Meta is buying solar power beamed from space. Memory chips are booming because AI can't stop eating data. And an AI just learned to teach itself.

Overview Energy signed its first commercial contract — with Meta. The startup uses satellites in orbit to capture solar energy and beam it back to Earth. The key detail: it works at night. No day-night cycle dependency. Continuous clean power from space.

This is Overview Energy's inaugural deal, so it's still early. But the fact that Meta is the first customer tells you where the demand is. AI data centers are power-hungry and getting hungrier. The big tech companies are racing to lock down energy sources that can actually scale with compute.

Space-based solar has been theoretical for decades. This is the first time a major tech company put a contract behind it.

Micron and SanDisk stocks surged as analysts confirmed memory demand will stay elevated through at least 2030. AI data centers need massive amounts of storage — training datasets, model checkpoints, inference data. That's driving NAND demand alongside the HBM and DRAM boom.

The whole memory sector is tight because the big DRAM makers shifted capacity toward HBM, which is more profitable for AI workloads. That shift created shortages that ripple across DRAM and NAND pricing broadly.

The two biggest memory device providers are riding the wave. When compute scales, storage has to scale with it — and right now, supply can't keep up.

OpenAI released an open-source, on-device model that sanitizes enterprise datasets by removing personal information before the data ever touches an AI system.

It runs locally — no cloud round-trip needed. The model identifies and strips PII from datasets so companies can use their data for AI without exposing employee or customer information.

This is OpenAI's first major open-source play focused purely on privacy. By making it on-device and open-source, they're removing two of the biggest objections enterprises have to feeding data into AI systems.

At his Senate confirmation hearing, Kevin Warsh laid out four priorities: revert to a strict 2% inflation target, switch the preferred inflation gauge from core PCE to trimmed-mean PCE, shrink the balance sheet meaningfully, and end forward guidance and the dot plot entirely. He called it a 'regime change' toward a quieter Fed with 'messier' meetings.

The trimmed-mean PCE switch is the tell. It currently reads about 2.3% versus core PCE at 2.8%. That's a methodological change that makes the inflation problem look smaller on paper — which gives political cover to cut rates. Warsh has also argued in writing that AI is a major disinflationary force and that productivity gains justify lower rates than would otherwise be warranted.

But the tension is real. Low job growth plus inflation still elevated is a stagflationary setup. Stocks are at all-time highs, housing is outpacing wages, and asset prices are frothy. Warsh himself warned in 2009-era speeches that prolonged low rates 'distort asset prices, encourage leverage, and weaken incentives for financial institutions.' Applying that same framework today would argue against cuts.

David Silver — the DeepMind researcher behind AlphaGo — left to start Ineffable Intelligence, a British AI lab. They just raised $1.1 billion at a $5.1 billion valuation, led by Sequoia Capital.

The approach: build AI systems that learn through reinforcement learning without relying on human-generated data. No curated datasets, no human annotations. The model teaches itself.

This is a significant departure from how most AI is built today. Current models are trained on massive amounts of human-created text, code, and images. Silver is betting that the next breakthrough comes from removing that dependency entirely.