Together AI Completes $800M Series C Funding at an $8.3B Valuation

TL;DR

Together AI announced on July 1 that it had completed an $800 million Series C funding round, achieving a post-money valuation of $8.3 billion—a significant increase from its previous valuation of $3.3 billion at the start of 2025. The company also disclosed that its annualized bookings for the last quarter exceeded $1.15 billion, with customers saving between 6x and 60x in costs by using open-source models on its platform compared to proprietary models.

On the other hand, pressure on AI infrastructure is becoming evident. According to Axios, citing Bloomberg, Meta is considering selling access to its AI models and excess computing power through a new cloud business. Oracle’s annual report through May 31, 2026, also disclosed risks related to long-term data center leases, power purchase commitments, and changing customer demand.

This comparison puts the question before investors: whether AI computing power is scarce is no longer the only variable. The more realistic challenge is who can consistently keep expensive electricity, GPUs, and data centers fully utilized.

Let’s break it down simply: Training teaches the model, while inference is the model answering questions, writing code, handling customer service, and generating content on a daily basis. The former is like building a factory; the latter is the daily output once the factory is operational.

Together's growth has primarily come from inference. It specializes in offering more affordable AI cloud services using open-source models, enabling developers and AI application companies to reduce their reliance on proprietary large model APIs. For customers, the key variable is whether the cost per inference call can continue to decrease.

The company’s disclosed clients include AI application companies such as Cursor, Cognition, and Decagon. Together stated that Decagon reduced its inference costs by approximately six times after using its platform. The company also cited industry data indicating that the usage of open-source models has tripled over the past 12 months.

This explains why capital is willing to assign an $8.3 billion valuation. For AI applications to move from demos to everyday use, inference costs must decline. As long as usage grows faster than unit prices fall, cheaper compute power will amplify overall demand.

Together CEO Vipul Ved Prakash’s statement is typical: intelligence is becoming a foundational resource like electricity, bandwidth, or capital, and an open ecosystem enables innovation to be cheaper and faster. This is the core belief of the optimists and the foundation for continued investment by firms such as Aramco Ventures and NVIDIA.

However, annualized bookings are not actual revenue. They more closely reflect the company’s disclosed order and contract momentum, indicating demand strength, but do not mean cash has been received or that renewals will occur annually.

If you look only at Together, the conclusion might be that AI computing power remains in short supply. However, signals from Meta and Oracle indicate that infrastructure investment is beginning to enter a tiered phase.

Meta is reportedly considering selling excess computing power and model access to external customers, which shouldn't be viewed as a negative development. For large tech companies, offering unused computing resources that aren't fully utilized for internal training or product demands is a natural way to improve asset utilization.

The issue is that this also indicates the construction speed has become so fast that external consumption channels must be actively sought. Computing power is no longer simply a matter of buying as much as you can; it’s now about whether you can consistently keep it filled with paid tasks.

Oracle's annual report provides more specific constraints. According to the filing, as of the end of May 2026, the company had $260 billion in unstarted lease commitments, primarily related to data center arrangements, with terms ranging from 15 to 19 years. Its capital expenditures are projected to rise from $21.2 billion in fiscal year 2025 to $55.7 billion in fiscal year 2026, primarily to expand data center capacity.

These figures cannot directly prove industry oversupply. Risk disclosures in public companies’ annual reports are inherently conservative. However, they highlight the most vulnerable aspect of AI infrastructure investment: capital expenditures occur upfront, while revenue follows later; electricity and leasing commitments are long-term, but customer demand may change more rapidly.

Together’s $8.3 billion valuation cannot be simply explained by AI hype. It assumes the company will not only secure orders but also convert open-source inference demand into sustainable long-term revenue through sufficiently high utilization, stable renewals, and healthy gross margins.

Another key concept is megawatt capacity. Megawatt refers to the power budget of a data center, determining how many GPUs it can support. Locked capacity means a company has secured the electricity and data center resources needed for future expansion, but it does not mean these resources have been deployed or are fully occupied by paid workloads.

For AI cloud companies, capacity is a double-edged sword. Failing to secure enough power and GPUs means missing out on surging demand, while overcommitting and failing to meet customer absorption leads to depreciation, electricity, and leasing costs hitting the income statement first.

This is also the difference between Together and large cloud providers. Together’s advantage lies in its focus on open-source inference, where customers may prioritize cost, speed, and model selection. Large cloud providers, on the other hand, serve enterprise clients, offer full-stack services, and have stronger balance sheets.

More likely, open-source inference demand will continue to grow, enabling specialized players like Together to achieve high growth. Meanwhile, some large-scale cloud and data center investments may yield lower returns than early market expectations due to contract mismatches, customer concentration, or slow utilization ramp-up.

AI infrastructure has not yet reached the evidence stage of a bubble burst, nor should demand be assumed to be unlimited just because Together raised funding. A more reasonable assessment is that the industry is transitioning from a phase of resource acquisition to one of validating resource monetization capabilities.

The variables the market focuses on will become increasingly specific. Funding size and valuation can only indicate that capital is willing to take a risk; they cannot replace metrics such as utilization rate, renewal rate, gross margin, and customer composition. If orders primarily come from well-funded early-stage AI companies, demand elasticity will be higher. Only when customers can be retained as long-term production environments will the valuation be more solidly supported.

Meta’s cloud computing efforts will provide the market with a price reference. When hyperscale companies offer their internal computing power for sale, the pricing power and differentiated services of external AI cloud providers will be tested. Oracle’s long-term commitment will also continue to remind investors that electricity and data center resources are not free options.

This round of funding underscores that demand for open-source inference continues to grow. But for investors, the evaluation has shifted: AI compute is not valuable simply by being built—it becomes a true infrastructure asset only when used consistently and with high margins.

律动 BlockBeats