Run:ai
Last reviewed
Jun 3, 2026
Sources
9 citations
Review status
Source-backed
Revision
v1 · 1,436 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 3, 2026
Sources
9 citations
Review status
Source-backed
Revision
v1 · 1,436 words
Add missing citations, update stale details, or suggest a clearer explanation.
Run:ai (legal name Runai Labs Ltd.) is an Israeli software company that developed a Kubernetes-based orchestration and scheduling platform for graphics processing unit (GPU) resources used in artificial intelligence and machine-learning workloads. Founded in Tel Aviv in 2018, the company built software that pools GPUs into a shared resource, slices physical GPUs into fractions, and dynamically schedules training and inference jobs in order to raise the utilization of expensive accelerator hardware. In April 2024 Nvidia agreed to acquire Run:ai, and the deal closed on December 30, 2024 after antitrust reviews in the United States and the European Union. Upon closing, Nvidia announced that it would open-source the Run:ai software so that it could run on hardware beyond Nvidia's own GPUs.
Run:ai was founded in 2018 by Omri Geller, who served as chief executive officer, and Ronen Dar, who served as chief technology officer. The two met as graduate students in the School of Electrical Engineering at Tel Aviv University, where they were taught by Professor Meir Feder. Feder, an information-theory specialist who holds the Information Theory Chair at the university, joined as the company's third co-founder. The company was headquartered in Tel Aviv, Israel, with additional operations in the United States.
The founders set out to solve a practical bottleneck in AI development: GPUs are scarce and costly, yet they are frequently left idle or underused because traditional infrastructure tools were not designed to share accelerators efficiently across many users and jobs. Run:ai's premise was to introduce a virtualization and orchestration layer that would let organizations treat their GPUs as a single elastic pool rather than as fixed assets bound to individual machines or teams.
Run:ai's platform, marketed for a time under the name Atlas, was built on top of Kubernetes, the open-source container-orchestration system that has become a standard substrate for AI and cloud infrastructure. The software added a scheduling and virtualization layer on top of Kubernetes primitives, giving data-science and IT teams a centralized way to manage shared compute across on-premises clusters, public clouds, and hybrid environments.
Key capabilities of the platform included:
The company reported that customers were able to raise average GPU utilization substantially, with figures cited in the range of roughly 25 percent to 75 percent. The platform was designed to be tool-agnostic, supporting common AI frameworks and third-party tooling so that researchers could keep their preferred workflows. Run:ai had been a close technology partner of Nvidia since 2020, and its software integrated with Nvidia products including DGX systems, DGX SuperPOD, Base Command, NGC containers, and Nvidia AI Enterprise.
Before its acquisition, Run:ai raised a total of approximately 118 million US dollars in venture capital across several rounds. Early investors included the Israeli firms TLV Partners and S Capital VC, with later participation from larger growth investors. The Series B round, announced in January 2021, was led by Insight Partners and brought the company's total financing to 43 million dollars at that point. The Series C round in March 2022 was co-led by Tiger Global Management and Insight Partners and was the company's largest disclosed raise.
| Round | Date | Amount | Lead investor(s) |
|---|---|---|---|
| Series A | 2019 | ~$13 million | TLV Partners (early backer) |
| Series B | January 2021 | $30 million | Insight Partners |
| Series C | March 2022 | $75 million | Tiger Global Management, Insight Partners |
| Total | ~$118 million |
At the time of the Series C round, Run:ai reported rapid commercial growth, including a roughly ninefold increase in annual recurring revenue over the prior year and a tripling of headcount.
On April 24, 2024, Nvidia announced that it had entered into a definitive agreement to acquire Run:ai. Neither company disclosed the financial terms publicly. People familiar with the deal told TechCrunch and other outlets that the price was approximately 700 million dollars, the figure most widely cited. Israeli business media, including Calcalist, reported a somewhat higher total of around 800 million dollars when accounting for a retention component of roughly 200 million dollars payable to employees in Nvidia shares. Earlier reports during negotiations had speculated the deal could reach as high as 1 billion dollars.
In its announcement, Nvidia framed the acquisition as a way to help customers get more out of their AI infrastructure as workloads grow more complex. Alexis Bjorlin, an Nvidia vice president associated with DGX Cloud, noted that orchestrating generative AI, recommender systems, and search workloads requires sophisticated scheduling to optimize performance across systems and across the data-center fabric. Run:ai chief executive Omri Geller said the company had collaborated closely with Nvidia since 2020 and shared a focus on helping customers make the most of their infrastructure. Nvidia stated that it intended to maintain Run:ai's existing business model and continue investing in the product roadmap, including ongoing work to integrate the technology with DGX Cloud.
Because Nvidia is the dominant supplier of AI accelerators, the acquisition drew regulatory scrutiny over whether bringing a leading GPU-orchestration vendor in-house could harm competition. The United States Department of Justice opened an antitrust review of the transaction, reported to be underway by at least August 2024, and the European Commission also examined the deal.
On December 20, 2024, the European Commission unconditionally approved the acquisition, concluding that it would not raise competition concerns because customers would still have access to other hardware options compatible with the Run:ai software. With regulatory hurdles cleared, the parties closed the transaction on December 30, 2024.
When the acquisition closed, Nvidia announced that it would open-source the Run:ai software. At the time, the platform worked only with Nvidia products. Open-sourcing was intended to extend the software's availability to the broader AI ecosystem, meaning that, in principle, hardware from Nvidia competitors such as AMD and Intel could be supported. The company stated that opening the code would help the community build better AI faster. The move was widely read as an effort to address competition concerns while expanding the reach of Nvidia's infrastructure software.
The Run:ai acquisition reflected Nvidia's strategy of building out a full-stack AI computing business that spans hardware and software. Although Nvidia is best known for its GPUs and for networking assets acquired through Mellanox, the company has increasingly invested in the software layers that schedule, orchestrate, and optimize those chips. Run:ai's Kubernetes-based scheduler complements offerings such as Nvidia AI Enterprise and DGX Cloud by addressing the operational problem of keeping fleets of accelerators busy, a growing concern as organizations spend heavily on GPU clusters for training and serving large models.
The deal also fit a broader pattern of Nvidia acquiring and investing in AI-infrastructure startups, a category that includes its later interest in GPU-cloud orchestration vendors such as Lepton AI. By placing the Run:ai scheduler at the heart of its software stack and committing to open-source it, Nvidia positioned the technology both as a value-add for its own customers and as a way to influence how GPU orchestration is done across the industry.