IBM Orders $240M NVIDIA B300 Cluster as AI Infrastructure Financing Hits $500B

IBM Bets $240M on Open-Source Inference

IBM signed a multi-year, $240 million agreement with Together AI to build its first large-scale dedicated inference cluster on IBM Cloud using NVIDIA HGX B300 systems and Spectrum-X Ethernet networking, with availability scheduled for Q1 2027. The cluster will run open-source model inference at scale, operated by Together AI on IBM's infrastructure.

This matters because it gives regulated enterprises a path to run Llama, Qwen, and other open models on a big-blue cloud with familiar contracts and support, without adding a second hyperscaler or relying on proprietary foundation models. The NVIDIA HGX B300 hardware indicates top-tier performance and premium pricing—expect IBM's AI infrastructure SKUs to sit at the high end of GPU hourly rates, but with enterprise support and compliance built in.

For CIOs who have been hesitant to run open models on bare-metal or smaller GPU clouds, this is a managed, high-capacity option with data residency controls and regulated industry positioning. The $240 million scale signals vendor viability and may push IBM to compete aggressively on reserved capacity pricing against AWS Bedrock, Azure OpenAI, and Google Cloud's TPU stacks.

NVIDIA Unlocks $500B in Third-Party Capital for GPU Buildout

NVIDIA announced strategic partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to create AI compute infrastructure financing platforms targeting over $500 billion of third-party capital over time. The structure is designed to fund AI data center buildouts across NVIDIA's ecosystem—frontier AI labs, enterprises, and GPU cloud providers—and NVIDIA plans to deliver 2 million additional GPUs as part of these initiatives.

This directly addresses the capital cost and financing friction that has kept data-center operators and cloud providers from scaling NVIDIA-based GPU clusters. By solving capex hurdles with cheap institutional capital, NVIDIA makes it easier for hosters, telcos, and second-tier GPU clouds to standardize on NVIDIA GPUs instead of switching to AMD Instinct, Intel Gaudi, or cloud-native silicon like AWS Trainium or Google TPU.

For enterprise buyers, this should translate to more regional GPU capacity and more vendors offering NVIDIA clusters over a 2-to-5-year horizon. It does not guarantee lower prices in the near term, but it improves the odds that capacity constraints—not financing—stop being the main bottleneck. Expect continued premium pricing on top-end GPUs, but more competitive pricing on H100, H200, and eventually B100 as the 2 million additional GPUs are deployed. The trade-off: this deepens global dependency on NVIDIA, so enterprises wanting multi-vendor resilience should balance with strategies involving AMD, Intel, or ARM-based accelerators, even if these remain secondary.

Oracle Ships Nemotron 3.5 Lightning and H100 Multi-Node Serving

Oracle published its August 2026 AI update detailing new capabilities in OCI Enterprise AI. NVIDIA Nemotron 3.5 Lightning is available day-zero in OCI Enterprise AI, making Oracle one of the first cloud providers to offer NVIDIA's latest customizable open model for always-on AI agents. OCI Enterprise AI now supports H100 multi-node serving for imported models, enabling deployment of large models across multiple H100 GPU nodes, and added additional hardware unit shapes for Cohere and Meta models.

This gives Oracle customers who already run workloads on OCI a path to deploy the latest open models without waiting for Azure or AWS to certify them. The H100 multi-node serving capability is particularly relevant for enterprises running fine-tuned Llama or Mistral models that exceed single-node memory, as it allows horizontal scaling without refactoring. The additional hardware shapes for Cohere and Meta models let buyers optimize cost versus performance by choosing smaller GPU configurations for lower-concurrency workloads.

What to Watch

The IBM-Together AI cluster lands in Q1 2027—watch for pricing announcements in late 2026 and compare reserved capacity rates against AWS Bedrock and Azure OpenAI for open models. NVIDIA's $500 billion financing structure will take years to deploy, but enterprises should track which second-tier GPU cloud providers announce expansions in 2026 and whether they pass financing savings to customers. Oracle's day-zero support for Nemotron 3.5 Lightning sets a precedent—monitor whether Azure and AWS match this speed for future NVIDIA releases or if Oracle maintains a time-to-market advantage for open models on OCI.