Deep Dive
The AI Graveyard: What's Actually Failing
InvestAnswers opens with a blunt assessment: AI hasn't failed, but plenty of AI players are getting left behind. The biggest shock is that someone in Korea allegedly reverse-engineered Claude's closed-source code in Python overnight, and because it's new code, it can't be taken down for copyright infringement. The broader question: what's a moat in AI anymore? Meanwhile, a Forbes article called OpenAI's history an 'graveyard'—Sora killed, Grok abandoned, Stargate dead. The company keeps killing products, which raises real questions about vision and execution at the world's most valuable startup.
Memory Wins, Software Loses—At Least for Now
The actual winners in AI aren't the AI companies—they're memory and semiconductors. Micron crashed from $475 to around $160 after Google announced 8X more efficient memory, but it's got a buy signal at $165 and demand is still through the roof. Meanwhile, AI as a category has underperformed compared to data centers and chips. SaaS companies have been destroyed, proving AI is real and competitive. The market is slowly figuring out that the real money is in chips and infrastructure, not software.
OpenAI's Secondary Market Collapse and Legal Trouble
This is the dark part. OpenAI has $600 million in shares sitting on the secondary market with zero demand, while Anthropic shares are flying off shelves. The reason became clear today: Sam Altman's sister filed an amended lawsuit alleging sexual abuse as recently as 2006. Lawsuits in actual courts don't disappear quickly, and they can kill IPO plans. InvestAnswers's advice is blunt—if someone offers you OpenAI shares on the street corner, don't buy them. Too risky until this dust clears.
Data Center Bottlenecks and the China Problem
Half of planned US data centers for 2026 will be delayed or cancelled. The chokepoint isn't space or chip availability—it's transformers, the huge electrical equipment that channels power from the grid. Elon called this out a year ago. The US doesn't manufacture enough of them and has to import from China. Out of 35 to 36 gigawatts of announced data center power, only 4 or 5 are under construction. The other 32 haven't broken ground. Cooling is eating 10% of global electricity, and it's getting worse as AI scales. This is why space data centers matter—cooling is easier and cheaper up there.
Physical AI and Space: The Real Next Wave
Tesla's Full Self-Driving just hit 9 billion miles, and it took only 41 days to go from 8 billion to 9 billion. The first billion took three and a half years. FSD 14.3 is rolling out to employees this month and should hit wide release next week, with unsupervised driving as the goal. Jensen Huang calls the next phase the physical AI revolution—self-driving cars, humanoid robots. Meanwhile, StarCloud is raising money to build data centers in space using SpaceX launches at roughly $500 per kilogram. If Starship scales and costs drop under $100 per kilogram, space becomes economically viable and SpaceX wins big. The APO is slated for June and could be the biggest IPO ever.