Sep 9, 2026 - An AI Just Proved Fermat's Last Theorem in Eleven Days; eRacks Says the Server That Runs Work Like It Now Fits on a Desk

Sep 9, 2026 - An AI Just Proved Fermat's Last Theorem in Eleven Days; eRacks Says the Server That Runs Work Like It Now Fits on a Desk

Campbell, CA - Sep 9, 2026

FOR IMMEDIATE RELEASE

An AI Just Proved Fermat's Last Theorem in Eleven Days; eRacks Says the Server That Runs Work Like It Now Fits on a Desk

Anthropic's machine-checked proof was eleven days of sustained, verifiable AI work. The open-source server maker explains why that changes how businesses should size private AI hardware, and publishes the numbers.

On September 4, Anthropic published the first complete, computer-checked proof of Fermat's Last Theorem. Claude wrote it in Lean, a programming language built so a computer can verify every step of a mathematical argument instead of a human reviewer: eleven days, largely unattended, 13 million lines, about 29,500 intermediate theorems, and nothing assumed beyond Lean's three standard axioms. Andrew Wiles's 1995 proof ran 129 pages and took months of expert review to check.

eRacks Open Source Systems, which has built open-source servers since 1999, reads the result as a hardware story.

"That was not a chat reply. It was eleven days of structured work that a machine could check, and that is the shape of job we spec servers for now," said Joseph Wolff, founder and CTO of eRacks. "The frontier models live in their makers' datacenters. What a business can own is the open-weight class right behind them, running on hardware it controls, on its own documents, with no per-token bill and nothing leaving the building."

Open-weight models are those whose trained weights are published for anyone to run, such as Qwen, Llama, DeepSeek and Gemma. On the eRacks bench, a dual Intel Arc Pro B70 server runs Qwen3-14B at 54 tokens per second and Qwen3.6-27B at 26, with the models resident in 64GB of GPU memory; a token is about three quarters of a word. The measurements, the software stack and the fixes it took to get there are published in the company's August benchmark report.

The company also published the arithmetic. Using the defaults of its free rent-versus-own calculator at https://eracks.com/tco/ (24 hours a day, 15 cents per kilowatt-hour, $1,200 a year of overhead, a three-year payback), an eRacks/AINSLEY at $21,395 pencils out against an AI bill of about $800 a month. A team of ten on $100 seats is $1,000 a month, indefinitely. Every field in the calculator is editable.

eRacks builds each server to order, burns it in, and benchmarks it before shipment, with the numbers in the box. The AINSLEY configurator and the full analysis are at https://eracks.com/products/ai-rackmount-servers/AINSLEY/ and https://blog.eracks.com/2026/09/ai-proved-fermats-last-theorem-what-it-means-for-your-next-server/

About eRacks Open Source Systems

eRacks Open Source Systems, founded in 1999 and headquartered in Campbell, California, specializes in open-source servers and storage. The company builds custom-configured rackmount servers, NAS (network-attached storage) systems, HPC (high-performance computing) clusters, and AI inference servers, all shipped with Linux or other open-source operating systems. More at https://eracks.com

Media Contact

Joseph Wolff eRacks Open Source Systems joe@eracks.com https://eracks.com