AI supercomputers double their performance every 9 months

The development of AI supercomputers is picking up speed! According to a new study by Epoch AI, leading systems are doubling their performance every nine months—that's 2.5x growth every year. This acceleration is driven by more and better chips, with the number of chips increasing 1.6x annually and performance per chip growing at the same pace.

The dimensions are impressive: while systems with more than 10,000 chips were rare in 2019, xAI's Colossus now has over 200,000 AI chips. These giants now cost billions (Colossus: approx. $7 billion) and consume as much electricity as medium-sized cities – around 300 megawatts in the case of Colossus, comparable to the electricity consumption of 250,000 households.

The shift in the balance of power is also interesting: while only 40% of global AI computing power was in private hands in 2019, this share rose to 80% by 2025. Geographically, the US dominates with 75% of global AI computing power, followed by China with 15%.

Where will this development lead? If the trends continue, the leading AI supercomputer in 2030 would require around 2 million AI chips, cost $200 billion, and consume 9 gigawatts of electricity – equivalent to nine nuclear reactors!

Why it matters: The explosive growth of AI supercomputers points to a fundamental realignment of technological power, with private companies increasingly taking control of the most advanced AI infrastructure. The extreme concentration of these resources could have a decisive influence on who will be able to develop and control the most powerful AI systems in the future.

Most slide decks look the same—static, boring, and forgettable.

Leave them behind with Prezi AI, the AI presentation software that builds unique, dynamic presentations just from your prompt. It’s fast, intuitive, and designed to fit your ideas perfectly.

Stand out with Prezi’s unique format that uses zooming, cinematic movement to wow your audience.

Try for free today

Elon Musk's platform X is suing Minnesota over a law that bans AI-generated deepfakes in election campaigns. X invokes freedom of speech and refers to §230, which protects platforms from liability. The case could redefine the role of platforms in shaping political opinion.

The Trump administration is considering measures to deny DeepSeek access to US technology and block the service for American users. This is due to concerns about possible circumvention of export controls through the use of Nvidia chips and the potential use of AI for Chinese military purposes.

Despite international concerns, BMW plans to integrate DeepSeek AI into new vehicle models in China. This underlines the technological dependence of Western companies on the Chinese market and raises questions about data security and ethical responsibility.

AI Digest reports that OpenAI’s new o3 and o4-mini agents show a dramatic leap in coding performance, aligning with a 2024–2025 trend of rapid acceleration. These results suggest coding agents' time horizons are now doubling approximately every four months, marking a sharp break from previous progress.