A Million Times Faster: Zenithon AI's Quest to Simulate Extreme Physics

Key points

Key takeaways from a Scaling Europe interview with Zenithon AI founder Alex Higginbottom (September 2026):

What Zenithon builds. A frontier AI lab creating world models for physics - machine-learning models that reproduce complex simulations in milliseconds rather than the 10 hours to multiple days a single high-performance-computing run takes today, roughly a million times faster.

The $10m seed. The round, announced in September 2026, brings in Backed, Lunar, Seraphim Space, MMC and SOSV; Higginbottom argues the mission - keeping pace with the exponentially more complex engineering of the 21st century - is both generational and economically significant.

First model next week. Zenithon's inaugural world model launches within days and is, per Higginbottom, the first to bring different fusion confinement types (tokamaks, laser fusion and every other approach) into one model; the company plans new iterations every one to two months with broader capability.

Trillions of designs. A fusion reactor or semiconductor wafer has trillions of possible configurations, so today's one-simulation-at-a-time workflow is like picking random grains of sand off a beach; the accelerated model lets engineers actually search that design space.

Founder's road. Higginbottom spent three years helping commercialise fusion with around 20 fusion companies before starting a PhD in AI for plasma physics in Cambridge; he raised a pre-seed, dropped out and co-founded Zenithon in 2025 with CTO Abetharan 'Abby' Antony, building a research team drawn from Meta, Nvidia and Anthropic.

Sim-to-real is next. After speed, the second value wave is training directly on experimental data to close the sim-to-real gap; the long-run North Star is running a million designs in a world model and then building only the final one.

Chipmaking opportunity. Plasma physics underpins wafer etching in semiconductor manufacturing - Lam Research's CTO has cited around 100 trillion ways to make a wafer - so Zenithon sees its world models applied across fusion, chipmaking and aerospace.

Where the money goes. Most of the $10m goes into compute and data to train multi-billion-parameter models, with the team expanding in London and San Francisco near customers; early work with leading but as-yet-unnamed fusion companies is already in daily workflows.

Read more: YouTube - Scaling Europe