On June 5, 2026, the ProShares Ultra Technology ETF (ROM) plunged 13.45% in a single day, wiping out over $10 billion in market value. The trigger wasn’t macroeconomic turmoil but a subtle shortfall in growth expectations for two core holdings: Broadcom and NVIDIA. Despite Broadcom reporting nearly 300% year-over-year growth in AI semiconductor revenue, its “cautious guidance” for the next quarter was enough to spark a sell-off. This episode reveals a deeper truth: beneath the surface of AI compute euphoria, the semiconductor industry’s valuation logic is undergoing a silent yet seismic recalibration.
Broadcom’s predicament is emblematic. As one of the few companies offering a full-stack AI infrastructure solution, its custom ASIC business is deeply embedded in the training clusters of OpenAI, Microsoft, and even Google. Yet this concentrated customer base has become a double-edged sword. Any slowdown in deployment or shift toward in-house silicon by these hyperscalers immediately pressures Broadcom’s revenue trajectory. Goldman Sachs recently downgraded its price target precisely due to “overreliance on top cloud vendors”—not technical shortcomings. This risk is especially acute in 2026: as NVIDIA’s Blackwell platform enters mass delivery, cloud giants are evaluating second-source strategies to rebalance bargaining power. While Broadcom benefits from software synergies post-VMware integration, it still struggles to breach NVIDIA’s hardware-software ecosystem moat.
More telling is Apple’s absence. As the world’s most valuable tech company, Apple remains conspicuously silent in the AI chip race. Its M-series chips excel at on-device inference but lack datacenter-scale training capabilities and offer no external AI acceleration services. Consequently, Apple is systematically marginalized in the current capital narrative dominated by training compute. ETFs like ROM, weighted toward high-growth, high-volatility AI infrastructure stocks, naturally dilute Apple’s influence. This isn’t just a portfolio issue—it’s a signal of shifting industrial power: the locus of compute authority is migrating from consumer devices to cloud infrastructure.
Goldman Sachs plays a pivotal role here. As lead financial advisor on Broadcom’s 2024 VMware acquisition, Goldman helped craft the narrative of Broadcom’s transformation into an “end-to-end AI infrastructure provider.” Yet when markets began questioning that story’s sustainability, Goldman swiftly pivoted, now highlighting “customer concentration” and “peak capex cycle” risks. This duality underscores Wall Street’s dual identity in semiconductor cycles: both catalyst of strategic reinvention and amplifier of sentiment swings. Data shows Goldman’s proprietary trading desk significantly reduced positions in ROM constituents a week before June 5, with research tone shifts often preceding market moves by days.
The fracture isn’t just inter-corporate—it runs through the entire semiconductor capital ecosystem. ROM’s crash isn’t isolated; it’s a precursor to localized AI bubble deflation. For two years, investors equated AI with limitless growth, ignoring three hard constraints: capacity ramp limits, power bottlenecks, and customer concentration. Broadcom’s “miss” is essentially a correction of that linear thinking. I judge that H2 2026 will see more such adjustments: firms propped up by AI hype without real order validation will face further selling, while those with genuine heterogeneous integration capabilities—bridging training and inference—will command premiums.
Apple’s silence may be strategic. Rather than battling NVIDIA in the crowded training chip arena, it could be betting on edge AI maturity. But this path carries risk: if cloud models keep evolving and dominate user interfaces, the value of client-side silicon could erode further. Broadcom, meanwhile, stands at a precipice: it must prove it’s not merely OpenAI’s contract manufacturer but a platform player capable of building its own ecosystem.
As ROM holders watched their paper wealth evaporate, what they truly lost wasn’t stock value—but faith in an outdated growth paradigm. The semiconductor industry has never been so dependent on a handful of companies, customers, and technical pathways. Can this fragile prosperity withstand the next payroll report or earnings guidance shock? The answer will determine who truly endures in Act II of the AI era.