Nvidia is paying Poolside $6B to license the software system that built its Laguna model and bringing roughly 109 of the model's engineers onto its own payroll, while separately investing $1B in the remaining company at a $12B pre-money valuation. The founders describe the structure as neither an acquisition nor an acquihire, and it mirrors the shape of Nvidia's $20B Groq deal from December: absorb the model-layer talent and IP, leave the corporate shell nominally independent, and skip the antitrust review a straight purchase would trigger. Nvidia's dealmaking increasingly runs on structures like this rather than plain M&A. (The Information)
Private Companies
Muon Spacefundraise
Muon Space closed a $250M Series C at a $1.5B valuation, with Google and Salesforce Ventures joining Eclipse Capital and existing backers. The company builds and launches its own satellites — 11 already in orbit — and is now directing capital toward orbital AI computing, following Starcloud's push to put GPUs in space. The round scales manufacturing toward 500 satellites a year by 2027, positioning Muon as infrastructure for a compute layer that doesn't yet have a cost structure competitive with terrestrial data centers. (Bloomberg)
Rebellionsother
Reportedly, Nvidia CEO Jensen Huang is discussing a technical partnership, an investment, or an acquisition with Rebellions — talks are preliminary and may not lead to a transaction. Rebellions is a South Korean NPU designer for AI inference backed by SK Hynix, Samsung Ventures, and Arm at a $2.3B valuation. (Bloomberg)
Starcloudfundraise
Starcloud raised a $250M extension to its March Series A, more than doubling its valuation to $2.3B from $1.1B five months earlier, with Nvidia contributing $25M alongside Cisco, Benchmark, and EQT. The capital funds a larger manufacturing facility and Starcloud-3, the company's largest orbital data center, designed to fly on SpaceX's Starship. Starcloud is the only operator running an Nvidia H100 in orbit and is feeding those results back into Nvidia's in-development Vera Rubin Space-1 chip. (TechCrunch)
Public Markets
Micron will invest $10B over a decade in a Boise research lab for memory and AI-compute architectures — the first dedicated US memory-research hub, and a sign suppliers now see enough AI demand to fund long-horizon R&D themselves. (company press release)
Broadcom is in talks to raise up to $100B in debt, including a $30B junior tranche with Blackstone and Apollo, extending the June deal that backs Anthropic's compute buildout toward 20+ GW of capacity by 2028. (Bloomberg)
Emerging
Export-control fallout: Super Micro's board-led investigation found no evidence that CEO Charles Liang or senior management knew of the scheme, tied to a former co-founder, to divert $2.5B of Nvidia-chipped servers to China in violation of export controls. The company disclosed personnel terminations but no restatement of prior financials, closing out a five-month overhang that had clouded its standing with hyperscale customers since the March indictment. (Fortune)