GPU cloud operator Together reportedly in search of $100M+ in latest funding

Together Computer Inc., one among several startups operating cloud platforms geared toward artificial intelligence workloads, is predicted to lift a fresh round of funding.

The Information late Wednesday cited sources as saying that the corporate is in search of greater than $100 million. Salesforce Inc.’s enterprise capital arm, which has emerged as a significant generative AI investor in recent quarters, is predicted to steer the round. It’s believed that Coatue will participate as well.

The Information’s sources indicate that the round is poised to value San Francisco-based Together at over $1 billion, or about double what it was price after its previous raise in November. That investment included contributions from Nvidia Corp. and several other other institutional investors. The report cautioned that the terms of the corporate’s latest raise could change before the transaction closes.

Together’s cloud platform provides access to top-end H100 and A100 graphics processing units from Nvidia. Customers can deploy those GPUs in clusters of between 16 and a couple of,048 chips. The platform spares firms the effort of constructing AI infrastructure from scratch, a fancy undertaking made especially difficult by the incontrovertible fact that high-end GPUs are currently in brief supply.

Alongside hardware, Together offers several software features designed to ease the duty of coaching large language models. Most notably, the corporate’s cloud is supplied with an open-source performance optimization tool called FlashAttention-2.

GPU clusters include several kinds of memory that AI models can use to store the info they process. The fastest memory type, SRAM, is built directly into the graphics cards and is out there in just limited quantities. FlashAttention-2 allows AI models to leverage SRAM more extensively than would otherwise be possible and thereby reduce their use of a GPU’s cluster slower memory pools, which increases performance.

Together’s customers even have access to a training dataset called RedPajama-Data-v2. The corporate, which developed the dataset in-house, says that it includes greater than 100 billion documents written in five languages. Together claims customers can easily filter documents that don’t align with an AI project’s requirements, which will likely be a highly time-consuming process.

Once a neural network exits the event phase, it’s often made available to workloads via an application programming interface. Together says its cloud lends itself to that use case as well. The platform monitors the quantity of API requests sent to AI models running on its platform and mechanically adds capability when needed.

The funding round Together is reportedly raising could make it easier for the corporate to finance product development efforts. As of last November, a single H100 graphics card reportedly retailed for between $25,000 and $40,000. Alongside GPUs, a cloud provider in search of to expand its AI infrastructure must also buy a big amount of auxiliary hardware that further drives up procurement costs.

Together’s competitors have also ramped up their fundraising efforts recently. Lambda Inc., one other startup with an AI-optimized cloud platform, today announced that it has closed a $320 million round at a $1.5 billion valuation. Last yr, CoreWeave Inc. raised $2.3 billion in debt financing from Coatue, one among the investors expected to back Together’s latest round, and a half-dozen other institutional backers. 

Photo: Unsplash