AI Winter Arrives: Tech Sell-Off Exposes the Valuation Reckoning

The technology sector's AI-fuelled rally has hit a wall, and the wreckage is instructive. On a single trading session, the Nasdaq fell 2.21% while the S&P 500 dropped 1.44%. In South Korea, where the semiconductor supply chain keeps the world's electronics fed, the Kospi index tumbled 10%—hard enough to trigger a 20-minute circuit breaker designed to cool overheated markets. This is what happens when narrative fatigue meets margin compression.

The semiconductor stocks that had been treated as infrastructure plays for the AI boom are taking the heaviest damage. SK Hynix and Samsung, two of the world's dominant memory chipmakers, both fell more than 12%. SanDisk, Micron Technology and Arm all plunged more than 10%. Marvell, Analog Devices, Western Digital, Texas Instruments and Qualcomm—the entire ecosystem of companies supplying the chips that power everything from data centres to consumer devices—all dropped around 9%. Even Nvidia, the largest publicly traded company in the world and the central narrative of the AI boom, tumbled 4.15%. That last figure matters. When your biggest winner starts losing ground, the thesis has shifted.

The sell-off reveals what has been quietly crystallising among institutional investors: the gap between AI hype and AI profitability is wider than the market had priced in. Wall Street spent months embracing a simple story—companies pour billions into artificial intelligence infrastructure and software, and those investments translate into faster revenue growth and higher profits. It was a clean narrative. It did not require investors to ask uncomfortable follow-up questions. But markets eventually do ask follow-up questions, and when they do, valuations respond.

Wall Street is now demanding something it should have demanded from day one: evidence. Not optimism. Not roadmaps. Evidence that the spending will pay off. This is where the air starts leaving the balloon. The industry has forced companies to spend and borrow tens of billions of dollars to build and develop the technology without the immediate results to show for it. That is sustainable for a quarter or two. It becomes harder to sell after that.

What is happening now is a bifurcation of the AI narrative. Investors are beginning to separate the companies selling the scarce infrastructure behind artificial intelligence—the picks and shovels play—from those spending heavily to build and run the models without yet showing the profit gains that would justify their elevated valuations. This distinction matters. A chipmaker selling $50 billion in silicon annually to data centre operators has a clear path to revenue and margin expansion. A company that has invested $20 billion into large language models and is still figuring out how to monetise them has a more complicated story.

The broader implication is stark enough that even the capital-raising ambitions of the sector's most celebrated players are flagging. OpenAI is considering delaying its IPO because of the recent market volatility that could make it difficult for the company to fetch its desired $1 trillion valuation. When the company at the centre of the AI boom is deciding that now is not the time to ask public markets for a breathtaking valuation, you know the momentum has shifted.

This is not the end of artificial intelligence. The technology remains genuinely transformative, and the long-term demand case is intact. But the AI story is no longer treated as a single monolithic trade. It has fractured into a series of more granular bets, each one requiring its own justification. Some will survive this reckoning unscathed. Others will discover that multiples compress quickly when the earnings growth narrative runs into the reality of margin pressure and delayed payoffs. The market is now asking which is which. The answers will be ugly for some.