- Bloomberg reported on May 4, 2026 that Nvidia has backed AI inference platform DeepInfra in a $107 million funding round.
- The investment continues Nvidia‘s pattern of strategic investments in AI infrastructure companies that deepen demand for its GPUs.
- DeepInfra operates a serverless inference platform offering open-weight model APIs at lower per-token rates than the major frontier-lab APIs.
- The Bloomberg source is a video page; specific deal details — additional investors, valuation, and use of proceeds — should be confirmed against Bloomberg’s full coverage or DeepInfra’s own announcement.
What Happened
Nvidia backed AI inference platform DeepInfra in a $107 million funding round, Bloomberg reported on May 4, 2026. The Bloomberg source is a video segment rather than a written article, so detailed deal terms — co-investors, valuation, board seats, use of proceeds — are best confirmed against Bloomberg’s full reporting or DeepInfra’s own announcement.
Why It Matters
DeepInfra is one of a handful of serverless inference platforms — alongside Together, Fireworks AI, Anyscale, and Replicate — that host open-weight models and offer them via API at substantially lower per-token rates than the major frontier-lab APIs. The category has expanded rapidly through 2026 as the open-weight model landscape has densified: DeepSeek V4 Pro, Xiaomi MiMo-V2.5-Pro, Moonshot Kimi K2.6, Zhipu GLM 5.1, and Meta’s Llama family all run on these inference platforms. Nvidia’s investment positions it strategically: every inference workload running on these platforms uses Nvidia GPUs, so the company benefits commercially regardless of which open-weight model wins.
Technical Details
Detailed company specifications, deal structure, and named co-investors were not retrievable from the Bloomberg video URL during research. DeepInfra’s prior public footprint includes its serverless inference API, model catalog covering major open-weight families (Llama, Mixtral, DeepSeek, Qwen, and others), and per-token pricing competitive with the lowest tiers in the category.
Nvidia’s strategic-investment program — operating through NVentures, the corporate venture arm — has accelerated through 2026. The arm’s recent investments include Legora ($50 million Series D extension at $5.6 billion valuation announced April 30, NVentures’ first reported legal-AI investment) and now DeepInfra. The pattern is consistent: invest in AI infrastructure and applications layers that drive sustained GPU demand without competing directly with Nvidia’s own product lines.
Who’s Affected
DeepInfra’s existing customer base of developers using open-weight models gains a more capitalized inference partner. Together, Fireworks AI, Anyscale, and Replicate — DeepInfra’s main competitors — face a competitor with strategic Nvidia backing. Open-weight model providers (Meta, DeepSeek, Xiaomi, Moonshot, Zhipu) gain a more reliable inference distribution channel. End customers — particularly cost-sensitive deployments using DeepClaude-style routing tools — gain a stronger third-party inference option for the price-versus-capability tradeoff. Nvidia gains a deeper relationship with the inference layer where the next several years of GPU consumption will land.
What’s Next
DeepInfra’s product roadmap announcements over the coming months will indicate whether the new capital funds geographic expansion, dedicated GPU capacity reservations with Nvidia, or vertical integrations. Watch for parallel announcements from Together and Fireworks responding to the competitive pressure. We will follow up with deeper coverage once the full deal details are publicly available.
