SpaceX and Groq Deals Show Frontier AI Compute Is Now Sold by the Month

Two deals announced on June 22 show how quickly the market for AI compute is changing shape. SpaceX agreed to rent Nvidia GB300 capacity at its Colossus 2 data center to Reflection AI, an open-weight model lab, in a contract worth up to $6.3 billion, TechCrunch reported. The same day, chipmaker Groq confirmed a $650 million raise to rebuild itself as a cloud provider after Nvidia licensed its technology and hired its founder, according to TechCrunch. Together they point to a market where frontier-class compute is increasingly sold by the month, and where the line between AI lab, chipmaker and landlord is blurring. That matters to enterprises deciding where their models run and who they depend on.

What did SpaceX and Reflection AI agree?

Reflection will pay $150 million a month starting July 1 through 2029 for immediate access to GB300 chips and supporting hardware near Memphis, Tennessee. After the first three months, either side can end the deal with 90 days' notice.

Colossus was originally built by xAI for its own model work. xAI is now part of SpaceX, and as its internal efforts struggled, SpaceX began renting capacity to other labs, TechCrunch said. Reflection is the third customer. TechCrunch put SpaceX's earlier deals at $1.25 billion a month with Anthropic and $920 million a month with Google, both running through July 2029.

Reflection, founded in 2024 by two former Google DeepMind researchers, is using its first compute deal to promote an open-weight alternative to closed labs. "Recent events highlight how important open source is to the AI ecosystem," a spokesperson told TechCrunch. The publication noted that open-weight models have drawn more interest since U.S. government action against Anthropic's Fable and Mythos models.

Why is Groq turning into a neocloud?

Groq's story runs in the opposite direction, from chip designer to service provider. About six months before the raise, Nvidia took a non-exclusive license to Groq's technology and hired founder and CEO Jonathan Ross, president Sunny Madra and other staff. Nvidia announced its own LPU-based system at its GTC conference in March.

What remained of Groq has shifted to its cloud business. Co-founder Doug Wightman is now CEO, and the company hired a new COO, CTO and chief product officer. Groq says it serves more than five million developers and thousands of AI companies from 13 data centers across North America, Europe, the Middle East and Asia-Pacific. The round was led by Disruptive, whose founder Alex Davis is Groq's chairman. The company did not disclose a new valuation.

Is renting frontier compute a sign of strength or weakness?

There are two readings, and both have evidence behind them.

The bullish view is that the market is maturing. Compute that once sat inside a single lab is now available to others, including smaller open-weight developers that could never build a Colossus themselves. More suppliers, from SpaceX to neoclouds like Groq, should mean more choice and better pricing for everyone downstream, including enterprises that consume those models through APIs or host them privately.

The cautious view is that the arrangements are less durable than their headline values suggest. Elon Musk has publicly downplayed the three-year terms, saying the contracts can be canceled at any time, TechCrunch reported. Reflection's own deal has a 90-day exit after three months. Groq is rebuilding its leadership team and its business at the same time. And SpaceX began renting capacity because its own AI ambitions struggled, not because it planned to be a landlord.

For enterprises, that tension is the point. A model provider's capacity can depend on contracts that either side may exit on short notice. A $6.3 billion headline is a ceiling, not a guarantee.

What does this mean for enterprise model strategy?

The week's deals strengthen the case for keeping options open at the model layer.

Open-weight models are becoming a continuity hedge. Reflection's pitch is that open models give buyers control. For enterprises, the practical benefit is portability: an open-weight model can run on several clouds or on-premises if one provider's access changes. That benefit is real even if an enterprise never self-hosts.

Self-hosting is still expensive. AI Connect speakers have urged caution here. "Hosting your own models is a really big investment," Jonathan Lounsbury, Principal Engineer at Alteryx, said in a AI Connect expert panel webinar, pointing to developer time, cost and scarce GPUs. His advice was to prove use cases with major providers first. The full discussion on self-hosted LLMs and lock-in lays out the tradeoffs.

Right-sizing reduces exposure. Mistral AI's Aaron Delp, Global Head of AI Technical Marketing, argued during a AI Connect webinar that the best return comes from "the right model at the right time in the right place." Smaller, task-specific models are easier to move between providers and less sensitive to frontier capacity crunches. AI Connect covered his approach in building enterprise agents with right-sized models.

What should enterprise leaders do now?

• Ask where your models run. For each critical AI service, find out which data centers and capacity contracts sit behind it, and what happens if those change.
• Test at least one open-weight fallback. Even a limited proof of concept shows how hard a switch would be and what quality you would give up.
• Read the exit clauses. Short notice periods upstream can turn into service disruptions downstream. Match your own contract terms and contingency plans to that reality.
• Watch neocloud pricing. New entrants such as Groq need customers. That can create negotiating room for enterprises with steady inference workloads, as AI Connect's coverage of AI factories and enterprise infrastructure has noted in the context of utilization and cost.

Reflection's payments begin July 1. Whether SpaceX adds more tenants, and whether Groq's rebuilt leadership can win enterprise workloads, will show how stable this new rental market turns out to be.

Frequently asked questions

What is the SpaceX and Reflection AI deal?

Reflection AI will pay SpaceX $150 million a month from July 1, 2026 through 2029 for Nvidia GB300 capacity at Colossus 2 near Memphis, worth up to $6.3 billion.

Why did Groq raise $650 million?

After Nvidia licensed Groq's technology and hired its founder, Groq refocused on its cloud business and raised $650 million led by Disruptive.

What should enterprises take from these deals?

Model providers' capacity can rest on contracts with short exit terms, so enterprises should know where critical models run and test fallback options.

Filed under: Data & AI Infrastructure, AI Strategy & ROI