could generate $75 billion in 2027 and $113 billion in 2028.
Due to its direct competition with Google’s Gemini products, OpenAI would be a very significant client.
When a competitor leverages processing capacity to service its AI clients, Google would essentially be profiting.
That’s already taking place elsewhere. Oracle is preparing to provide Google TPUs via its own cloud infrastructure, and Anthropic has committed to use a significant portion of Google’s TPU capacity.
Google’s attempt to make bespoke silicon a company outside of its own ecosystem would be further validated if OpenAI were to purchase the chips directly.
The timing is noteworthy. Alphabet CEO Sundar Pichai said on the company’s Oct. 28 earnings call that Google Cloud revenue increased 103% from a year earlier and that the company is seeing strong demand for TPU systems across a broad range of customers.
This implies that Google might make more money regardless of the AI helper a customer selects, which is positive news for Alphabet investors.
ChatGPT may be used in place of Gemini, but the infrastructure supporting it may still generate revenue for Google.
OpenAI may have found a cheaper way to run ChatGPT.Heather Diehl / Getty Images
AI prices have already been moving lower
There is evidence that consumers may benefit from declining computer costs, but not always via more affordable monthly subscriptions.
Currently, ChatGPT Plus costs $20 a month. Additionally, OpenAI charges $8 per month for ChatGPT Go in the United States.
Through its application programming interface, OpenAI has consistently reduced the cost for developers to use some of its models. CEO Sam Altman said during the firm’s September developer conference that the business aimed to continue enhancing the quantity of intelligence clients receive at every pricing tier.
TheStreet’s Mwangi Enos recently examined how dramatically OpenAI’s model prices have fallen, including large reductions in the cost of its newest models compared with previous generations.
This change is significant because OpenAI may pass on reduced infrastructure costs to clients in a number of ways.
It could reduce prices. It could keep monthly prices unchanged while increasing message limits. It could move more capable models into cheaper subscription tiers. Or it could simply protect margins while spending more on new products.
Customers shouldn’t believe that Google’s chips would always result in a less expensive ChatGPT subscription.
However, OpenAI has greater negotiating leverage because of competition among chip providers. And having multiple hardware alternatives lessens its reliance on any one provider.
Cheaper chips may matter more as AI use explodes
The rapid expansion of AI use is driving an increasing demand to reduce computer costs.
Google said its first-party AI application programming interfaces are now processing nearly 32 billion tokens per minute, more than four times the volume from a year ago.
OpenAI has the same fundamental issue. As more individuals use AI, they require more infrastructure to meet their needs.
This produces a unique dynamic.
A profitable software business often gains from the low cost of acquiring new clients. But every time a consumer uses an AI company’s product, the company still incurs significant computational costs.
Chip efficiency is clearly a factor in consumer price. Thus, the rumored Google acquisition of OpenAI is significant to investors outside of Alphabet, Nvidia, and Broadcom.
Consumers ultimately pay for the massive data centers, processors, and energy that power the AI products they use. So businesses that can lower such expenses have greater control over what clients are willing to spend.
Google is attempting to join that group. In addition, the most significant outcome for ChatGPT customers may not be the logo on the processor if OpenAI really begins purchasing its competitors’ chips.
This may happen if increased competition lowers the cost of producing each AI response.
Related: Google gives Fitbit a major health upgrade with one catch