Together AI raises $800M at $8.3B as open-model demand triples

The open-model economy just got its strongest price signal yet. Together AI, the neocloud built for running open-source models, raised an $800 million Series C at an $8.3 billion valuation, per the company’s announcement and TechCrunch’s reporting. That is a 2.5x markup on its valuation from roughly sixteen months ago.

The round was led by Aramco Ventures with Nvidia, Vista Equity Partners, General Catalyst and others participating, and the numbers behind it are the interesting part: the company claims annual bookings above $1.15 billion, and separately secured commitments for more than 500 megawatts of compute capacity to be built by investors.

Open-model inference found its business model

Together’s bet is that enterprises increasingly want frontier-adjacent capability without frontier-lab lock-in, rent the infrastructure, run open weights, keep the workload portable. A billion dollars of bookings says that demand is real, and it slots directly into the pattern we mapped in the AI capital stack: specialized capacity providers converting demand into financeable contracts, the same layer where Crusoe is reportedly raising at $30 billion.

There is also a quiet geopolitical note: Saudi capital leading the round for open-model infrastructure, while closed-model labs subsidize their own lock-in, sketches the two competing economics of the inference market.

Reading the bookings number

The $1.15 billion figure deserves the scrutiny any bookings claim does. Bookings are commitments, not recognized revenue: multi-year contracts counted at full value, some portion of which depends on customers still needing the capacity in year three. The neocloud sector has learned to lead with the bigger number, and investors have learned to ask what the annualized recognized figure underneath it is. Together has not broken that out publicly, which is normal for a private company and exactly why the conversion question tops the watch list.

The 500-megawatt commitment is the more unusual structure: investors committing to build capacity the company will operate shifts the capital burden off Together’s balance sheet, echoing the financing patterns in IREN’s prepaid contracts and Nscale’s bank revolver. And the competitive backdrop is the real test of the thesis: Together’s margin depends on the spread between what open-model inference costs to serve and what closed-model APIs charge. Every frontier-lab price cut narrows that spread, and the open-weight releases arriving from Thinking Machines and Moonshot keep raising the quality of what the open stack can serve. Together wins if open models stay good enough and closed prices stay high enough, a corridor that both sides of the market are actively moving.

What to watch

Watch whether bookings convert to recognized revenue at scale, how the 500-megawatt buildout is financed and sited, and whether open-model demand keeps compounding as frontier labs cut prices to defend share. SambaNova’s new $11 billion valuation is the same investor thesis applied to inference hardware rather than model access.