NVIDIA Bets on Ilya Sutskever's Secretive SSI With Vera Rubin Access

Safe Superintelligence Inc. has never shipped a product, published a paper, or revealed what it is building. That opacity is deliberate, and NVIDIA just bet on it anyway. The two companies announced a long-term strategic partnership pairing an undisclosed NVIDIA equity investment with priority access to the Vera Rubin compute platform, promising to multiply SSI's compute capacity by an order of magnitude over the next 12 months.

The most secretive lab in AI just got the most powerful hardware in AI

NVIDIA gained rare access to SSI's closely guarded research before committing, evaluating Ilya Sutskever's track record in foundational AI and SSI's new research direction before agreeing to both invest and collaborate long-term. The partnership goes beyond a hardware sale: SSI's research feeds back into how NVIDIA designs future chips, and the two companies will work together on the technical advancement of NVIDIA's current and upcoming compute platforms.

Who is SSI, and why does everyone keep throwing money at it?

Safe Superintelligence Inc. is an AI safety research company with a single stated objective: build a safe superintelligence and nothing else. Founded in 2024 by Ilya Sutskever and Daniel Levy, the company draws much of its credibility from Sutskever's prior work on AlexNet, AlphaGo, and the GPT model family.

Sutskever has described AI history in three phases: a research era from 2012 to 2020, a scaling era from 2020 to 2025, and a new research era beginning in 2026 in which algorithmic innovation, rather than raw compute, drives progress. He has told associates he is pursuing techniques distinct from those used at OpenAI, describing it as "a different mountain to climb," though he has not disclosed specifics and SSI has published no papers.

Unlike OpenAI or Anthropic, which are deploying chatbots and productivity tools, SSI is deliberately withholding commercialization. Sutskever has said the company's first product will be safe superintelligence and "it will not do anything else up until then" — no API, no chatbot, no enterprise deals, no interim releases.

The numbers, and the ones that are missing

The deal terms are thin. The announcement omits how much NVIDIA invested, how many megawatts the expansion covers, and when the systems arrive. What is known is the broader financial picture around SSI:

• SSI's first funding round closed in September 2024, raising $1 billion at a $5 billion valuation.
• A subsequent round led by Greenoaks brought in $500 million, with Andreessen Horowitz and Lightspeed Venture Partners also participating.
• According to Reuters, the company is now valued at $32 billion.
• Total capital raised stands at approximately $3 billion.

An order of magnitude is a ratio, and SSI has never published the base. The lab has not disclosed how many accelerators it runs, which sites house them, or who supplies the power, so a tenfold increase resolves to a number no reader can size. For comparison, NVIDIA's deal with Thinking Machines Lab named at least one gigawatt of Vera Rubin systems targeting early 2027 deployment. SSI's announcement carries no equivalent anchor.

What SSI is actually getting: Vera Rubin

NVIDIA's flagship configuration is the Vera Rubin NVL72: 72 Rubin GPUs and 36 Vera CPUs connected through NVLink 6, delivering 3.6 EFLOPS of NVFP4 inference and 2.5 EFLOPS of training. EFLOPS means exaFLOPS, one quintillion floating-point operations per second, making a single NVL72 rack one of the densest compute units ever built.

The figure that shapes the platform's design is 20.7 terabytes of HBM4 memory pooled across a single 72-GPU rack. Modern reasoning models spend most of their inference cost moving data between GPUs rather than computing on it, and Rubin's answer is to make memory and interconnect bandwidth the headline spec rather than raw FLOPS.

Vera Rubin is in full production, with Rubin-based products shipping in the second half of 2026. SSI's expansion therefore depends on a hardware ramp still in its early stages.

NVIDIA's real play: a portfolio of futures

This deal follows a visible pattern. NVIDIA is securing guaranteed demand for its most expensive hardware while acquiring a window into research it otherwise could not access. The financial mechanics show up in NVIDIA's own books: in its first-quarter results for fiscal 2027, the company reported $43.4 billion in non-marketable securities, up from $22.3 billion three months earlier. That portfolio nearly doubled in a single quarter, while data-center revenue reached $75.2 billion. Some of the demand NVIDIA reports now comes from customers it has also funded.

SSI is also hedging on the chip side. Earlier this month, Alphabet announced that Google Cloud will sell SSI access to tensor processing units (TPUs), its in-house AI chips, to support SSI's research toward safe superintelligence. SSI is running Google's TPUs for some workloads and NVIDIA's Vera Rubin for the scale-up phase.

Who wins, and what it unlocks

The beneficiaries are clear:

• SSI gets frontier-scale compute without generating revenue to pay for it, preserving its no-product-until-superintelligence stance.
• NVIDIA gets equity upside in a $32 billion lab, a guaranteed Vera Rubin customer, and a technical collaboration channel into research nobody else has seen.
• SSI's investors (a16z, Sequoia, DST Global, Greenoaks) see the lab's compute credibility jump, which protects their $32 billion valuation mark.

The subtler cost falls on NVIDIA's competitors. AMD's Helios rack with MI400-series GPUs also targets the second half of 2026, claiming 2.9 EFLOPS FP4 and 31 TB HBM4, but AMD has no equivalent relationship with SSI. If SSI's research direction turns out to be the next paradigm shift the way transformers were in 2017, NVIDIA will have had a front-row seat for years before anyone else.

As of mid-2026, SSI remains in a research-only posture with no announced timetable for any product or paper, describing its program as a "straight shot" toward superintelligence. Outside observers have suggested the company will eventually need to disclose at least some of its work to satisfy investors and participate in the broader research community, but no public milestones have been set.

Three things to watch as this partnership develops:

1. The size of the stake. NVIDIA itemizes material equity positions in its quarterly filings. The next earnings report will show whether the SSI investment clears that disclosure threshold.
2. A physical site and a power figure. Compute at the implied scale needs a data center home, and neither company has named one.
3. What SSI actually publishes. The official announcement describes a "new research direction" compelling enough for NVIDIA to invest. At some point, the world will need to see it.

NVIDIA has purchased a look at two years of the most closely guarded AI research on the planet. Whether that research represents the next leap in machine intelligence or a very expensive dead end is the only question that ultimately matters here.