Three raises, one week, one direction
Between June 29 and July 1, 2026, three separate funding rounds closed for AI infrastructure companies that have nothing to do with building a better chatbot. Together AI raised $800 million. A chip-architecture startup called OXMIQ raised $35 million. A data center coolant-monitoring company called Omen AI raised $31 million. None of the three compete with each other, and none of them make a foundation model. Read side by side, they describe where a growing share of AI capital is actually going: not into the next flagship release, but into the plumbing that keeps flagship releases running at scale.
The headline number: Together AI at $8.3 billion
Together AI closed an $800 million Series C on July 1, more than doubling its valuation to $8.3 billion from $3.3 billion at the start of 2025, according to the company's announcement and reporting from TechCrunch and Data Center Dynamics. Aramco Ventures led the round, with NVIDIA, General Catalyst, Vista Equity Partners and Emergence Capital participating.
Together AI runs a cloud platform that lets companies train and serve open-source models, including Llama, Mistral and Qwen, at a lower cost than proprietary alternatives from the closed labs. The company says it will use the new capital to expand its inference and fine-tuning products and to grow its infrastructure footprint roughly 50-fold over the next five years. That last figure is a company projection, not an audited result, and should be read as a stated plan rather than a settled fact. What is verifiable is the round itself and the investor list, which puts NVIDIA on both sides of the open-model economy: selling the chips and now backing one of the largest platforms for renting access to them.
The smaller rounds tell the same story
OXMIQ, founded by former Intel and AMD chip architect Raja Koduri, raised a $35 million Series A on July 1, bringing its total funding to $60 million, per its own announcement and coverage in HPCwire and Electronics Weekly. The round was co-led by Fundomo and Samsung Catalyst Fund, with MediaTek and Pegatron Venture Capital among the strategic backers.
OXMIQ's pitch is narrower than Together AI's: it licenses a GPU core design, called OxCore, that bundles a CUDA-compatible compute engine, a tensor processing engine and an orchestration engine into a single architecture that other chipmakers can build into their own silicon. Rather than manufacturing chips itself, OXMIQ wants to be licensed the way Arm licenses processor designs, lowering the cost for semiconductor companies that cannot afford to design custom AI silicon from scratch.
Omen AI closed a $31 million Series A on June 30, led by Nava Ventures with CRV and Sheryl Sandberg among the backers, according to the company's release and reporting from TechCrunch and SiliconANGLE. Its product has nothing to do with models or chips: it is a spectroscopic sensor that attaches to a data center's liquid cooling system and continuously tracks more than 21 elemental signatures in the coolant, flagging degradation before it forces a costly shutdown. The company says data centers representing $200 billion in assets and 10 to 14 gigawatts of capacity already use its monitoring, a figure that comes from Omen AI itself and has not been independently verified.
Who this actually affects
For enterprise buyers, the practical effect is more competition at the inference layer. Together AI's expansion gives companies training or running open-weight models another well-capitalized option beyond the closed labs, which should keep pressure on inference pricing across the market. For chipmakers and system integrators, OXMIQ's licensing model lowers the capital required to bring a custom AI chip to market, a bet that demand for specialized silicon will keep outpacing what NVIDIA and a handful of others can supply directly. For data center operators, Omen AI's raise is a signal that liquid cooling failures have become expensive enough, as GPU racks push thermal loads higher, to justify continuous monitoring as its own product category rather than a feature bolted onto existing facilities software.
None of these three rounds are large enough individually to be the story. Together, closing in the same three-day window, they show investors spreading AI infrastructure bets across compute access, chip design and the physical systems that keep it all cool, rather than concentrating everything on the next model release.