Higgsfield puts its video models on Together AI after its $400M Series B

Why it matters

Higgsfield's funding is flowing directly into compute. Its use of Together AI alongside Nebius, Nvidia and OpenAI shows how AI application companies are using multiple suppliers to control capacity and inference costs.

Video version

Higgsfield co-founder and CEO Alex Mashrabov (@alexmashrabov) has put the AI video platform's models on Together AI, adding another infrastructure supplier two days after Higgsfield closed a $400 million Series B.

Together AI disclosed the customer relationship on Wednesday, saying Higgsfield's video models run on its cloud. The post identifies Higgsfield as a production customer but does not name the models, workloads or financial terms covered by the agreement.

Mashrabov built Higgsfield with CTO Yerzat Dulat (@codentropy) and chief strategy officer Mahi de Silva. Before Higgsfield, Mashrabov led generative AI at Snap and co-founded AI Factory, which Snap acquired for $166 million in 2020, according to TechCrunch. That background explains Higgsfield's initial focus: turning model output into repeatable camera movements, advertising formats and social video rather than selling access to a raw generation model.

A customer win tied to an expensive workload

Higgsfield's Cinema Studio combines image and video generation with controls for camera movement, lenses, lighting, storyboards and multi-shot sequences. Higgsfield also routes work across proprietary and third-party models. An OpenAI customer profile published in January described Higgsfield using GPT-4.1 and GPT-5 for planning and Sora 2 for video generation.

That orchestration creates a heavy infrastructure requirement. Video generation involves processing sequences of high-resolution frames, making small improvements in GPU utilization, model serving and generation time financially significant at production volume.

Together AI sells dedicated container inference specifically for video, audio and image models, alongside GPU clusters, model training and serverless inference. The Higgsfield deployment gives Together AI a high-volume generative media customer weeks after Together AI raised an $800 million Series C on July 1st. Together AI also said it had secured commitments for above 500 megawatts of compute capacity, financed separately from the equity round.

The pairing illustrates where a portion of the application-layer AI funding boom goes next. Investors fund companies such as Higgsfield to increase generation volume and build enterprise products; infrastructure providers raise their own rounds to acquire the GPUs and power capacity needed to serve that demand.

Higgsfield is spreading workloads across suppliers

Together AI joins an infrastructure stack that has already included Nebius, Nvidia and OpenAI. An Nvidia case study says Higgsfield used Nvidia Blackwell systems on Nebius and other Nvidia cloud partners for training and inference. Nvidia said the move reduced training time by 30%, a vendor-supplied performance figure tied to Higgsfield's migration to newer hardware.

The existing relationships make the Together AI announcement an expansion of Higgsfield's supplier base rather than evidence of a wholesale migration. For Mashrabov and Dulat, that approach limits dependence on one cloud provider and creates room to route different models or workloads according to capacity, performance and cost.

That flexibility matters because Higgsfield's core product already behaves like a routing and workflow layer. Higgsfield's current product materials describe a system that plans camera motion, pacing and visual emphasis before selecting generation tools. Customers buy the finished workflow while Higgsfield absorbs the complexity and cost of the underlying model providers.

The $400M round is buying capacity

Higgsfield announced its Series B on August 17th at a $5.4 billion post-money valuation, up from the $1.3 billion valuation attached to its Series A extension in January. DST Global led the financing. Participants included Tribe Capital, Goldman Sachs Alternatives, Smash Capital, Fifth Wall, Valor Capital, Intel Capital, Liberty Global Tech Ventures, Mirae Asset Capital and NTT Docomo Ventures. Existing backers Accel, Menlo Ventures, AI Capital Partners, GFT Ventures, Capra Ventures, BAM Corner Point and BroadLight Capital also invested, according to Axios.

Higgsfield says annualized revenue reached $700 million in August, with above 30 million users and work produced for 390 Fortune 500 companies. Those are self-reported figures. Annualized revenue extrapolates the current pace over a full year and does not establish recognized revenue, margins or the cost of serving each generation.

Higgsfield said the Series B would finance infrastructure, research, hiring and international sales. The Together AI agreement puts a named supplier behind the first item on that list. Higgsfield's ability to hold its reported revenue pace will depend partly on whether Mashrabov and Dulat can push generation costs down as usage rises. The $400 million round gives them considerable room to negotiate for capacity. It also gives infrastructure vendors a well-funded customer with every incentive to keep the GPUs busy.