Safe Superintelligence Inc. (SSI)

Contribution to AI

A laboratory that ships nothing is a wager about how frontier research can be paid for, and the wager was placed here at a size nobody had attempted. Every other well-capitalised group working towards general AI after 2020 funded itself by selling access to intermediate systems — an API, a chat product, enterprise contracts — which meant releasing capabilities on a commercial clock and defending them afterwards. The alternative proposed in 2024 was a single long run: no interim releases, no revenue, no published papers, on the argument that the pressure to ship is itself a safety problem rather than a distraction from one. Capital agreed. A billion dollars arrived in September 2024 at a five-billion valuation, and roughly two billion more by April 2025 at a reported thirty-two, for an organisation of around twenty people with nothing external to point at. That repriced what a thesis and a track record are worth in the total absence of a product, and gave the safety-first position an independent institution rather than a team embedded in a company selling something else. Two decisions entered the record alongside it: training on Google's TPUs instead of the industry-standard Nvidia hardware, and rebuffing an acquisition approach from Meta in 2025 at the cost of a co-founder. Whether the straight-shot strategy produces anything remains open, which is exactly why the experiment matters.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).