OpenAI (Private Code: OPAI.PVT) President Greg Brockman believes the entire AI industry will face a sustained shortage of computing power for the foreseeable future.
Speaking at a roundtable forum in New York on Thursday, Brockman stated: "My assessment is very clear. No matter how much the industry increases its investment, a tight supply of computing power will persist."
He explained: "At this stage, we have to make difficult choices, deciding which models to truly train and which products can be scaled for deployment."
OpenAI, Anthropic (Private Code: ANTH.PVT), and other AI labs, alongside major tech companies like Microsoft, Amazon, and Meta Platforms, Inc., are simultaneously training new models and offering AI services to users. The available computing power is insufficient to meet the surging demand.
On January 5, 2026, on the eve of the International Consumer Electronics Show in Las Vegas, Nevada, OpenAI President Greg Brockman attended an AMD press conference.
AI training and operation heavily rely on mass-scale computing hardware like NVIDIA graphics cards. The industry has cumulatively invested hundreds of billions of dollars in building data centers, yet the growth rate of computing power still cannot keep pace with the expansion of demand.
During Alphabet's second-quarter earnings call on Wednesday, Chief Financial Officer Anat Ashkenazi stated that even with Alphabet's heavy investment in building its own data centers, it would still need to purchase external computing power from third parties in the third quarter. Alphabet finalized a deal in June to lease AI computing power from SpaceX (SPCX) at a monthly cost of $920 million.
Brockman noted that he often sees OpenAI's computing capacity expanding rapidly, leading him to believe there might be a surplus. However, as OpenAI continuously develops new models and expands AI use cases, the demand for computing power surges again.
"The era of a computing-power-driven economy has arrived, and supply-demand tension will become the norm," Brockman interpreted.
The executive also stated that as long as AI is implemented properly, various clients are gradually seeing a return on their investment. He does not endorse the current industry trend of "tokenmaxxing" — a phenomenon where many companies and employees recklessly consume AI tokens on computing power just to prove their AI capability, without pursuing actual value.
Brockman said: "If you use the right scenarios, companies can get real returns. If you are just mindlessly increasing token usage, you will only end up wasting computing power. But if you focus on business value and use AI to solve real business problems, you can realize the benefits."
Comments