Nvidia backs rivals, hires safety team

- Nvidia on August 4 helped launch new AI-security guidelines at Black Hat while expanding support for open-weight models and customers that could reduce dependence on it. - More than 120 organizations are now in the Open Secure AI Alliance, and AMD said its “backed, open-source software stack” is a selling point. - Nvidia’s next public marker is continued SAFE guideline work through the Linux Foundation and Open Secure AI Alliance participants.

Nvidia is trying to hold two positions at once. It is still the dominant supplier of the AI chips and software stack that power the current boom, but it is also funding and tooling parts of the ecosystem that could eventually weaken its grip on the most profitable layer. At the same time, the company is putting more visible structure around AI security and safety work as open-weight models become a bigger part of the market. The result is a strategy that looks less like lock-in at all costs and more like owning the infrastructure around a more open AI market. ### Why would Nvidia support companies that could become alternatives to its own products? Nvidia has been backing major AI customers whose long-term goals include reducing dependence on outside model providers and, in some cases, on Nvidia’s highest-margin software and services layer, according to reporting cited in the source briefings. That is not as contradictory as it sounds. If more companies build models, deploy inference systems, and train specialized AI, they still need accelerators, networking, and developer tools. Jensen Huang has been making a version of that argument for months: AI will be built by many companies and many countries, not just a few frontier labs, according to public comments referenced in recent reporting on open-weight models. Nvidia’s commercial logic is that a broader AI field can still run on Nvidia hardware even if the application layer fragments. ### What changed this week on security? Black Hat in Las Vegas gave Nvidia a venue to push AI security in public. The Linux Foundation on August 4 shared a request for comments on Shared AI Findings Exchange, or SAFE, a proposed framework meant to turn agentic AI incidents into shared defensive information across the industry. Nvidia said members of the Open Secure AI Alliance, now more than 120 organizations, are contributing to that work. (africa.businessinsider.com) Nvidia had already announced the Open Secure AI Alliance in late July with founding members across software, infrastructure, and security. The company said the group would build and share open tools for AI safety and security, tying its security push directly to a broader argument that open software is “accessible and observable” in ways that can help defenders. ### How does an internal safety team fit with a push for open-weight models? (blogs.nvidia.com) Business Insider reported that Nvidia is staffing a new AI safety team as it doubles down on open-weight models. That pairing matters because Nvidia is not presenting openness and safety as opposing goals. Its public trust and security materials frame the company’s position around securing models, infrastructure, and data rather than restricting access by default. (blogs.nvidia.com) The company’s recent public messaging after the launch of the alliance made the same case more explicitly. Nvidia said open and closed models can be complementary, while arguing that more transparent systems can improve forensics and shared defense when incidents happen. That does not settle the policy debate around open models, but it does show where Nvidia is placing its institutional weight. (businessinsider.com) ### Where does AMD see an opening? AMD used its second-quarter results this week to underline openness as a competitive theme. In an August 4 earnings release, AMD reported $11.5 billion in second-quarter revenue and continued promoting ROCm as an open software stack for AI workloads. The company has separately described ROCm as a “backed, open-source software stack,” emphasizing support and portability rather than openness as a slogan. (africa.businessinsider.com) That argument is aimed at Nvidia’s biggest software advantage: CUDA. If customers want easier migration paths, less proprietary dependence, or broader ecosystem participation, AMD wants ROCm and related tools to be seen as a practical alternative. Nvidia’s answer, at least this week, is not to retreat from openness but to define it on terms that still keep Nvidia central to deployment, security, and scale. (ir.amd.com) ### Why do investors seem comfortable with that tension? Nvidia’s public posture suggests it believes ecosystem expansion can outweigh the risk of enabling future rivals. As long as model builders, cloud providers, and enterprises keep buying accelerated compute, Nvidia can benefit even if more of the software and model layer becomes portable or open. That is consistent with the company’s recent emphasis on platforms, security tooling, and industry alliances rather than a narrower defense of closed systems. (amd.com) The next test will be execution. SAFE is now in request-for-comments form through the Linux Foundation, and the Open Secure AI Alliance says participating companies will keep developing shared guidance and tools as the Black Hat process continues. (blogs.nvidia.com 1) (blogs.nvidia.com 2)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.