Musk Buys Cursor, DOJ Shields His Datacenter, Military Buys AI Advice
SpaceX swallows Cursor for sixty billion, the DOJ kills a pollution suit on national security grounds, and generals start letting models pick targets.
The Musk Singularity Tightens
SpaceX exercised its option on Cursor today, an all-stock deal valuing the coding startup at sixty billion dollars. Cursor's engineers now report, in some abstract org-chart sense, to the same man who runs the rockets, the cars, the satellites, the social network, and xAI. The justification is the usual one. Coding agents are infrastructure, infrastructure is strategic, strategic things belong to Elon. The market agrees enthusiastically. The new Powerlaw closed-end fund, which holds nothing but SpaceX and OpenAI stakes, debuted trading at a premium to its own net asset value, because retail investors have decided that paying a dollar ten for a dollar of Musk and Altman exposure is rational portfolio construction.
This is what consolidation looks like when nobody is willing to call it that. A handful of private entities now own the models, the chips, the datacenters, the power contracts, and increasingly the developer tools that everyone else uses to build on top of them. Cursor was the last independent piece of the coding-agent stack with real mindshare. It is independent no longer.
The DOJ Picks a Side
The more revealing story arrived from the Justice Department, which moved to dismiss a pollution lawsuit against a SpaceX datacenter by invoking national security. The cited justification is SpaceX's role in the ongoing Iran conflict. The implication is that any litigation touching a Musk-owned facility involved in defense work is now potentially a state secrets matter. Environmental plaintiffs can go home.
This is the part of the AI buildout that the keynote slides skip. Datacenters need water, power, and permissive regulators. When permissiveness fails, national security is the override switch, and the override is now being pulled openly. Pennsylvania offers the polite version of the same fight. Governor Shapiro wants datacenters statewide; the lawmakers in whose districts they would actually be built do not. Shapiro will probably win, because the alternative is watching the load go to Texas or Virginia, and no governor wants to explain that trade to donors. But at least in Pennsylvania the fight is happening in public.
Regulation Remains a Rumor
After last week's Anthropic incident, legal scholars are still trying to figure out whether the Trump administration actually has the authority to suspend an AI company's operations. NPR's reporting suggests the answer is a confident maybe. There is no statute that cleanly grants the power. There is also no statute that clearly forbids it. Everyone is improvising, which is how you end up with executive actions that get reversed in district court six months later while the underlying models keep training.
Into this vacuum, Jensen Huang offered his contribution to the discourse. Society, he said, must fundamentally change to accommodate AI. New norms are inevitable and necessary. It is a remarkable framing. The technology is treated as a weather system, and humans are asked to update their building codes. Notably absent from the Huang formulation is any obligation on the part of the people selling the GPUs.
Generals With Chatbots
The heaviest item in the slot is the MIT Tech Review report that multiple militaries are now deploying AI models as decision-making advisors inside command structures. Not logistics optimization, not satellite image triage, advisors. Models that get asked what to do and whose outputs shape what gets done.
The problem is not that the models are evil. The problem is that they are confident, fluent, and wrong in ways that are difficult to audit under time pressure. A commander who asks a model for a recommendation and gets a crisp paragraph back is going to weight that paragraph more than the institutional process would suggest is wise. The models hallucinate. The models reflect their training data. The models have no concept of the consequence of being wrong, because consequence is not in the loss function. Wiring them into kill chains, even advisory ones, is the kind of decision that looks reasonable in a briefing deck and catastrophic in an after-action report.
The Boring Item
For balance, HPE is giving away free VMware licenses to poach Broadcom-alienated customers. It is the most normal piece of technology news in the slot, which is itself a kind of indictment. When the most reassuring story of the day is enterprise virtualization knife-fighting, the rest of the day was not reassuring.
- HPE Offers Free VMware Licenses to Poach Customers · Ars Technica · 2/10
- Pennsylvania Governor Battles State Over AI Datacenter Expansion · The Guardian · 3/10
- Musk's SpaceX Acquires AI Startup Cursor for Sixty Billion · New York Times · 4/10
- Militaries Deploy AI Models as Decision-Making Advisors · MIT Tech Review · 7/10
- Trump Administration's AI Regulation Authority Remains Unclear After Anthropic Incident · NPR · 5/10
- Jensen Huang Says Society Must Fundamentally Change for AI · Josh Boak · 4/10
- DOJ Invokes National Security to Kill Pollution Lawsuit Against Musk · New York Times · 8/10
- Powerlaw Fund Trades Above Value Betting on Musk and OpenAI · · 2/10