IBM Strikes $240M Deal with $8.3B Startup Together AI for Massive Inference Infrastructure

IBM Partners with Together AI on $240M Inference Deal

IBM and San Francisco-based startup Together AI have entered a $240 million, multi-year agreement to build a large-scale artificial intelligence infrastructure within IBM’s cloud, Reuters reported on Tuesday. The deal is centered on inference—the process by which already-trained AI models respond to user requests—a segment of the AI stack where demand is exploding.

The infrastructure will be built around Nvidia’s latest technology, specifically the HGX B300 platform using the new Blackwell processors, combined with Nvidia’s Spectrum-X networking. Together AI’s platform already allows enterprises to run and fine-tune open-source models from developers like DeepSeek, MiniMax, and Kimi, and this partnership will bring that capability directly onto IBM’s cloud, with significant scale.

Together AI was valued at $8.3 billion in a July funding round, underlining the market’s appetite for specialist inference providers. For IBM, the deal injects a high-profile AI-native partner into its hybrid cloud portfolio at a time when it is repositioning its watsonx platform and broader cloud strategy to compete with hyperscalers.

The partnership spans multiple years, suggesting a strategic commitment rather than a one-off capacity purchase. It reflects the broader industry shift where cloud providers are no longer just renting raw compute but are integrating specialized AI platforms to attract developers and enterprises focused on deploying open-source models at scale.

Behind the IBM–Together AI Alliance

Why Inference Infrastructure Is Suddenly the Hot Market

Until recently, the AI infrastructure spending frenzy focused on training large models. Now, as models move into production, the bottleneck is inference—the real-time generation of answers to user queries. Every chatbot interaction, code assistant, or enterprise AI application consumes inference compute. Demand is growing faster than training capacity, and Together AI has positioned itself as a pure-play inference cloud for open-source models, a niche that hyperscalers are racing to fill. IBM’s deal gives it a ready-made, optimized stack to offer customers looking to deploy models like DeepSeek or Llama variants without managing the underlying infrastructure.

Where IBM Fits in the AI Cloud Race

IBM has struggled to keep pace with AWS, Azure, and Google Cloud in the public cloud market, but its strength lies in hybrid cloud and enterprise workloads. By embedding Together AI’s platform—specifically tuned for open-source inference—IBM can offer a differentiated product to enterprises that are wary of vendor lock-in or want to run AI on-premises or in a regulated environment. The Nvidia Blackwell hardware partnership also signals that IBM is willing to invest in cutting-edge silicon to stay competitive. However, the $240 million figure, while substantial, is modest compared to the tens of billions hyperscalers are pouring into AI infrastructure annually, suggesting this is a targeted play rather than an attempt to outspend the leaders.

Together AI’s Strategic Leap

For a startup that was valued at $8.3 billion just weeks ago, locking in a multi-year, quarter-billion-dollar deal with an established enterprise giant like IBM provides a major credibility boost and guaranteed revenue. It moves Together AI from a niche cloud provider to a partner capable of landing global enterprise clients via IBM’s sales channel. The risk is that IBM’s cloud may not attract the same volume of AI-native developers as hyperscalers, and Together AI could become overly dependent on a single distribution deal. Still, the startup’s focus on open-source inference aligns with a clear market trend away from black-box proprietary models, potentially giving it a long runway.

What This Means for Cloud AI Strategies

For cloud and AI decision-makers at enterprises

• If your organization is already using IBM Cloud or evaluating open-source models, the IBM–Together AI stack could simplify deployment: it offers HGX B300-powered inference with Spectrum-X networking, optimized for models like DeepSeek and Kimi. Ask for a trial workload to compare performance against self-managed GPU clusters.
• The deal signals IBM’s seriousness about becoming a competitive inference partner. Enterprises should pressure IBM for concrete SLAs on latency and throughput, as inference workloads are latency-sensitive. Pricing models are not yet public; negotiate based on tokens per second, not just GPU hours.
• For startups and AI-native companies, the partnership validates the market for dedicated inference clouds. However, vendor lock-in is a real risk. If you adopt this platform, ensure your model formats and deployment scripts remain portable to other Nvidia-based clouds.

For investors and competitors

• Together AI’s $240M contract win suggests it is winning enterprise trust beyond its initial developer user base. Competitors like CoreWeave or Lambda Labs should note that pairing with a legacy enterprise cloud can shift the sales motion from self-service to long-term enterprise contracts.
• IBM’s deal does not automatically change its position in AI cloud rankings, but it gives the company a tangible, branded offering to pitch at upcoming earnings calls. Watch for any mention of Together AI in IBM’s next quarterly release as a gauge of deal momentum.