Investment Memo: SSI (Safe Superintelligence Inc.)
Prepared for: IC Review — Hypothetical Series A participation Date: 2025
---
Thesis
SSI is the purest expression of a "person bet" in venture history. Ilya Sutskever co-founded OpenAI, was chief scientist behind GPT-1 through GPT-4, and is credibly one of the five most important researchers in deep learning's modern era (AlexNet, seq2seq, scaling advocacy). The company's stated plan — build safe superintelligence directly, ship no intermediate products, take no commercial detours — means the investment is a binary option on the single most valuable technological outcome imaginable.
The fund-returner logic: if any lab reaches superintelligence (or even credibly approaches it), equity in that lab is worth more than any company that has ever existed. SSI's structure — no revenue pressure, no enterprise sales, no safety-vs-shipping tension — may actually be an advantage in a race where OpenAI, Anthropic, and Google are diverting research talent to product, inference costs, and revenue. If the frontier is won by focused research rather than commercial flywheels, SSI is positioned better than anyone. That is a real "if," and the entire memo hinges on it.
---
Product & Wedge
There is no product, by design. The "wedge" is:
- Talent gravity. Sutskever plus co-founders Daniel Gross (ex-Apple AI, though Gross departed to Meta in mid-2025 — a material negative signal) and Daniel Levy (ex-OpenAI optimization lead) attract researchers who want to work on frontier problems without product distraction.
- Strategic focus. SSI claims its research agenda diverges from the industry's scaling consensus — Sutskever has publicly suggested pretraining scaling is plateauing and that new paradigms are needed. If he's right and everyone else is over-invested in the current recipe, SSI's clean-sheet approach is the wedge.
- One customer, eventually: superintelligence itself, monetized however the winner chooses. There is no articulated go-to-market, and the company is explicit that there won't be one until the mission is achieved.
---
Market & Competition
The market is "all cognitive labor," which is unfalsifiable and therefore analytically useless; the honest frame is the frontier-lab race.
- OpenAI — massive compute (Microsoft/Stargate), revenue north of $10B annualized, largest research org. Product distraction is real but so is their data/capital flywheel.
- Anthropic — the closest philosophical competitor (safety-first framing), but chose the commercial path; Claude revenue funds research.
- Google DeepMind — deepest compute pool and TPU independence; Gemini closing frontier gaps.
- Meta Superintelligence Labs — aggressively poached talent (including Gross and, reportedly, an unsuccessful run at Sutskever and an acquisition attempt on SSI itself).
- xAI — Musk's capital and Colossus cluster.
SSI competes on none of these axes except research quality. It has, by construction, less compute, less data exhaust, and no revenue to recycle. The bet is that ideas, not flywheels, decide the outcome.
---
Traction & Business Signal
Publicly known:
- Raised ~$1B (Sept 2024) at ~$5B valuation (a16z, Sequoia, DST, NFDG, SV Angel); subsequently raised at a reported ~$32B valuation (April 2025, Greenoaks-led, ~$2B).
- Offices in Palo Alto and Tel Aviv; headcount reportedly small (tens of researchers).
- Reported Google Cloud TPU arrangement for compute.
- Meta acquisition interest rebuffed; Sutskever became CEO after Gross's exit.
Unknown: research progress, benchmarks, architecture direction, compute committed, burn rate, revenue (zero by design), timelines. There is literally no external evidence of technical progress. Every traction signal is a financing signal, which is circular — investors validating investors.
---
Risks (the three that kill the deal)
- The scaling-capital gap. If superintelligence requires $50–100B+ in compute — the trajectory OpenAI, Meta, and Google are on — SSI cannot get there without either revenue or dilution so severe the Series A is crushed. "No product" means every dollar of compute is financed by equity. The structure that makes SSI pure also makes it structurally undercapitalized against rivals spending revenue.
- Key-person concentration at unprecedented severity. Gross already left. If Sutskever departs, is incapacitated, or is simply wrong about the post-scaling paradigm, there is no product, no revenue, no brand, and no fallback. The asset is one man's research intuition, and his headline achievements came inside OpenAI's resource envelope.
- No feedback loop. Rivals learn from hundreds of millions of users, RLHF data, and deployment failures. SSI forgoes all of it. If frontier progress increasingly depends on deployment-derived data and iteration (as the last three years suggest), SSI is running blind by design — and we'd have no milestones to mark the position against until it's too late.
---
Verdict: Invest — small, at Series A pricing only.
At the actual $5B Series A, a 1–2% fund position is a rational lottery ticket on the highest-EV outcome in technology, with acceptable loss-given-failure. At the current $32B mark, pass — you're paying for the option after the cheap volatility is gone.
The one question I need answered first: What is Sutskever's concrete technical thesis for why superintelligence is reachable without deployment-scale data and $50B-class compute — and what internal milestone in the next 18 months would falsify it? Without a falsifiable answer, this isn't an investment; it's a donation to a worldview.