NVIDIA Backs Ilya Sutskever’s SSI as It Prepares to Scale Compute 10x

Introduction

Ilya Sutskever’s secretive AI lab, Safe Superintelligence Inc. (SSI), has entered a long-term strategic partnership with NVIDIA.

The two companies announced on July 27, 2026 that NVIDIA had made a substantial investment in SSI and would give the lab access to its next-generation Vera Rubin computing platform. Together, the investment and hardware access are expected to increase SSI’s available compute by roughly an order of magnitude.

SSI’s own announcement was even more direct: the company said its research had reached a point where it was worth scaling.

SSI said the NVIDIA partnership would help increase its compute capacity by 10x over the next 12 months.

Sutskever summarized the moment in a short post:

"Time to scale that SSI."

The companies did not disclose the financial terms in their official announcement. Reuters, citing a person briefed on the deal, reported that NVIDIA's equity investment is $5 billion. Bloomberg had previously reported the same figure.

Accuracy note: The $5 billion amount is a reported figure, not a number disclosed by NVIDIA or SSI in the official partnership announcement.

SSI and NVIDIA Form a Long-Term Alliance

Safe Superintelligence was launched in June 2024 by Ilya Sutskever, Daniel Gross, and Daniel Levy with an unusually narrow objective: build safe superintelligence without being distracted by ordinary product cycles.

The company's website still reflects that focus.

SSI describes itself as a "straight-shot" lab focused on one product: safe superintelligence.

Unlike most frontier AI companies, SSI has not spent the past two years releasing chatbots, developer APIs, model cards, public demos, or research papers.

Its stated strategy is to work on capabilities and safety together as technical problems, while keeping the business insulated from short-term commercial pressure.

That makes the NVIDIA partnership especially notable. The chipmaker says it decided to deepen the relationship after receiving rare access to SSI's closely guarded

research.

According to NVIDIA’s official announcement, SSI has spent the past two years pursuing a new research direction intended to produce powerful AI that remains robustly aligned.

NVIDIA CEO Jensen Huang said Sutskever has been responsible for foundational breakthroughs in modern AI, beginning with AlexNet, and said NVIDIA is eager to see what SSI can discover using the Vera Rubin platform.

Sutskever’s statement focused on the same point:

“We have research that is worthy of scaling up, and having access to a big NVIDIA computer will let us do so.”

Vera Rubin Gives SSI Much More Compute

The partnership gives SSI access to NVIDIA’s new Vera Rubin platform.

NVIDIA says Vera Rubin is designed for large-scale training and inference, particularly for multitrillion-parameter models, long-context workloads, mixture-of-experts systems, and agentic AI.

For SSI, the important number is not a benchmark score but capacity.

The company says the combination of NVIDIA’s investment and access to Vera Rubin will allow it to expand its compute by about 10x.

SSI’s social-media announcement says that increase is expected over the next 12 months.

NVIDIA and SSI will also collaborate on the technical evolution of current and future NVIDIA computing platforms, with SSI contributing feedback based on its own AI research.

A Reported $5 Billion Equity Investment

The official press release describes NVIDIA’s investment only as “substantial.”

Reuters subsequently reported that the equity investment is worth $5 billion, citing a person briefed on the transaction.

If accurate, it would represent an unusually large strategic investment in a company that has not yet released a commercial product.

The partnership reportedly came together within a matter of weeks.

That speed is part of what makes the story interesting: NVIDIA appears to have gained enough confidence from private access to SSI’s research to commit both capital and scarce next-generation computing infrastructure.

Why Investors Have Backed SSI Before Seeing a Product

SSI has attracted large amounts of capital despite revealing very little publicly.

In September 2024, roughly three months after its launch, the company raised more than $1 billion** at a reported valuation of about **$5 billion.

Investors included Andreessen Horowitz, Sequoia Capital, DST Global, SV Angel, and NFDG.

At the time, SSI said the money would be used primarily to acquire computing capacity and recruit researchers and engineers.

In April 2025, the company reportedly raised another $2 billion** at a valuation of **$32 billion. Greenoaks led the round, with participation from investors including Andreessen Horowitz and Lightspeed. Alphabet and NVIDIA also backed the company.

That means SSI reached a $32 billion private valuation without releasing a model, product, or conventional public technical roadmap.

Part of the explanation is Sutskever’s track record.

He was one of the authors of AlexNet, the 2012 neural network that helped establish deep learning as the dominant approach in computer vision. He later worked at Google Brain and contributed to research including

sequence-to-sequence learning and AlphaGo.

In 2015, he co-founded OpenAI and became its chief scientist. He was involved across multiple generations of the company’s large-scale model research before leaving in May 2024.

A month later, he launched SSI.

The new company was deliberately designed to avoid the commercial tension that comes with shipping consumer products while simultaneously pursuing advanced AI research.

Its website says:

“SSI is our mission, our name, and our entire product roadmap.”

That remains essentially the only public description of what the company is building.

What Did NVIDIA See?

This is the biggest unanswered question.

SSI has published almost nothing about its current research.

There is no public model, no detailed architecture paper, no benchmark table, and no technical demo that allows outsiders to independently evaluate the work behind the partnership.

NVIDIA’s announcement nevertheless makes one unusually specific statement: the company received rare access to SSI’s closely guarded research before choosing to accelerate the partnership.

NVIDIA says SSI has been developing “a new research direction” aimed at powerful and robustly aligned AI.

SSI, meanwhile, says that research has now reached the point where scaling it is justified.

Those statements are meaningful, but they are still descriptions from the two companies directly involved in the transaction.

Without a paper, model, or reproducible experiment, outsiders cannot yet determine:

• What the new research direction actually is
• Whether it changes model architecture
• Whether it changes training objectives
• Whether it relies on new data or synthetic data
• Whether it changes post-training or reinforcement learning
• How its safety approach differs from existing alignment techniques
• What evidence convinced NVIDIA that the work should be scaled

For now, the strongest public signal is NVIDIA’s willingness to commit both funding and access to Vera Rubin.

The Return of Scaling

The partnership is especially striking because of Sutskever’s long association with the idea of scaling AI systems.

In 2020, OpenAI researchers published work showing that language-model loss follows predictable power-law relationships with model size, dataset size, and training compute.

That line of research helped reinforce an industry-wide strategy: if larger models trained on more data and compute improve predictably, then one of the most reliable ways to build stronger systems is to scale.

GPT-3 became one of the most visible demonstrations of that approach.

Over the following years, scaling drove increasingly large training runs across OpenAI, Google DeepMind, Anthropic, Meta, xAI, and other labs.

More recently, however, the industry has debated whether ordinary pre-training scaling is delivering diminishing returns. Research groups have increasingly explored reasoning, reinforcement learning, synthetic data, test-time compute, multimodal training, and new architectures.

Sutskever himself has publicly argued that AI is entering a period in which fundamental research matters again.

That makes his latest line unusually interesting.

SSI is not

It is simply saying it wants more GPUs.

It is saying that it has first spent roughly two years searching for a research direction, and now believes that direction deserves more compute.

In other words, the sequence appears to be:

1. Find a new research approach.
2. Test it privately.
3. Decide that it is promising enough to scale.
4. Expand compute by approximately 10x.

That is different from scaling an unchanged architecture simply because more hardware is available.

A New Scaling Curve — or Just a Bigger Experiment?

The original report frames the NVIDIA partnership as Sutskever returning to the “scaling arena.”

That interpretation is plausible, but the available evidence does not yet tell us what exactly SSI plans to scale.

It could involve a conventional frontier training run. It could also involve a different learning objective, architecture, safety mechanism, or post-training process.

The distinction matters.

A 10x increase in compute only becomes strategically important if the underlying method benefits from it.

SSI’s public wording suggests that the company believes it has crossed that threshold.

NVIDIA appears convinced enough to help provide the infrastructure.

The rest of the AI community still has to wait for evidence.

FAQ

What is Safe Superintelligence Inc.?

Safe Superintelligence Inc., or SSI, is an AI research company founded in 2024 by Ilya Sutskever, Daniel Gross, and Daniel Levy. Its stated goal is to build safe superintelligence without being distracted by consumer products or short-term product cycles.

Did NVIDIA officially invest $5 billion in SSI?

NVIDIA officially confirmed that it made a substantial investment, but it did not disclose the amount. Reuters reported a $5 billion equity investment based on a person briefed on the deal.

How much will SSI increase its compute?

NVIDIA says the partnership will increase SSI’s computing capacity by an order of magnitude. SSI’s own announcement describes that as roughly a 10x increase over the next 12 months.

What hardware will SSI use?

SSI will gain access to NVIDIA’s Vera Rubin platform. Vera Rubin is NVIDIA’s next-generation AI computing platform for large-scale training and inference.

Has SSI released a model yet?

No public SSI model was available at the time this article was prepared. The company has also released very little technical information about its research.

Why is SSI valued so highly without a product?

Investors appear to be betting heavily on Ilya Sutskever’s research track record and SSI’s potential to produce a major AI breakthrough. The company reportedly reached a $32 billion valuation in a 2025 funding round.

What did NVIDIA see in SSI’s research?

NVIDIA says it received rare access to SSI’s closely guarded work before entering the partnership. The companies have not publicly disclosed the underlying research details.

Is SSI simply scaling a larger language model?

That has not been confirmed. SSI says it has developed a new research direction that is now worth scaling, but it has not publicly described the architecture or training method.

Related Tools

• [Safe Superintelligence Inc.

](https://ssi.inc/): SSI’s official website and public mission statement.

• NVIDIA Vera Rubin: NVIDIA’s next-generation platform for large-scale AI training and inference.
• NVIDIA Developer: Technical resources for NVIDIA accelerated computing and AI infrastructure.
• Google Cloud TPU: Google’s purpose-built AI accelerator platform, which has also been used by SSI.
• JAX: A high-performance numerical computing framework widely used for large-scale AI research.
• PyTorch: An open-source machine-learning framework used for large-scale model research and training.

Related Links

• NVIDIA and SSI Official Partnership Announcement: NVIDIA’s official July 27 announcement covering the investment, Vera Rubin access, and 10x compute expansion.
• Safe Superintelligence Official Website: SSI’s original mission statement and company description.
• Reuters: NVIDIA to Invest $5 Billion in SSI: Independent reporting on the investment amount and transaction.
• OpenAI Scaling Laws for Neural Language Models: The 2020 research that formalized key empirical relationships between model performance, compute, data, and scale.
• GPT-3 Paper: The paper describing GPT-3 and the performance gains achieved by scaling language models.

Summary

NVIDIA and SSI have formed a long-term strategic partnership that gives Ilya Sutskever’s lab access to Vera Rubin systems and enough additional capacity to increase its compute by roughly 10x.

The companies officially disclosed a substantial NVIDIA investment but not its size. Reuters reports that the equity investment is worth $5 billion.

The most important part of the announcement may be SSI’s own explanation: after nearly two years of highly secretive research, the lab now believes its work has reached a stage where further scaling is justified.

The partnership does not reveal what SSI has discovered, but it gives the company the compute required to find out whether that private research direction continues to improve at scale.