AI Infrastructure Accelerates
Together AI Secures $800M Series C to Fund 500MW AI Capacity, Reaches $8.3B Valuation
Key Points
Together AI counts concrete customers by name, including Cursor, Cognition, and Decagon, illustrating real-world adoption beyond the broad 'thousands of paying customers' claim.
The Series C included a $1 billion primary tranche plus a $44.89 million extension, with a on-paper valuation of about $7.5 billion and reporting indicating an effective valuation above $8 billion.
Pearl Research Labs’ Pearl Network provides subsidies to inference costs via a Proof-of-Useful-Work protocol, with subsidies reportedly reducing costs by more than 25% for certain models.
Together AI’s serverless inference layer is positioned as delivering roughly 2x the performance of the fastest available alternatives.
Nvidia participated in the round as an investor, underscoring continued backing from hardware partners aligned with Together AI’s neocloud strategy.
Together AI raised $800 million in a Series C round led by Aramco Ventures, pushing the open-source AI infrastructure company to an $8.3 billion valuation, according to Business Wire. The deal nearly triples the company's $3.3 billion valuation from just 16 months ago and comes with a commitment to build more than 500 megawatts of compute capacity — enough to rival a small city's power demand.
The company reports annual bookings above $1.15 billion and says usage of open-source models on its platform has tripled in the past year. Nvidia joined the round as an investor, alongside Vista Equity Partners and General Catalyst, signaling that hardware makers are doubling down on the neocloud model — specialized clouds built specifically for AI workloads.
Saudi Oil Money Bets Big on AI Infrastructure
Aramco Ventures led the round, with its Prosperity7 arm managing the investment. Abhishek Shukla, managing director at Prosperity7, said building AI infrastructure "will be the biggest infrastructure project in human history" and praised Together AI for making open-source models "genuinely usable at enterprise scale," according to Business Wire. The move signals that Gulf sovereign capital is shifting bets from frontier AI labs toward the unglamorous but profitable infrastructure layer.
CEO Vipul Ved Prakash framed the company's mission in blunt terms. "Intelligence is becoming a foundational resource," he said, "every bit as essential as electricity." SiliconAngle notes that Together AI has evolved from a simple GPU broker into an owner-operator of data centers, with its Maryland facility going live in July 2025.
ATLAS Engine and Serverless Layer Cut Costs in Half
Together AI's core product is a serverless inference layer — a service that runs AI models without customers needing to manage GPUs or networks. Let's Data Science reports the layer delivers roughly 2x the speed of the fastest competing alternatives. Under the hood, an engine called ATLAS uses speculative decoding: a lightweight model drafts answers fast, and the main model checks and refines them, cutting response times while keeping accuracy intact.
For less urgent tasks, Together AI offers Batch Inference, which can cut costs by up to 50% by processing jobs in off-peak windows. The company says customers like Cursor, Cognition, and Decagon are already using the platform, with some reporting cost savings of 6x to 60x compared to closed-model APIs like those from OpenAI, according to Business Wire.
Blockchain Partnership Adds a 25% Subsidy for Some Models
In May 2026, Together AI struck a deal with Pearl Research Labs, whose Pearl Network uses a "Proof-of-Useful-Work" protocol. The idea: AI matrix math — the same calculations that run model inference — doubles as the computational work that secures a blockchain. That lets Pearl subsidize inference costs by more than 25% for certain model endpoints, according to Crypto Briefing.
Not everyone is sold. Tom's Hardware warned the partnership could spark a GPU mining rush and that subsidies may become unstable if the PRL token price drops. The risk is real: a volatile token price could erase the cost advantage overnight, leaving enterprise customers exposed to sudden pricing swings.
500 Megawatts of Capacity Bets on Winning the Power Race
The 500 MW capacity commitment is the round's most striking detail. Most major data center campuses run between 20 MW and 100 MW. Securing 500 MW — funded independently by investors, not from the $800 million itself — positions Together AI to sidestep the power-grid bottlenecks that have stalled rivals, according to Let's Data Science.
The company now serves over 450,000 developers, according to TechBuzz AI. With bookings at $1.15 billion and a valuation of $8.3 billion — roughly 7x bookings — the pressure to justify that multiple is real. But with Nvidia, Aramco, and General Catalyst all on the cap table, Together AI has the backing to make the infrastructure bet pay off.
Sentiment
Mostly PositivePublishers
13
Articles
86
Reach
99
