Two thousand Nvidia Blackwell chips are already spoken for, and IBM has not finished building the room to hold them. On August 11, IBM and Together AI signed a multi-year, $240 million agreement to deploy a dedicated inference cluster on IBM Cloud, built on Nvidia HGX B300 systems paired with Spectrum-X Ethernet networking (IBM, 2026). IBM calls it the first large-scale cluster built specifically for inference on IBM Cloud using that hardware. IBM does not expect the cluster to go live before the first quarter of 2027.
Equity Funds Growth. A Cluster Lease Funds Capacity.
Together AI's chief revenue officer, Kai Mak, told Reuters the capacity will sell out two to three months before the cluster opens, with full offtake committed ahead of service (BNN Bloomberg, 2026). Together AI raised an $800 million Series C round in July at an $8.3 billion valuation and earmarked that capital to expand what it calls its AI Native Cloud (IBM, 2026). Alongside that raise, it signed a separate $240 million, multi-year commitment with IBM for dedicated capacity.
One deal buys headcount and product development. The other buys a guaranteed spot on a specific chip.
Inference Costs Scale With Every Request, Not With a Funding Round
Together AI reports serving 400 trillion tokens monthly through its inference product, and that volume is what is driving the IBM deal (IBM, 2026). Training a model needs a burst of compute once. Serving it needs compute every time a customer sends a request, indefinitely, at a cost that scales with adoption rather than tapering off after launch. IBM secured 2,000 Blackwell B300 units for the cluster, and that queue position is what the lease buys.
IBM Is Building Trust Twice, at Two Different Layers
This is IBM's second nine-figure bet on open-source AI trust infrastructure this year. In May, IBM and Red Hat committed $5 billion to Project Lightwell, a clearinghouse that signs and certifies open-source software components for regulated enterprises (Bellamkonda, 2026). That deal secured the code. This one secures the compute the code runs on. Read together, the pattern positions IBM as the trust layer open-source AI needs at both ends, supply chain and data center, while AWS, Microsoft Azure, and Google Cloud compete mostly on price and proprietary model access.
Alan Peacock, general manager of IBM Cloud, framed the deal around the race to adopt agentic AI at scale (IBM, 2026). Dion Harris, Nvidia's senior director for HPC and AI infrastructure solutions, went further, comparing AI factories to electricity and telecommunications, infrastructure enterprises expect to exist rather than build themselves (IBM, 2026).
A funding round buys equipment, not a place in line for the next chip.
Two IBM commitments in three months, one for code and one for compute, add up to a single claim: enterprises adopting open-source AI need a trusted intermediary at every layer, and IBM wants to be that intermediary before AWS, Azure, or a neocloud gets there first.
IBM Cloud Sits Underneath the Inference Layer Enterprises Choose
Enterprises evaluating Together AI for open-source model inference are also choosing IBM Cloud as the underlying data center. The lease runs for multiple years, and capacity planning at this scale means IBM allocates Blackwell supply across customers. That allocation decides how much capacity a given workload gets during periods of high demand.
Ask your inference vendor what happens to your workload when their underlying cloud lease comes up for renewal, and get that answer in writing before you build on top of it.
IBM. "IBM and Together AI Sign Multi-Year Agreement to Scale Open-Source AI Inference with NVIDIA AI Infrastructure on IBM Cloud." IBM Newsroom, 11 Aug. 2026, newsroom.ibm.com.
Reuters. "IBM, Together AI Ink US$240 Million Deal for Nvidia-Powered AI Inference Cluster." BNN Bloomberg, 11 Aug. 2026, www.bnnbloomberg.ca.
Bellamkonda, Shashi. "IBM and Red Hat's $5B Project Lightwell Reveals a Patch Velocity Crisis." shashi.co, 28 May 2026, www.shashi.co.
Disclaimer: This blog reflects my personal views only. Content does not represent the views of my employer, Info-Tech Research Group.
