Over the same period, OpenAI’s compute grew from 0.2 gigawatt (GW) in 2023, to 0.6 GW in 2024 and about 1.9 GW in 2025, Friar said in a Sunday (Jan. 18) blog post.
“And we firmly believe that more compute in these periods would have led to faster customer adoption and monetization,” Friar said in the post.
Saying that compute is “the scarcest resource in AI,” Friar said that OpenAI has shifted from having one compute provider to “a diversified ecosystem” and that the company manages this portfolio in a way that makes AI viable for users’ everyday workflows.
“As these systems move from novelty to habit, usage becomes deeper and more persistent,” Friar said. “That predictability strengthens the economics of the platform and supports long-term investment.”
Noting that OpenAI added a free ad-supported tier to its business model that also includes consumer and team subscriptions and usage-based application programming interfaces (APIs) tied to workloads, Friar said this model “closes the loop.”
“Where this goes next will extend beyond what we already sell,” Friar said. “As intelligence moves into scientific research, drug discovery, energy systems and financial modeling, new economic models will emerge. Licensing, IP-based agreements and outcome-based pricing will share in the value created.”
OpenAI announced Friday (Jan. 16) that it was bringing its $8-a-month ChatGPT Go subscription tier to the United States and everywhere ChatGPT is available after launching the tier in 171 countries since August. The tier joined the existing Pro and Plus plans that cost $20 and $200 per month, respectively.
The company also said it plans to begin testing ads in the U.S. for its Free and Go tiers within weeks. It said it will not include ads in the Plus, Pro, Business and Enterprise subscription plans.
On Wednesday (Jan. 14), OpenAI said it will integrate 750 megawatts of ultra-low latency compute from chipmaker Cerebras in several stages, beginning this year and continuing through 2028. OpenAI said the compute will accelerate the response time of its AI models.
