CoreWeave’s attempt to acquire Marimo and Core Scientific got shut down when the shareholders at Core Scientific balked at a bid rumored to have been close to a $9 billion all-stock acquisition. “Today Core Scientific and all its equipment, including the SJ01, will always remain owned by Core Scientific,” said Christie Weyant, who has since left her role at Core Scientific and is now head of Developer Advocacy at Marimo. In industry minutes, CoreWeave turned around and managed to acquire the open-source Python notebook project Marimo, signaling a shift to valuables which include a development tool even as megadeals fall apart due to high valuations. Ultimately, growth stage establishes the right valuation metric and seated power value.
The Core Scientific deal’s collapse highlights a fundamental truth: all-stock acquisitions are challenging to execute if the buyer’s stock is lifted by AI enthusiasm, and the target’s valuation ascends more rapidly. CoreWeave serves as a primary illustration. Originally a crypto miner, a pivot into AI compute firm with tight Nvidia integration developed, and CoreWeave’s market capitalization has risen from roughly 14 billion dollars at IPO to 66 billion dollars or more today, driven by an investor desire for scarce GPU capacity.
The lack of availability is inherent. Power and permitting are limited; transformer waiting time is lengthy, and the best accelerators are restricted to the amount provided. Major bank figures, such as those released by Goldman Sachs, predict AI-linked capex may climb to 200 billion dollars annually soon, resulting in data center firms, utilities, and chipmakers in a cost-free struggle to expand capacity. As a consequence, Core Scientific ops – huge power consumption, expandable sites, experience in high-density deployments – are becoming a strategic asset in their own right.
The other curve: crypto-native power producers have quietly created expertise in immersion cooling, rapid construction, and price-efficient power sourcing. These abilities may be utilized in AI hosting with partial retraining; if it turns viable, this is why investors believe AI landlords switched to miners can close a gap in valuation with GPU clouds playing pure.
From Racks To Notebooks CoreWeave Buys Marimo
With the Core Scientific transaction declined, CoreWeave opted to reinforce its software story by acquiring Marimo, an open-source counterpart to Jupyter Notebook, for a disclosed sum. According to PitchBook, Marimo has raised around 5 million dollars, a small amount for an infrastructure purchase but excellent leverage for developer engagements.
Python notebooks drive AI work in two ways: first, they prototype models, stitch data pipelines, and package demos that scale into production apps, which presents the purchase funnel for public cloud or “AI-first” services; and second, through the same notebook buyer, a provider gains a channel to direct a core function of its business – in CoreWeave’s case, its GPU cloud – rather than serving as a lead gen expense line until work gets steered away toward it. In both plotlines, CoreWeave fails to capture the last mile of its users’ value creation: For CoreWeave to direct AI workloads into its cloud—very similarly to how Databricks’ native notebooks do or Google’s Colab does or collaborative tools like Hex and Deepnote can—is a game-changer. If, critically, Marimo’s open-source roots persist, CoreWeave can cultivate a community flywheel while seeding valuable integrations for model serving, vector databases, and inference endpoints on its infrastructure – and help the company “move up the stack” away from pure capacity sales to stickier, higher-margin services around the developer workflow. So what does that mean for Core Scientific and CoreWeave? The path forward for Core Scientific looks clearer: monetize the multi-year capacity deal it has wisely signed, convert power and real estate toward AI-ready capacity, and add GPUs or partnerships further up the stack to capture more of the value chain. Other former miners such as Crusoe and Hut 8 have signaled that same painful transition, suggesting a broader reshaping of the entire sector from crypto-centric hosting to AI-first hosting. The strategic logic for CoreWeave is twofold: hardware remains the momentum of growth, but owning the developer entry point will likely compress sales cycles and improve unit economics. The company will keep shopping for tooling that complements inference, fine-tuning, and data prep – all areas where buyers increasingly expect integrated solutions rather than raw GPU power.
The larger lesson: In an AI space that is moving faster than valuations can be set, boards and chairs do not feel it is urgent to buy targets with power and contracts. Look for additional collapses, more program packages like Marimo, and a high price for operators with power that can transform megawatts into AI throughput with predictable economics. The trend is still in full swing — it’s just migrating up the stack.