Take a closer look at Nvidia's buy of Run.ai, European Commission told

Updated A left-of-center think tank along with other non-profits are urging the European Commission to "fully investigate" Nvidia's purchase of workload management startup Run:ai amid worries its will help to tighten the GPU titan's grip on the AI industry.

The price of the takeover, first revealed in April, has yet to be confirmed, although Tel Aviv-based AI-centric Kubernetes orchestration biz Run.ai was valued at $700 million after raising $18 million in cash over four funding rounds since its formation in 2018.

Run.ai's platform offers a central Unser interface and control plane that lets customers use a variety of Kubernetes flavors, incorporating some of the same tools for managing namespace, resource allocations and the like.

As we've pointed out before, the main point of distinction for Run.ai is that platform is built to integrate with third party AI tools and frameworks with respect to GPU accelerated container environments. It already supports Nvidia's DGX compute platform, and Nvidia's has pledged to maintain the same biz model for Run.ai.

Nevertheless, the Open Markets Institute, and eight other organizations have written to the EC, calling on competition officials to inspect concerns that the transaction would allegedly "help Nvidia reinforce its super-dominance" in supplying GPUs by "integrating Run.ai software to build an additional barrier around its chip empire."

"We are a coalition of civil society organisations deeply concerned about the extreme and growing concentration of power in the artificial intelligence technology stack – particularly in the supply of computing power – which this merger threatens to make worse," the submission states [PDF].

In addition to the Open Markets Institute, other signatories include campaign groups Article19, FoxGlove, Balanced Economy Project, AI Now Institute, Gentium, SOMO, IT for Change and Idec.

According to some - perhaps conservative - estimates, Nvidia already has a dominant market share in terms of supplying the processors that train and deploy large language models. Revenue from AI semiconductors is estimated to be up 33 percent year-on-year in 2024 to $71 billion.

Nvidia's sales in the nine months ended October 27, 2024 were $91.2 billion versus $38.82 billion for the corresponding period in the prior year. Operating income was up to $57.4 billion compared to $19.35 billion.

The submission to the EC from the Open Markets Institute and co urges the EC to "launch a full investigation into the merger, focusing on the risks it poses to the entire AI sector."

"The merger will reinforce Nvidia's control over the supply of the advanced chips that are a critical input in downstream markets, particularly cloud and AI," it adds. "Nvidia has become a key input provider across the entire AI sector since it supplies both the dominant cloud providers – despite their efforts to develop their own competing chips – as well as smaller 'neo-cloud' startups such as Coreweave, Crusoe or Lambda1, all of whom subsequently provide cloud computing and AI services to a wide range of businesses in Europe and worldwide."

Nvidia is "booked up" in terms of inventory sales allocation for the next year and, as we've already pointed out, is the kingmaker with respect to which companies do or do not get its chips.

The submission continues:

The letter says the signatories estimate Nvidia has an 88 percent GPU sales market share, worldwide, and is leaving Intel and AMD to eat its dust. It claims Nvidia is using CUDA to shut out competitors, "developing its software offer to build a moat around its chips. It has built a closed environment in which its GPUs and proprietary software are tightly integrated."

The letter continues: "It is against this background that Nvidia is acquiring Run:ai. By optimising the use of scarce computing resources, Run:ai's GPU orchestration and virtualization services have the potential to significantly improve the performance of Nvidia's (already market-leading) GPUs and reduce the computation costs associated with AI development. Orchestration is already considered an important asset by GPU users as it scales resources across workloads to maximize performance and reduce cost, and it is therefore quickly becoming a standard market practice."

• Cheat codes for LLM performance: An introduction to speculative decoding
• Broadcom says VMware is a better money-making machine than it hoped
• China gorging on silicon before Uncle Sam slams the door
• China strikes back with Nvidia antitrust probe as US tightens tech chokehold

Foreclosing access to Run.ai from rivals is a major concern, the letter states, and Nvidia will have the "ability and incentive to tie or bundle Run:ai's services with its GPUs."

The letter adds "the Commission must not be misled by Nvidia's announcement that Run:ai services will remain open source.

"There are different degrees of openness and AI companies frequently use 'open' rhetoric in ways that actually reinforce their market power. For instance, 'open source' can include serious limitations to competition such as the prohibition to use or develop a competing product. Such an approach would not preclude Nvidia from restricting access to the relevant technology or shutting it off completely at a later stage."

In a statement sent to The Register, a spokesperson at Nvidia, said:

"We look forward to welcoming the Run:ai employees to our team. Although Run:ai supports only NVIDIA GPUs today, we plan to open-source Run:ai, making it available to all, as soon as the regulatory process is wrapped up. Not surprisingly, we haven't heard complaints from customers, partners, or even competitors about our plan."

The European Commission told us the deadline to "take a decision" on Nvidia's purchase of Run.ai is December 20, and it has no further comment to make at this stage. ®

Updated to add on December 20:

The European Commission cleared the acquistion today, concluding that the transaction would raise no competition concerns in the European Economic Area.