Nasdaq Loses $2.5 Trillion as Samsung Races to Build AI's Chip Supply

Almost $2.5 trillion in market value vanished from technology stocks in a single session, as the Nasdaq posted its worst day since April 2025 amid fears of an AI slowdown. The sell-off, which began on Wall Street, spread to markets in Asia and Europe, where investors also cited expectations of tighter AI oversight. The so-called Magnificent Seven companies lost more than $1 trillion combined, and Tesla and SpaceX shed over $420 billion together. Jim Carroll, a senior wealth adviser at Ballast Rock Private Wealth, told Reuters how the mood on trading floors felt: "The steamroller showed up today."

The correction lands at a moment when physical investment in AI infrastructure keeps accelerating. At Pyeongtaek, south of Seoul, Samsung operates three semiconductor fabrication plants around the clock, is bringing a fourth online, and has thousands of workers racing to build a fifth. Generative AI models depend on stacks of specialized memory chips, and the sprawling South Korean campus exists to meet that soaring demand. For R&D managers, the divergence matters: paper valuations are falling while the capital equipment and fab construction that supply the compute layer continue to expand.

How close are domestic humanoid robots to autonomy?

A reporter for MIT Technology Review signed up for a new humanoid-robot cleaning service in San Francisco after seeing it advertised on X. The machine that arrived belongs to Tau Robotics, a local startup whose long-term goal is autonomous domestic cleaning. The technology is not there yet. For now, human operators pilot the robots remotely, watching their work through onboard cameras. The trial produced unexpected challenges alongside moments the reporter described as a glimpse of the future. The mechanism is worth noting for anyone tracking embodied AI: teleoperation (a human controlling the robot from afar) is carrying the commercial service until perception and manipulation software can run unsupervised.

Is AI now accelerating AI research?

Researchers report that Anthropic's Claude models are helping the company improve its successors nearly twice as fast, according to The Information. The claim signals a measurable R&D feedback loop at one frontier lab. The same coverage cautions that recursive self-improvement, the process of AI systems building better AI systems with minimal human input, will take time to mature.

What else moved this week?

• Political pressure on Big Tech. Emboldened Democrats are preparing to haul tech CEOs before Congress. Senator Bernie Sanders has proposed a national data center moratorium, and Representative Yassamin Ansari plans legislation to pause advanced AI development. Reuters reports the backlash could delay Anthropic's IPO.
• Industrial refrigeration hacks. The US government warned of an "ongoing cyber threat" to commercial fridges and freezers after a spate of hacks. Russia has denied claims it is behind the attacks.
• Volcanism on Mars. Scientists have new evidence of active volcanoes at Alba Mons, which may boost the chances of finding life on the planet.
• Covid vaccine economics. A new analysis places the economic benefits of Covid vaccines at up to $64 trillion, roughly double earlier estimates.
• Ocean color change. The Great Pacific Garbage Patch has brightened the water around it, Scientific American reports.
• NASA's X-70. The agency unveiled a mysterious experimental aircraft expected to take flight within two years.
• Cybertruck wind-down. Elon Musk has signaled the end of the model, with Tesla winding down production in Texas next year after one of last year's biggest tech flops.

What does a niche weather app show about data strategy?

OpenSnow, a startup built by two broke skiers, has become the best snow-forecasting service for winter sports, outperforming federally funded tools and big-name brands. It combines government data with its own AI models and decades of alpine experience, and its forecasters write "Daily Snow" reports for locations worldwide. The app proved especially vital during last winter, one of the weirdest on record, and made microcelebrities of its forecasters. The case illustrates how proprietary modeling layered on public datasets can beat both government services and incumbent brands in a narrow vertical.

In lighter items, 7,000 users of the RunPee app signed up for the three-hour film "You Can See Everything," Arizona's Chiricahua became a US national park, a 1960s Futuro flying-saucer house listed for $1.5 million, and a bear named Backpack won Fat Bear Week. Whether the market correction deepens or stabilizes, the week's fab construction schedules and research-speed claims will give R&D planners harder data than the ticker.