Three stories dropped in the last 24 hours that together paint a pretty clear picture of where AI is headed in the second half of 2026. Big compute bets, serious scientific talent moving around, and China making a very deliberate move. Let's get into it.
SpaceX Just Wrote a $6 Billion Check to an AI Startup
SpaceX signed a compute lease with Reflection AI starting July 1 at $150 million a month. Run that through the full 2029 term and you land at $6.3 billion. That is not a small experiment.
Reflection AI is an open-source shop founded by researchers who came out of Google DeepMind. They have been quiet, but apparently not quiet enough for SpaceX to pass on them. Either Elon sees something in the open-source AI stack that the rest of the industry is sleeping on, or SpaceX needs that much raw compute for their own internal projects and this is a way to offset the cost. Probably both.
Either way, a $150 million per month compute deal is a signal. Infrastructure is still the game. Whoever controls the chips and the power controls the ceiling for what anyone can build on top of it.
The Guy Who Cracked Protein Folding Just Jumped to Anthropic
John Jumper, one of the co-creators of AlphaFold2, is leaving Google DeepMind after nine years to join Anthropic. He won a Nobel Prize for that work. A literal Nobel Prize. And now he is going to Anthropic.
That is a headline worth slowing down for. AlphaFold changed biology in a real way. It solved a problem that the scientific community had been stuck on for decades. The fact that the person behind that is now at Anthropic tells you a couple things. First, Anthropic is doing something interesting enough to pull someone with that kind of track record. Second, the overlap between deep scientific research and frontier AI modeling is only going to get more important from here.
I have said before that the next wave of AI value is not going to come from better chatbots. It is going to come from AI doing actual hard science. Jumper moving to Anthropic fits that thesis pretty well.
China Sees an Opening and Is Not Wasting It
China's AI sector pulled in $7.4 billion in a fresh funding round, timed pretty directly to the US regulatory pressure building around export controls and access to frontier models. When the US restricts something, capital finds a way around it. This is just that, at scale.
It would be easy to dismiss this as reactive, but that is probably the wrong read. Chinese AI developers have been building seriously for a while now. What the US restrictions actually do is give them a cleaner market narrative. They can go to investors, to governments, to enterprise buyers, and say the American alternative may not be available to you. That pitch lands in a lot of places.
Restricting access to frontier AI without a credible open alternative is a tricky game. It might slow some people down in the short term. It also hands other people a very good reason to fund alternatives at $7.4 billion a pop.