The Aramco-Led Leap
Nobody in Sand Hill Road expected a continuation of the crazy 2024 multiples, but here we are. Together AI just locked down an $800 million Series C. Lead investor Aramco Ventures didn’t blink at an $8.3 billion valuation, dragging along a syndicate of usual suspects: General Catalyst, Emergence Capital, Nvidia, March Capital, Pegatron, Vista Equity Partners, and SentinelOne's S Ventures. Contrast that with their early 2025 Series B. They raised $305 million at a $3.3 billion valuation back then. To more than double your valuation in roughly eighteen months during a supposed capital correction is a neat trick. It shows that sovereign wealth and strategic tech money are now dictating top-tier pricing, bypassing traditional venture discipline entirely.
We used to look at 10x forward revenue as aggressive. Today, if you control access to high-performance clusters, the rules change. Sovereign funds aren't playing for quarterly IRR; they are buying entry tickets to a new computing era. This $800 million check gives Together AI the balance sheet to go toe-to-toe with hyperscalers. According to the official blog post, they're scaling up immediately. You don't raise this kind of cash unless you plan to dump it all directly into hardware capital expenditures.
Bookings Meet Neocloud Realities
Let's look at the underlying math. Together AI is reporting an annual bookings rate of over $1.15 billion as of last quarter. That is a massive number for a company founded in 2022. But bookings are not recognized revenue, and GPU computing contracts have a way of looking inflated on paper. Neoclouds act as specialized landlords, buying or leasing massive GPU clusters to rent them out to startups. It is a capital-intensive business model disguised as a high-margin software play.
The core service is hosting open-source models. Instead of subscribing to closed APIs, enterprises lease compute to run their own fine-tuned models. It is cheaper. It is private. Startups like Cursor, Cognition, and Decagon are actively building their systems on Together's infrastructure. These developer-focused companies need high reliability and low latency, which means they are locked into Together's clusters. While the margins here look nothing like traditional SaaS—bandwidth and depreciation costs eat a massive chunk of every dollar—the top-line growth is too spectacular for investors to ignore. If you run the math, an $8.3 billion valuation on a $1.15 billion booking run rate implies a multiple under 8x bookings. In this market, that actually looks reasonable, even if the underlying margins are thinner than we'd like to admit. It was reported by TechCrunch that the deal was closed despite the shifting macro environment.
The Open-Source Infrastructure Boom
This capital injection is not happening in a vacuum. The industry is moving away from proprietary foundation models. The cost of subscribing to closed-source frontier models has driven developers to seek alternatives, and open-source models have filled the gap. Data from OpenRouter shows that open-source model usage tripled industry-wide in the past year alone.
This pivot has created a parallel demand for specialized hosting. You cannot just run these models on basic cloud servers. They require highly optimized GPU orchestration. Together AI specializes in this middle layer, offering a developer platform that handles the complexities of serving open-source weights. By focusing on open-source, they avoid competing directly with OpenAI's proprietary model ecosystem. Instead, they position themselves as the infrastructure provider of choice for developers who want complete control over their weights. It is a smart bet. If open-source continues to eat the market, Together AI's infrastructure becomes the default highway for enterprise intelligence.
The Crowd of Neocloud Contenders
Together AI is not the only player riding this wave. The neocloud sector is flooded with capital, and we are seeing a massive rush to secure GPU capacity. In June 2026, Upscale AI closed a massive $500 million combined Series A and extension round, pushing its valuation to $2 billion. Around the same time, TensorWave, a neocloud focusing on AMD's alternative silicon, secured a $350 million Series B at a $1.55 billion valuation.
This level of funding suggests a frantic land grab. Startups are raising hundreds of millions not to build software, but to prepay for chip allocations. The risk is obvious: if the demand for AI compute softens, these neoclouds will be stuck with depreciating hardware and massive lease liabilities. However, for now, the bottleneck is real. Companies are betting that Nvidia is not the only game in town, while Together AI is leveraging its early lead to lock down long-term customer commitments. The neocloud wars have officially begun, and the winners will be determined by who can maintain utilization rates when the supply of chips finally catches up with demand.
Foundational Roots and Strategic Paths
The technical foundation of Together AI explains why investors are willing to back them over generic server farms. The company was founded in 2022 by a pedigree-heavy trio: Vipul Ved Prakash, Percy Liang, and Ce Zhang. Prakash has a track record of building and selling; he sold his search startup Topsy to Apple back in 2013. Liang is a Stanford professor who directs the Center for Research on Foundation Models. Zhang is an associate professor at ETH Zürich and UChicago.
This combination of academic research experience and commercial execution is rare. It allowed Together AI to build proprietary software that optimizes cluster performance, rather than just renting raw chips. Their developer platform does not just host models; it makes them run faster and cheaper. This software edge is their primary defense against commoditization. While any sovereign fund can buy GPUs and build a datacenter, only a few teams can build the orchestration layer required to run them efficiently. That is the premium Aramco and General Catalyst are paying for. They are not buying silicon; they are buying the engineering talent capable of making that silicon useful.