OpenAI Claims Navier-Stokes Solution Amid Controversy, ProLogium Begins Battery Production

On September 8, 2026, OpenAI claimed to have solved one of mathematics’ most stubborn open problems, but the announcement immediately drew allegations that the company may have had access to a rival team’s unpublished work. The claim, if verified, would mark the first resolution of a Navier-Stokes Millennium Prize Problem in roughly 90 years.

The Navier-Stokes Claim and Its Controversy

OpenAI says an unreleased internal model, described by the company as “significantly more capable than GPT-6 Astra,” ran a coordinated system of roughly 10,000 concurrent AI agents for 88 hours to reach a solution, specifically, that the equations governing fluid motion can develop a singularity in finite time.

The controversy centres on Tristan Buckmaster, a professor at New York University who claims OpenAI accelerated its effort after learning of parallel progress he and a researcher at Anthropic had made on the same problem. Buckmaster also raised concerns that their work-in-progress, stored in OpenAI’s Codex model, could have been visible to OpenAI’s team. OpenAI denied using the pair’s work, stating its solution differed “significantly.” Despite the controversy, OpenAI says it does not intend to claim the $1 million prize.

OpenAI’s Mathematical Track Record in 2026

The Navier-Stokes announcement is the third major mathematical claim OpenAI has made this year. In May 2026, a general-purpose reasoning model disproved the 80-year-old Erdős unit distance conjecture in discrete geometry by identifying a structural connection between distinct mathematical domains, a link that had eluded human mathematicians for nearly eight decades. External mathematicians verified the proof.

Solid-State Batteries Hit the Production Line

On September 2, 2026, ProLogium Technology announced mass production of its Gen 3.5 Lithium Ceramic Battery at its Giga-level facility in Taiwan. Third-party TÜV testing confirmed that its 185.4 Ah large-format cells achieved 381 Wh/kg gravimetric energy density and 903 Wh/L volumetric energy density, compared with 160-200 Wh/kg for LFP and 240-300 Wh/kg for high-nickel NMC chemistries in current EVs.

The cells have also been classified as all-solid-state under China’s GB/T 43568-2026 standard, which took effect in July 2026. UL Solutions tested them in a six-hour vacuum at 120°C; mass loss came in below 0.05%, well under the 0.5% threshold for that classification. ProLogium, which has a technology cooperation agreement with Mercedes-Benz, is running its Taoyuan line at 0.5 GWh per year, enough for roughly 6,000 80 kWh packs annually. The scale-up story across next-generation battery and AI infrastructure shares a common thread: moving from lab performance to manufacturing yield is where most programmes stall, and ProLogium’s production numbers will be the metric to watch. For context on how infrastructure investment is shaping AI and adjacent hardware markets, see our coverage of next-generation AI infrastructure build-out.

Silicon Anodes and the Lab Pipeline

Also on September 2, Amprius Technologies announced its 500 Wh/kg SiCore® silicon-anode lithium-ion cell, targeting long-endurance aviation applications including High-Altitude Platform Stations and fixed-wing drones. The cell runs on conventional manufacturing equipment, with commercial availability set for Q4 2026, a meaningful contrast to early solid-state programmes that required bespoke tooling and constrained output.

Further up the research pipeline, a joint team from Tianmushan Lab and Tsinghua University has developed high-stability lithium metal pouch cells reaching 600 Wh/kg. Separately, researchers in South Korea achieved 1,270 Wh/L volumetric energy density in an anode-free lithium metal pouch cell, against a typical 650 Wh/L for current EV lithium-ion packs. Neither result is near commercial production, but both set a ceiling for what silicon and lithium-metal chemistries may eventually deliver at scale.