The term "neocloud" barely existed eighteen months ago. Now it commands billion-dollar checks. Together AI's $800 million raise at an $8.3 billion valuation marks the moment when GPU-as-a-service graduates from scrappy alternative to legitimate infrastructure layer — one that investors believe can compete with, or at least complement, the hyperscalers.
The San Francisco company, founded in 2022 by researchers from Stanford and Google, has built its business on a simple premise: not everyone wants to pay Amazon, Microsoft, or Google for AI compute, and not everyone can get Nvidia chips fast enough to build their own clusters. Together AI slots into that gap, offering inference and fine-tuning for open-source models like Llama and Mistral at prices that undercut the big three.
The economics of being second-best
Together AI's pitch works because the hyperscalers are distracted. AWS, Azure, and Google Cloud are busy selling their own proprietary AI services — Bedrock, Azure OpenAI, Vertex AI — which means their raw compute offerings often feel like afterthoughts. Together AI doesn't have a foundation model to push. It just wants to rent you GPUs and run whatever model you bring, which turns out to be exactly what a certain class of enterprise customer wants.
The valuation math is aggressive but not insane. At $8.3 billion, investors are betting Together AI can reach $500 million to $1 billion in annual revenue within a few years. The company hasn't disclosed current figures, but the neocloud market is growing fast enough — Anyscale, CoreWeave, and Lambda Labs are all scaling rapidly — that the bet isn't pure fantasy. CoreWeave reportedly hit $2 billion in annualized revenue earlier this year.
Why open-source models matter here
Together AI's fortunes are tied to the open-source AI ecosystem in ways that make its business model both promising and precarious. If Meta keeps releasing capable Llama models for free, and if Mistral and others continue to close the gap with proprietary systems, Together AI has an ever-expanding catalog of models to serve. But if OpenAI and Anthropic maintain a significant capability lead, enterprises may decide the premium for closed models is worth paying.
The Anthropic export restrictions have, ironically, helped companies like Together AI. With Claude unavailable in certain markets and use cases, developers are turning to open alternatives — and they need somewhere to run them. Together AI has reportedly seen significant growth in Asia-Pacific markets where Anthropic's models are now harder to access.
Our take
The neocloud category exists because the AI boom happened faster than the cloud giants could adapt. Together AI's $8.3 billion valuation is a bet that this window stays open — that enterprises will continue to want Switzerland-style neutrality in their AI infrastructure, and that open-source models will remain competitive enough to matter. Both assumptions are reasonable today. Whether they hold in 2028, when the hyperscalers have had time to respond and the model landscape has consolidated, is the $8.3 billion question.
