Explosive adoption of open-source AI models has propelled two key players to new heights. Together AI now generates $1 billion in annualized revenue. Hugging Face has expanded to 13 million users with more than 2 million public models hosted.
But the story runs deeper than raw numbers. These companies don’t just benefit from the wave of open models. They actively shape how developers, enterprises and researchers build with them. Their growth reveals a maturing market where free weights and community-driven innovation translate into serious commercial success.
The Information reported last week that paying subscribers to Hugging Face’s repository doubled between January and June, crediting the broader open-source momentum. https://www.theinformation.com/articles/open-source-growth-boosts-together-ai-hugging-face
That surge aligns with data from Hugging Face itself. Its Spring 2026 state-of-open-source report shows the platform nearly doubled users, models and datasets in 2025 alone. Activity has shifted. Users no longer simply download pre-trained systems. They create fine-tuned models, adapters, benchmarks and full applications. Independent developers now account for 39% of development activity, up from 17% before 2022.
Hugging Face Turns Community Scale Into Enterprise Revenue
Hugging Face began as a chatbot experiment. It evolved into the central hub for open AI artifacts. By 2025 it counted 13 million users, over 2 million public models and more than 500,000 public datasets, according to its own analysis. Robotics datasets exploded from 1,145 in 2024 to 26,991 last year. They now represent the largest category on the Hub.
Enterprise interest has followed. More than 30% of Fortune 500 companies maintain verified accounts. Legacy firms upgrade subscriptions at a faster clip. Big Tech contributes heavily too. NVIDIA leads in new repository creation. Chinese organizations such as Alibaba, Baidu, ByteDance and Tencent have dramatically increased their open releases. Baidu went from zero repositories in 2024 to more than 100 in 2025.
Monetization has kept pace. Sacra estimates Hugging Face reached $70 million in ARR by the end of 2023, a 367% jump from 2022. Later reports place 2024 revenue near $130 million. The company sits at a $4.5 billion valuation after raising roughly $396 million total, with backers including Salesforce Ventures, Google, NVIDIA and others. It has grown to serve 500,000 organizations.
Much of that revenue flows through enterprise subscriptions, inference services and a clever routing layer called Inference Providers. The system integrates third-party compute providers — including Together AI — directly into model pages and SDKs. Users get billed through their Hugging Face account. No extra markup. This setup funnels traffic to paid usage while keeping the core Hub open and free. https://sacra.com/c/hugging-face/
Partnerships amplify the effect. Microsoft now surfaces over 10,000 Hugging Face models in Azure AI Foundry. Google Cloud usage has grown 10x over three years, moving tens of petabytes of models monthly.
Thomas Wolf, chief scientist at Hugging Face, captured the shift in a 2025 MIT Sloan interview. He noted the platform now hosts close to 4 million models in some counts and serves roughly 10 million AI builders who both share and download artifacts. The focus remains accessibility and community feedback mechanisms that resemble social media upvotes.
Yet the real power lies in derivatives. Top 200 models drive nearly half of all downloads, but the long tail of community fine-tunes and adaptations creates stickiness. Organizations that rely solely on closed systems often face higher long-term costs, the Hugging Face report argues. Open artifacts generate downstream value that dwarfs initial expenses.
Together AI takes a different slice of the same pie. The company operates an AI-native cloud optimized for open models. It offers fast inference, fine-tuning and pre-training at competitive prices. Sacra’s May 2026 analysis pegs its annualized revenue at $1 billion as of February 2026 — a 400% increase from about $618 million at the end of 2025. Valuation stands at $3.3 billion following a $305 million Series B in early 2025. The startup has raised $533.5 million total and was reportedly in talks for a $1 billion round at a $7.5 billion pre-money valuation by March 2026. Sacra Together AI report
Its edge comes from specialized kernels, FlashAttention optimizations and a focus on 100-plus popular open models ranging from Llama variants to Mistral. Developers praise sub-second latency on coding agents and cost savings that can reach 60% versus some closed alternatives. The platform runs thousands of GPUs across its own data centers and partner capacity.
Integration with Hugging Face tightens the loop. In September 2025 the two announced that any compatible LLM on the Hub could be fine-tuned directly on Together AI’s infrastructure. Users gain one-click access without managing their own clusters. The partnership extends to inference endpoints where Together powers many of the fastest options listed on Hugging Face model pages.
Recent benchmarks highlight the advantage. Together’s inference for coding agents delivered 31% more tokens per second than TensorRT-LLM in some tests, with better time-to-first-token and significantly lower cost than certain proprietary models. Such performance draws startups and enterprises seeking to productionize open systems without massive infrastructure bills.
And the momentum shows no sign of slowing. On X, Hugging Face product lead Victor Mustar highlighted in June 2026 how open models like GLM-5.2 were offered free for limited periods across multiple providers including Together AI. Developers responded by integrating them into coding agents and seeing performance that rivals or exceeds closed options in specific tasks.
This flywheel — open models draw users to the Hub, the Hub routes paid inference and fine-tuning workloads, revenue funds further platform improvements — explains why both companies have scaled so quickly. It also explains why more than 30% of the largest U.S. corporations now maintain presence on Hugging Face. They want optionality. They want to avoid lock-in. They want to experiment cheaply before committing budgets.
China’s rising share of downloads adds another dimension. The country surpassed the U.S. in monthly Hugging Face downloads in recent periods. That shift raises questions around AI sovereignty, supply-chain resilience and the role of open systems in global competition. Independent developers and small teams increasingly steer meaningful portions of model innovation, a trend the Spring 2026 report describes as a move away from industry-dominated development.
Challenges remain. Quality varies across the long tail of community uploads. Bias mitigation tools still lag in some areas. Misuse risks grow as models become easier to deploy. Yet the data suggests organizations pay for governance, support and speed on top of the free base layer. Hugging Face’s move toward more predictable recurring revenue — API fees, referrals, enterprise plans — reflects that reality.
Together AI similarly balances open accessibility with enterprise-grade features. Its recent ISO 27001 certification targets production workloads at larger customers. The company continues to expand physical GPU capacity with new facilities in Maryland and planned sites elsewhere.
Look at the numbers again. Thirteen million users. Two million models. A billion dollars in run-rate revenue at one compute provider. These figures would have seemed fanciful a few years ago. They now describe the present.
The open-source wave hasn’t just boosted Together AI and Hugging Face. It has redefined where value accrues in AI development. The platforms that host the models, simplify their use and provide paid acceleration layers stand to capture a disproportionate share. Both companies have positioned themselves at exactly those intersections.
Future growth will test their ability to maintain community goodwill while scaling commercial operations. It will also test whether the cost advantages and customization freedom of open models continue to pull share from closed frontiers. Early evidence says yes. The numbers say the bet is already paying off handsomely.
WebProNews is an iEntry Publication