Morgan Stanley Says Humanoid Market Hits $5T; Memory Guys Already Counting

Morgan Stanley dropped a $5 trillion total addressable market estimate for humanoid robots this week, roughly doubling prior Street consensus and sending memory chip makers into full hype mode. Samsung and SK Hynix both told investors humanoids represent the next major DRAM demand cycle after AI data centers, with Korea JoongAng Daily reporting that each bot could require 16 to 64 gigabytes of high-bandwidth memory depending on task complexity. That's more than most laptops and about half what a single H100 GPU burns through.

The timing matters because hyperscaler capex—while still massive—is starting to spook bond markets. One wire service noted that AI infrastructure spending is now large enough to nudge sovereign borrowing costs higher in several Asian economies, a polite way of saying investors are repricing risk. DBS Group's CEO told reporters that financing discipline is returning as lenders get pickier about which AI bets to fund. Tencent, meanwhile, just sold $5 billion in bonds explicitly earmarked for AI capex, per Bloomberg, suggesting the credit window is open but narrowing.

XPeng announced it will roll out its IRON humanoid across Europe in the second half of 2027, joining a crowded field that now includes 120 robots on display at this week's Hong Kong electronics fairs. The South China Morning Post counted them. None of these are deployed at scale yet, but the demos are getting louder and the supply chain is already reacting. Memory fabs don't wait for purchase orders—they ramp on forecast.

JPMorgan analysts told clients that AI will drive 70 percent of earnings growth across Asian equities next year, with South Korea and Taiwan rated overweight on the strength of semiconductor exposure. The bank didn't break out humanoids as a separate line item, but the implication is clear: if robots are the next leg of the AI spending cycle, the chip guys who survived data center buildout are best positioned to capture it. Samsung and SK Hynix have been here before.

Elsewhere in the capex pile, Google Cloud revenue is up 82 percent year-over-year, and Alphabet is defending a $15 billion AI hub investment in India against local water-use concerns. ProLogium announced a 381 Wh/kg solid-state battery aimed at data center backup power, a reminder that AI infrastructure now spans everything from chips to real estate to utilities. An analysis from TradingView noted that 42 AI-related stocks have driven 67 percent of S&P 500 gains since 2024, which is either a sign of concentration risk or proof that the market knows exactly where the money is going.

The bull case is straightforward: if humanoids require meaningful onboard compute and memory, and if deployment scales anything like the data center ramp, then Samsung and SK Hynix print money for the next three years. The bear case is that $5 trillion is a made-up number, that most bots will stay in warehouses, and that memory pricing collapses the moment supply overtakes actual—not projected—demand.

For now, the memory makers are talking like believers. Whether they're front-running real orders or just front-running each other is the $5 trillion question.