Wall Street's AI Trade Hits a Air Pocket — But the Chip Orders Keep Coming

If you've glanced at a market summary this week, you've seen the word "momentum unwind" more than once. Semiconductor and AI infrastructure stocks have taken a beating — some chip-equipment names are down 20-25% over a matter of days — and the popular narrative is that the AI trade is finally cracking under its own weight. The more sober read from trading desks across the Street is less dramatic: this looks like a technical unwind of crowded positioning after a strong second-quarter run, not a reassessment of whether companies actually need more AI computing capacity. Momentum-tracking indices are down in the range typically seen during ordinary consolidation, not the kind of drawdown associated with a genuine change in the fundamental story.

The clearest evidence that the underlying demand story remains intact came this week from Apple and Broadcom, who announced an expanded, multiyear custom silicon and wireless-chip partnership worth more than $30 billion — Apple's largest domestic manufacturing commitment to date. The deal is expected to produce more than 15 billion U.S.-made chips and includes a $1.5 billion investment to modernize Broadcom's Colorado manufacturing facility. Whatever one thinks of the political packaging around "American-made" chip announcements, the dollar figure is real, and it lands directly on top of similar multibillion-dollar commitments other chipmakers have made this year to expand domestic capacity. Separately, Broadcom's own management, in meetings with analysts this week, expressed confidence that demand for its AI networking and custom silicon products remains higher than what's currently reflected in next year's guidance — an unusually bullish signal from a company that has already benefited enormously from the AI buildout.

There's a genuine tension worth watching, though. China's DeepSeek, the AI lab that rattled markets last year with a cheaper training approach, is reportedly now developing its own AI inference chips, a move aimed at reducing dependence on both Nvidia and Huawei. A separate industry survey found Chinese firms plan to allocate nearly half their AI hardware budgets to domestic suppliers going forward. That's a slow-moving but structurally important story: the world's AI hardware supply chain is bifurcating along geopolitical lines, and over the medium term it could dent the addressable market for American chip designers in China even as demand at home stays robust.

For the broader economy, the practical takeaway is this: the stock-price gyrations in chip names reflect trader positioning and seasonal profit-taking more than a change in the multi-year capital spending cycle among the hyperscale cloud companies. Amazon, Google and Meta are all reportedly still raising their own capital expenditure plans, not cutting them. That capex is a meaningful driver of GDP growth and of the demand for electricity, data-center construction, and skilled trades labor — including on Long Island, where several utility and construction firms feed into the data-center supply chain. A cooling in stock momentum doesn't necessarily mean a cooling in the physical buildout, and readers should be careful not to conflate the two.