AI Storage Giants Post Surging Profits but Stocks Plummet: A Valuation Rift in the Capex Cycle
AI Chip and Memory Giants Beat Expectations—Yet Stocks Plunge: The Valuation Rift Beneath the Capex Frenzy
When SanDisk reported third-fiscal-quarter revenue of $5.95 billion—26% above consensus—and adjusted EPS of $23.41—61% above expectations—while Western Digital simultaneously posted a staggering 50.2% gross margin and 45% year-on-year revenue growth, the market should have celebrated the full realization of the AI hardware cycle. Yet reality defied logic: SanDisk plunged 8% after hours; Western Digital fell 6.4%. This paradox—“soaring profits, collapsing stock prices”—is no random fluctuation. It marks a pivotal watershed: the AI semiconductor supply chain’s transition from exuberant narrative to ruthless validation. It exposes three structural fault lines beneath inflated valuations: (1) an emerging inflection point in marginal growth, (2) intensifying customer concentration risk, and (3) mounting doubts over the sustainability of capital expenditures.
Profits Surge on AI Demand—but Growth Structure Hides Deep Vulnerabilities
Financial results unmistakably confirm AI’s transformative impact on the memory layer. SanDisk’s fourth-fiscal-quarter guidance—$7.75–$8.25 billion in revenue—far exceeds the consensus of $6.65 billion. Its primary drivers? Mass deployment of HBM3 and PCIe 5.0 SSDs. Industry surveys show HBM packaging and test capacity utilization has held above 98% for six consecutive quarters in 2024, while lead times for premium enterprise SSDs have stretched to 22 weeks. Western Digital’s extraordinary 50.2% gross margin directly reflects its pricing power in the AI server SSD market—its U.2/U.3 enterprise SSDs now account for 65% of the supply chain for NVIDIA’s DGX GB200 clusters.
But this high-growth trajectory rests on a perilously fragile customer base. SanDisk’s top two customers—NVIDIA and Microsoft Azure—account for 73% of its AI-related revenue; Western Digital’s cloud-service-provider channel now contributes 58% of its total revenue. Such “single-point dependency” grows especially dangerous amid accelerating technology transitions: as NVIDIA’s next-generation Blackwell Ultra platform shifts toward CXL-based memory pooling architectures, traditional SSD interfaces may face structural obsolescence. During the earnings call, SanDisk’s CFO candidly admitted: “While HBM3 orders remain robust, customers are demanding that 40% of the contract value for HBM4 be paid as advance deposits in 2025—a de facto transfer of future risk onto suppliers.”
Valuation Overextension and Capex Misalignment: The Market Repricing the “AI Premium”
SanDisk currently trades at a forward P/E of 38x; Western Digital at 22x—both significantly above the semiconductor equipment sector’s average of 18x. The deeper logic behind the sell-off lies in systemic skepticism over the sustainability of AI-related capex. According to IDC’s latest report, storage modules now constitute 37% of global AI server procurement budgets in 2024—up from 29% in 2023. Yet capex growth among leading cloud providers is diverging: Microsoft Azure’s capex declined 5% quarter-on-quarter this quarter, while Meta—though maintaining high investment—explicitly stated it will “optimize SSD procurement strategies in 2025 by shifting toward hybrid memory architectures.”
More critically, memory vendors themselves are caught in a capex dilemma. SanDisk announced an additional $3.2 billion investment to expand its HBM packaging and test capacity—yet depreciation and amortization as a share of revenue has surged from 11% in 2022 to 19% in Q3 2024. When depreciation alone consumes nearly one-fifth of revenue, “high gross margins” are, in essence, an accounting illusion—temporary and unsustainable. In an unusually frank disclosure, Western Digital revealed in its earnings report: “Yield on our existing SSD production lines has improved to 99.2%, but yield on our newly commissioned HBM3 wafer-level packaging line stands at just 76%—meaning 24 out of every 100 wafers require rework, directly eroding gross margin elasticity.”
Apple’s Steadfast Performance Highlights a Valuation Schism: Consumer Electronics vs. AI Infrastructure
On the same trading day, Apple delivered an earnings beat: demand for the iPhone 17 was described as “extremely strong,” services revenue surpassed $30.9 billion, and the board approved a $100-billion share buyback program. Its stock rose 3.2% that day, with its forward P/E holding steady at 31x. This contrast is profoundly instructive. Apple embodies the time-tested “cash-flow generator” paradigm—hardware sales lock in users; service ecosystems generate durable, recurring free cash flow. SanDisk and peers, by contrast, represent the “capital-intensive” paradigm: every $1 of revenue requires $0.42 in capex (versus Apple’s $0.08), and profitability remains tightly tethered to a single technology roadmap and the creditworthiness of a handful of customers.
That Intel’s stock doubled in April—setting a Nasdaq record spanning 55 years—failed to lift the memory sector underscores how the market is now segmenting the AI theme: chip design (IP value), foundry manufacturing (capacity scarcity), and equipment (technology barriers) command premiums—while memory, the most standardized link in the chain, is being recategorized as a “cyclical capital good.” Deutsche Bank’s research note cuts to the core: “Investors no longer pay for the ‘AI story’—they pay only for ‘verifiable three-year free cash flow.’ SanDisk’s FCF yield over the next 12 months stands at just 3.1%, below the S&P 500’s average of 4.7%.”
Transition to the Profit-Realization Phase: A Structural Revaluation of the Entire Value Chain
This sharp correction signals a fundamental shift in AI hardware investment logic. The linear narrative of the past two years—“technical breakthrough → order surge → stock rally”—has ended. It is being replaced by a four-dimensional evaluation framework: “technology deployment → customer validation → cash-flow quality → return on capital.” For memory vendors, the market now asks pointed questions: Can HBM3 orders cover depreciation on new production lines? Does that 50% SSD gross margin include one-time inventory revaluation gains? Does concentrated cloud procurement mask implicit price pressure?
For the broader semiconductor ecosystem, this triggers a cascade of valuation pressures. Equipment makers face downward pressure unless they can prove their HBM etch tools remain irreplaceable beyond 2025; foundries risk amplifying utilization uncertainty if they cannot diversify AI customers into automotive or industrial long-tail markets. The sole clear winners are companies like Apple—those with end-device pricing power and proven service monetization capability. With $111.1 billion in revenue, Apple demonstrates a critical truth: when AI becomes infrastructure, the real moat lies not in stacking compute, but in owning user mindshare and closed-loop ecosystems.
SanDisk’s and Western Digital’s stock plunge does not signal the ebbing of the AI wave—it reveals the jagged rocks exposed once the tide recedes. Valuations once effortlessly buoyed by the word “AI” must now withstand rigorous scrutiny under a three-lens microscope: free cash flow generation, customer diversification, and intergenerational technology risk. When markets stop cheering “next-quarter growth” and begin calculating “three-year capital returns,” the semiconductor industry finally enters the deep waters of rational, sustainable prosperity.