What is a GPU without AI company? Not much more than an argument of ambitious investors.
The CEO of Openai, Sam Altman, admitted twice during the last month that he had underestimated the computing power needing two major launches: GPT-4.5 And chatpts image generator.
When GPT-4.5 was released at the end of February, Openai praised him as his best model to date (as AI companies tend to do so), with better emotional intelligence and “vibrations”. But the company has restricted availability at its $ 200 / month level, citing the shortages of GPU.
“It’s a giant and expensive model”, “Altman said at the time. “We really wanted to launch it on more and pro at the same time, but we grew up a lot and are out of the GPU. This is not how we want to work, but it is difficult to perfectly predict growth overvoltages that lead to GPU shortages.”
This week, the same fun has occurred with the new image generator. The company shouted the exit from the roofs and created media threw on social networks, largely via images inspired by the Ghibli studio Raised alarms on the violation of copyright. After a high demand, Altman announced a limit on the number of images that users could create, citing overworked GPUs again.
“It’s super fun to see that people like images in Chatgpt, but our GPUs melt”, he said. “We will temporarily introduce certain rate limits while we work to make it more effective. Hopefully will not be long!”
This tweet is currently not available. It can be loaded or has been deleted.
These GPU deficits could slow down the adoption of the mass on the AI market. In September 2024 blogAltman wrote: “If we do not build enough infrastructure, AI will be a very limited resource that wars take place and it mainly becomes a tool for the rich.”
Recommended by our publishers
A few months later, President Trump explained Altman as one of the faces of an IA infrastructure investment of $ 500 billion, nicknamed StargateWho “will build the physical and virtual infrastructure to feed the next generation of advances in AI,” said Trump in January.
Meanwhile, China In depth The model challenges the premise that Openai needs more money and infrastructure. Although its claims cannot be verified, it can be able to operate at the same level as the Openai models with a fraction of the computing power. (OPENAI complaints A little ironically that Deepseek used OpenAi models for training.)
OPENAI is mainly based on nvidia tokens, including its next Blackwell chip This promises enormous performance gains and 25% less energy consumption. He plans to extend to AMD offers in 2026 and also develops his own chip, the rod reports.
Get our best stories!
What’s new now
Register What’s new nowYour daily dose of the latest technological news, the best new products and expert advice from PCMAG publishers.
By clicking on register, you confirm that you are 16 years old and over and accept our Conditions of use And Privacy Policy.
Thank you for registering!
Your subscription has been confirmed. Keep an eye on your reception box!
About Emily Forini
Main journalist
