Summary
- Together AI announced it is teaming up with IBM and NVIDIA to build enterprise-grade AI inference infrastructure on IBM Cloud.
- The setup combines a dedicated NVIDIA B300 cluster with Spectrum-X networking, reportedly the first such configuration attempted on IBM Cloud.
- Following its $800 million Series C raise last month, Together AI has been rolling out partnerships in quick succession, accelerating its push to expand in the inference infrastructure market.
The First B300 Cluster on IBM Cloud
Together AI announced it is building new AI inference infrastructure on IBM Cloud together with IBM and NVIDIA. At the core is a dedicated cluster built on NVIDIA's latest GPU, the B300. This is paired with NVIDIA's high-speed networking technology, Spectrum-X — a combination Together AI says is being tried on IBM Cloud for the first time. Together AI described the offering as "enterprise-grade AI inference."
How the Three Companies Divide the Work
The structure is straightforward. NVIDIA supplies the hardware (B300 GPUs and Spectrum-X networking), IBM hosts it on its cloud and sells it to enterprise customers, and Together AI handles the inference software layer that actually runs the models on top. Together AI is reportedly planning to operate what it calls a "production-ready inference platform" on this cluster.
This three-way structure goes beyond a cloud provider simply renting out GPUs — it bundles in the software that actually serves the models. For enterprise customers, this means they no longer need to separately contract for GPU capacity and inference optimization; both can be handled at once.
Why Together AI Is Rolling Out Partnerships Now
On July 1, Together AI announced an $800 million Series C funding round, citing the accelerating shift toward open-source AI as its rationale. At the same time, it announced a dedicated GPU cluster partnership with Y Combinator, on-demand availability of B200, and serving support for MiniMax-M3. Its GPU lineup now spans a wide range, including GB300, GB200, B200, H200, and H100.
The momentum has continued since. On August 6, coding agent tool Roomote announced it had adopted Together AI as its inference provider. On August 11 — the same day as this IBM/NVIDIA partnership — Together AI unveiled "voice finder," a tool for searching and matching among more than 600 voices from its own TTS model. The following day, August 12, it began offering fine-tuning support for the DeepSeek V4 Flash 0731 model. Funding, major cloud partnerships, developer tools, and new model support have all landed within a single month.
| Date | Announcement |
|---|---|
| 2026-07-01 | $800M Series C raised, dedicated YC cluster, B200 on-demand |
| 2026-08-06 | Adopted by Roomote as inference provider |
| 2026-08-11 | voice finder launched (search across 600+ voices) |
| 2026-08-11 | Partnership with IBM and NVIDIA on IBM Cloud, B300 cluster |
| 2026-08-12 | Fine-tuning support begins for DeepSeek V4 Flash 0731 |
What Are B300 and Spectrum-X
B300 is NVIDIA's latest-generation data center GPU, used for inference workloads running large language models. Spectrum-X is an Ethernet-based networking platform built by NVIDIA that ties multiple GPUs together into a single cluster, reducing the bottlenecks that occur when data is exchanged between them. Industry observers note that inference speed often fails to scale linearly with GPU count due to such bottlenecks, and combining the two technologies can help reduce that loss.
What Changes as a Result
For enterprise customers already using IBM Cloud, this creates a path to plug into Together AI's inference stack directly, without needing a separate GPU cloud contract. In an inference infrastructure market where competition is increasingly centered on who can serve open-source models faster and more reliably, Together AI appears to be expanding its footprint through alliances with major cloud providers right after closing its funding round. The fact that a cloud provider like IBM, with its long-established enterprise customer base, is entering into this kind of partnership suggests that adoption of open-weight models is spreading beyond startups and into the more conservative enterprise market.
Comments