The Compute Reality Check for AI Safety
Ilya Sutskever left OpenAI in May 2024 with a very specific, almost single-minded mandate. He wanted to solve superintelligence safety far away from commercial product roadmaps, ad models, and enterprise sales targets. So he co-founded Safe Superintelligence, known simply as SSI, alongside Daniel Gross and Daniel Levy.
Now, after operating mostly in stealth and securing $1 billion in capital back in September 2024, SSI is making its biggest structural move yet. They've inked a long-term partnership with Nvidia to supply the massive compute capacity needed for their next research phase.
It's a huge deal. And it tells us everything about where AI research is heading.
Why Silicon Matters More Than Philosophy
Here's what most coverage misses about Sutskever's strategy. Outside observers assumed SSI would operate like an academic think tank. People thought they'd sit back, write theoretical papers on alignment, and leave the massive GPU clusters to commercial giants like OpenAI and Anthropic.
That was always a naive fantasy.
The reality is that safety algorithms don't run on good intentions or neat mathematical proofs. They run on silicon. If you want to study how superintelligent systems behave, you have to build systems that actually approach superintelligence. And to build those, you need tens of thousands of top-tier GPUs wired together with ultra-low latency.
When looking at the sheer scale required for modern frontier research, whether evaluating cloud infrastructure like AWS vs