Snowflake Surging to All-Time Highs

Snowflake shares rose more than 20% after revenue growth accelerated to 37% versus 35% expected, adjusted operating margin expanded to more than 15% from 11%, and management forecast roughly 36% revenue growth for fiscal 2027. AI workloads drove about half of the growth acceleration, although the stock's valuation remains elevated at about 20x forward sales and stock-based compensation diluted shareholders by more than 4% over the past year. Separately, data-center development faces growing opposition: New York and Texas have paused or restricted approvals, while activist groups reportedly blocked or delayed two-thirds of targeted projects in the first half; power bills rose about 5% over the past year and could increase another 6%. The discussion suggests the buildout is likely to continue but with slower, more constrained returns for AI-infrastructure suppliers after sharp declines in names including Marvell, Celestica, and Sterling Infrastructure.

Analysis

SNOW’s rerating now hinges less on another consumption upside and more on whether AI usage converts into durable enterprise commitments. A widening gap between usage growth and contracted demand would expose the model to optimization risk as customers move from experimentation to cost-controlled production; meanwhile, equity dilution remains a material drag on per-share FCF compounding. The near-term setup is asymmetric after the gap: sustained net revenue retention above 120% and incremental margin expansion can support the premium, but any deceleration in product revenue or a renewed rise in dilution would likely drive sharp multiple compression within 1-3 months.

The more investable implication from permitting resistance is a shift in AI-infrastructure economics, not a collapse in spending. Hyperscalers will increasingly need to internalize grid upgrades, behind-the-meter generation, and community concessions; that raises project cost and lengthens deployment schedules, pressuring near-term orders for network and construction suppliers such as MRVL, CLS and STRL. Conversely, constrained compute availability protects the strategic value of installed capacity and favors NVDA near term, but delayed energization eventually creates a mismatch between chip deliveries and revenue-generating cluster utilization.

Consensus appears too binary on data-center approvals. State-level frameworks can ultimately de-risk development pipelines, but the likely outcome is lower project IRRs and more capital concentrated in power-advantaged regions rather than a rapid return to unconstrained builds. Watch hyperscaler capex guidance alongside utility interconnection queues: capex remaining elevated while energized capacity slips would be bearish for suppliers with short order-to-revenue cycles, even if headline AI spending remains strong.

Over 6-18 months, the second-order winners are not necessarily the current AI hardware leaders but firms enabling power availability and construction execution in pre-approved locations. The article provides insufficient project-level backlog, cancellation, and geographic exposure data to underwrite a broad bottom call in CLS/MRVL/STRL; these should remain event-driven rather than treated as clean AI-beta longs.

Key Decisions for Investors

• Do not chase SNOW immediately after the gap; initiate only on a pullback or after the next print confirms product-revenue acceleration without deterioration in net revenue retention. Use a 3-6 month horizon; thesis fails if retention falls below 120%, dilution reaccelerates, or FY revenue guidance stops rising.
• Maintain NVDA as the higher-quality AI-infrastructure exposure versus MRVL and CLS for the next 1-3 months: supply scarcity and installed-base software economics provide better downside support than component suppliers exposed to timing slippage. Reduce if hyperscaler capex guidance or cluster deployment commentary turns materially lower.
• Avoid averaging into STRL, CLS, or MRVL solely on drawdowns. Establish an alert around backlog conversion, customer cancellations, and revised delivery schedules; a long basket becomes actionable only if order growth reaccelerates while project energization timelines stabilize.
• Consider a tactical pair: long NVDA / short an equal-dollar basket of MRVL and CLS over 1-3 months, sized modestly. It isolates the risk that permitting and power delays defer lower-margin infrastructure revenue while scarce accelerators retain pricing power; exit if supplier order commentary improves or NVDA supply constraints ease sharply.
• Monitor META and GOOG capex disclosures and state-level grid-cost allocation rules. Evidence that hyperscalers accept direct power-build obligations is structurally positive for build continuity but initially negative for free-cash-flow margins and could cap their multiples despite sustained AI revenue optimism.