Superpower Daily
University of Konstanz Finds AI Agents Coordinate Up to 1,000,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...Nvidia Takes a Minority Stake in Cloverleaf to Push AI Data-Cen...Watermarks Remover Tops 14,000 Stars; Claude’s Detector Isn’t P...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...University of Konstanz Finds AI Agents Coordinate Up to 1,000,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...Nvidia Takes a Minority Stake in Cloverleaf to Push AI Data-Cen...Watermarks Remover Tops 14,000 Stars; Claude’s Detector Isn’t P...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...
Businesspublished

Together AI Secures $240 Million IBM Cloud Buildout for Open-Source Inference

The agreement gives Together AI a planned dedicated IBM Cloud system for serving enterprise open-source models. The commercial test will be whether the 2027 deployment delivers better service performance and token economics at its reported scale.

By 3 min read
Together AI Secures $240 Million IBM Cloud Buildout for Open-Source Inference
Sign in to save and react

Listen to this story

The audio brief

About 1:28
0:001:28
Read transcript
Together AI and IBM have committed 240 million dollars to build a dedicated inference cluster on IBM Cloud, but the capacity will not be available until the first quarter of 2027. The project is designed to serve open-source models to enterprise customers, with Together AI running its inference software on the new system. The hardware combines Nvidia HGX B300 systems with Nvidia’s Spectrum-X Ethernet networking. In practical terms, this is a tightly integrated setup: computing, networking, and model-serving software aimed at handling real-time workloads at scale. Together AI says its platform already processes roughly 400 trillion tokens a month, which helps explain the size of the commitment. The stated goal is better inference performance and improved token economics—essentially, serving more model output efficiently—while expanding access to open-source models for businesses. But this is a planned infrastructure buildout, not a live increase in customer capacity. Nvidia says the underlying platform is designed to deliver up to 30 times more AI factory output than prior generations. That figure describes the platform’s design, not measured results from this specific IBM Cloud deployment. The key question is therefore deferred to delivery: when the cluster is expected to come online in early 2027, will it actually improve service performance and operating economics at Together AI’s reported scale?

Story brief

3 key points

Together AI’s $240 million IBM Cloud agreement is a capacity bet for enterprise open-source inference, not a currently available service. The planned multi-year deployment will use Nvidia HGX B300 systems and Spectrum-X networking, with availability targeted for Q1 2027. Together AI says its platform already processes about 400 trillion tokens monthly, giving the project a substantial demand rationale. For buyers...

  1. 01

    Expected Q1 2027 availability means the project is not yet a live capacity increase for Together AI customers.

  2. 02

    The hardware stack combines Nvidia HGX B300 systems with Spectrum-X Ethernet networking on IBM Cloud.

  3. 03

    Nvidia’s “up to 30×” figure describes platform design, not measured performance from this deployment.

Together AI is lining up a dedicated, large-scale IBM Cloud cluster to serve open-source models to enterprise customers. IBM and Together AI have put $240 million behind that planned buildout, seeking infrastructure for a service Together AI says is handling roughly 400 trillion tokens a month.

The companies signed a multi-year agreement to deploy the inference cluster on IBM Cloud. IBM says it will be its first dedicated, large-scale inference cluster built with Nvidia HGX B300 systems, and the deployment is expected to become available in the first quarter of 2027. That makes the announcement a committed infrastructure project rather than a live service today.

The stack is designed for serving models at scale

The cluster will pair Nvidia HGX B300 systems with Nvidia Spectrum-X Ethernet networking. IBM Cloud supplies the cloud environment, while Together AI plans to run its inference software on the combined system. The design brings the compute hardware, network and serving software into one deployment for Together AI’s enterprise open-source model service.

Together AI’s stated objective is to improve inference performance and token economics while providing open-source model inference to enterprise customers. It selected IBM and Nvidia based on their product roadmaps and their ability to supply GPU capacity at the scale required for expanding workloads. The deployment is also intended to support real-time AI services.

Nvidia’s performance figure is not a result from this cluster

Nvidia says the underlying infrastructure is designed to deliver up to 30 times more AI factory output than prior generations. That is a platform-level design claim, not a measured outcome from the announced IBM Cloud deployment. Together AI’s proposed gains in service performance and token economics are likewise goals for a system that has not yet reached its expected availability date.

A deeper IBM-Nvidia infrastructure relationship

For IBM, the agreement expands a broader relationship with Nvidia across AI infrastructure and software. Together AI says its AI Native Cloud spans inference, training, fine-tuning and agentic workflows, but this project is specifically aimed at inference: running open-source models for enterprise customers rather than adding a general-purpose platform announcement.

The next milestone is delivery

The immediate result is a dedicated infrastructure plan for Together AI’s enterprise inference service. The meaningful next test comes with the expected Q1 2027 availability, when the companies can show whether the B300-based system delivers the operating improvements they have promised.

Editorial analysis

Our Read

This is a commitment to control a larger slice of the infrastructure beneath Together AI’s service, not yet evidence that customers will receive a faster or cheaper offering. Together AI’s stated volume of roughly 400 trillion tokens a month makes the delivery outcome consequential: small operational changes can matter at that scale. The next useful evidence will be the Q1 2027 launch, followed by customer-facing measures of performance and token economics. The deal also belongs to a period of sizable multiyear compute commitments, including Hudson River Trading’s agreement with CoreWeave.

Related coverageHudson River Trading Makes a Multibillion-Dollar Bet on CoreWeave Compute

Sources

  1. pulse2.comIBM And Together AI Sign $240 Million Multi-Year Deal For Large-Scale NVIDIA B300 AI Inference Cluster

Related coverage

Continue reading