<\/p>\n\n\n\n
When S Capital first backed Run:ai in 2018, there was no AI boom to ride. Large language models weren\u2019t part of the mainstream vocabulary and ChatGPT was still far into the future. GPUs were powerful but inefficient, and the problem Run:ai set out to solve, orchestrating and optimizing GPU usage at scale, was largely invisible. \u201cThe word of AI was not as it is considered today,\u201d investor Aya Peterburg recalled. \u201cNo one knew to say even \u2018LLM\u2019 or what that means.\u201d<\/p>\n\n\n\n
Seen through that lens, NVIDIA\u2019s ultimate 2024 acquisition of Run:ai was not a bet on hype, but the continuation of a strategy that began four years earlier with its purchase of Mellanox Technologies. It first embedded NVIDIA into Israel\u2019s engineering ecosystem, anchoring its data-center, networking, and high-performance computing capabilities in the country.<\/p>\n\n\n\n
Run:ai and others would then extend that footprint into the software layer that determines how AI actually runs at scale. Together, the two acquisitions tell a coherent story: Israel is not just a distant satellite office for NVIDIA, but where some of its most critical infrastructure problems are solved. Run:ai identified a structural inefficiency at the heart of AI computing. Companies were paying for expensive GPUs while extracting only a fraction of their potential. As Peterburg explained, customers were \u201cpay[ing] more money to NVIDIA\u2026 but they got back in the day only 15% of what they paid for\u201d because GPUs running large workloads couldn\u2019t communicate efficiently.<\/p>\n\n\n\n