NVIDIA and Safe Superintelligence (SSI) announced a long-term strategic partnership on July 27, 2026. NVIDIA will invest in SSI and provide access to its next-generation AI infrastructure, "Vera Rubin." This is expected to increase SSI's computing resources by an order of magnitude—roughly tenfold. For SSI, which has continued its research behind closed doors, this marks a shift toward testing its research direction at massive computational scale.
Beyond access to Vera Rubin, the partnership includes an arrangement to jointly improve NVIDIA's current and future infrastructure. SSI gains access to large-scale computing infrastructure, while NVIDIA gains a pathway to incorporate SSI's research-driven requirements into the development of future infrastructure. However, neither company has disclosed the scale of usage, the form of access, or the launch date. Without clarifying what exactly constitutes this "roughly tenfold" increase, it's difficult to gauge the real impact of the partnership.
A Tenfold Compute Bet on Closed-Door Research
NVIDIA explains that it decided on the partnership after gaining unusual access to SSI's tightly guarded research. According to both companies, SSI has spent the past two years pursuing a new research direction aimed at achieving "powerful, robustly aligned artificial intelligence." In the joint announcement, Ilya Sutskever stated, "We have research that is worth scaling, and access to large-scale NVIDIA computers will let us do that."
NVIDIA committed capital and infrastructure after reviewing the research content. However, no material has been provided that would allow outside parties to verify the research. The joint announcement offers no details on model architecture or training methods, nor does it disclose any safety metrics or performance benchmarks achieved so far.
SSI was founded in 2024 and is currently led by Sutskever and Daniel Levy. Its stated goal and product remain singular: "safe superintelligence." The roughly tenfold increase in compute represents an expansion aimed at validating, at greater scale, the research direction chosen in pursuit of that single goal.
Vera Rubin Redefines the Unit of Scale
The Vera Rubin NVL72 connects 72 Rubin GPUs and 36 Vera CPUs via NVLink 6, integrating ConnectX-9 SuperNICs and BlueField-4 DPUs within the same rack. The CPUs handle data movement and control, while NVLink 6 supports GPU-to-GPU communication within the rack. Networking and storage processing are also built into this rack-scale design.
Furthermore, NVIDIA treats five distinct rack types—NVL72, Vera CPU, Groq 3 LPX, BlueField-4 STX storage, and Spectrum-6 SPX Ethernet—as components of a single AI supercomputer. Vera Rubin expands the unit of scale from individual GPUs to racks, and further to PODs composed of multiple racks. The "computing resources" SSI gains access to refer to this integrated infrastructure.
Compared to the Blackwell generation, NVIDIA claims Vera Rubin can reduce the number of GPUs required to train large Mixture-of-Experts (MoE) models by a factor of four, while boosting inference throughput per watt by up to tenfold and cutting the cost per token to one-tenth. These figures are not measurements derived from SSI's own research. Still, they make clear that the phrase "roughly tenfold increase in compute" doesn't necessarily mean a tenfold increase in GPU count. A combination of improved processing efficiency from the new generation and an increase in deployed units could together drive up total computing capacity.
The $5 Billion Report and the Link Between Capital, Compute, and Development
The official announcement describes NVIDIA's investment only as "significant," without disclosing financial terms. However, Bloomberg reported, citing multiple sources familiar with the matter, that the investment amounts to $5 billion. According to Bloomberg, SSI had previously raised approximately $3 billion and was valued at $32 billion as of 2025. If the $5 billion figure is accurate, NVIDIA's investment alone would exceed the total funding SSI had raised up to that point.
Under this arrangement, NVIDIA is injecting capital while also providing access to its own computing infrastructure. This allows SSI to scale up its research. In turn, NVIDIA states that it will use insights from SSI's future AI research to technically advance both its current and next-generation computing infrastructure. Capital, access to compute, and research feedback have all been folded into a single partnership.
However, it remains undisclosed whether the invested funds will flow back to NVIDIA as payment for Vera Rubin systems. It's also unclear whether SSI will own the systems outright, lease them via cloud services, or use facilities operated by NVIDIA. Understanding how capital and product demand are connected requires knowing how the deal is actually implemented.
Neither GPU Count Nor Launch Date Yet Visible
Vera Rubin has entered mass production, with NVIDIA planning to begin shipping products starting in fall 2026. This partnership represents a commitment to give SSI access to infrastructure ahead of that shipping launch. However, the joint announcement does not specify deployment locations, the number of GPUs or racks involved, or any phased expansion plans.
Even the scale of "roughly tenfold" lacks a clear baseline. Since SSI's pre-expansion computing resources—whether owned or leased—remain undisclosed, the absolute scale after the tenfold increase cannot be calculated. Similarly, Vera Rubin's performance figures are based on general workload comparisons provided by NVIDIA, with no guarantee that the same efficiency gains will apply to the specific research methods SSI has chosen.
The first real measure of this partnership's success will come once Vera Rubin's shipment or cloud deployment is confirmed and SSI discloses the scale of its usage. Beyond that, the key benchmark will be whether the expanded computing resources translate into reproducible research evaluations and verifiable safety methodologies. The roughly tenfold increase in compute is merely a starting point—it will be the first research results SSI shares publicly that ultimately determine the value of this major partnership.