The broad sell-off in semiconductor stocks is no accident. May’s U.S. non-farm payroll data—showing a gain of 172,000 jobs, more than double the expected 80,000—sent Treasury yields and the dollar surging, reigniting fears that the Federal Reserve will delay rate cuts. In this macro shift, high-valuation, high-growth AI chip stocks were the first to bleed. NVIDIA, TSMC (Taiwan, China), and Micron all slid, but it was Broadcom’s latest earnings report that truly exposed cracks beneath the surface.
Despite reporting 30% year-over-year revenue growth, Broadcom signaled slowing momentum in its AI segment. More critically, it lowered its next-quarter revenue guidance and hinted that major cloud service providers (CSPs) are shifting from “frenzied procurement” to “precision deployment.” This marks a pivotal transition: AI infrastructure investment is moving from land-grabbing to efficiency-driven optimization. For two years, NVIDIA dominated with its general-purpose GPU architecture, while Broadcom bet on custom AI accelerators—like those designed for Google and Amazon as TPU alternatives. But as customers tighten budgets, the high development costs and long payback cycles of custom chips become liabilities. I judge this not as a temporary dip but as an early signal of a paradigm shift in AI hardware investment.
Intel, meanwhile, finds itself in particularly perilous territory. Its stock plunged 6% in a single day—far exceeding the sector average. On the surface, this reflects disappointment over delayed foundry progress and underwhelming AI chip shipments. Deeper down, it reveals how the market is repricing Intel’s strategic ambiguity. The company tries to be three things at once: an IDM 2.0 reformer, an advanced packaging leader, and an AI accelerator challenger—yet has failed to establish decisive advantage in any. Its Gaudi 3 accelerator may rival NVIDIA’s H100 on paper, but without software ecosystem support, adoption remains minimal. Meanwhile, repeated delays in its 18A process node have pushed potential clients like Microsoft and Amazon toward TSMC’s (Taiwan, China) N2P. Intel is no longer an option—it’s becoming a risk.
Notably, Micron and Western Digital (parent of SanDisk) held up relatively well—not because they’re at the AI forefront, but precisely because they aren’t. They benefit from sustained demand for high-bandwidth memory (HBM) and enterprise SSDs in AI servers. Micron is already shipping HBM3E to NVIDIA at scale and plans HBM4 volume production by 2026. This “pick-and-shovel” strategy insulates them from direct competition while securing steady orders from AI infrastructure build-outs. It underscores a neglected truth: in the AI arms race, the safest bets are often not the guns, but the bullets and shovels.
Broadcom’s report also reveals another trend: hyperscalers are accelerating vertical integration. Google’s TPU v5e, Amazon’s Graviton+Trainium combo, and Microsoft’s custom MI300X variants with AMD—all aim to reduce reliance on general-purpose GPUs. Broadcom, as a custom chip supplier, should benefit, yet its earnings show these deals carry significantly lower margins than its legacy networking business. Even big wins may not translate into profit growth. NVIDIA’s moat isn’t just hardware—it’s the CUDA ecosystem’s switching cost. Without a comparable software stack, Broadcom’s AI chip business risks remaining trapped in a “high-cost, low-return” cycle.
This market correction is finally deflating the hype accumulated over the past two years. Investors are now distinguishing true AI beneficiaries from mere AI-themed tickers. Companies with only concepts—but no real orders or technical moats—face painful valuation resets. Those with genuine scarcity—TSMC’s (Taiwan, China) advanced nodes, Micron’s HBM capacity, NVIDIA’s software dominance—will endure.
Can Intel survive this reckoning? Its fate hinges on two questions: Can it prove 18A’s reliability and cost competitiveness by 2026? And can it secure large-scale Gaudi 4 adoption from at least one top-tier CSP? Failure on both fronts could relegate the former semiconductor titan to irrelevance—a marginal player in foundry services. Broadcom’s challenge, meanwhile, is balancing customization with scale. Without that equilibrium, it may become the most expensive spectator in the AI boom.
As markets sober from euphoria, the semiconductor industry’s real inflection point is only now emerging.