By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel is bearish on the current AI infrastructure rally, citing high valuations, unsustainable capital intensity, cyclical nature of optical gear, and potential disintermediation by hyperscalers' custom silicon. They warn of valuation traps and risk of multiple compression if growth disappoints.
Risk: Acceleration of custom silicon (ASICs) adoption by hyperscalers, potentially leading to margin compression and disintermediation of hardware supply chain.
Opportunity: Potential margin relief for SMCI and MU if GPU supply eases in H2 2025, allowing for a mix shift.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
What happened: AI infrastructure stocks jumped on Wednesday after blowout results from AI server hardware maker Supermicro (SMCI) and neocloud providers CoreWeave (CRWV) and Nebius Group (NBIS ). AI hosting peers Applied Digital (APLD) and IREN (IREN) moved higher.
Other areas of the AI infrastructure trade also gained, including optical product maker Lumentum (LITE), and the memory and storage complex trade also gained with the Roundhill Memory ETF (DRAM).
What's behind the move: Signs of accelerating demand for everything from computing power to AI server racks to liquid cooling technology sent the entire AI infrastructure complex higher as companies within the space posted their results.
Nebius stock surged after the company posted quarterly revenue that topped analyst expectations. Supermicro reported fourth quarter results that beat analyst expectations on earnings . The company's upbeat forecast sent shares higher by more than 6%.
Neocloud and AI hosting players Applied Digital (APLD) and IREN (IREN) also moved higher after artificial intelligence cloud provider CoreWeave (CRWV) posted quarterly results highlighting accelerating demand and a surging backlog.
And Lumentum (LITE) shares gained after the maker of optical and photonic products for data centers posted fiscal fourth quarter revenue more than doubled to $1.01 billion. Peers Coherent (COHR) and Ciena (CIEN) both jumped.
The memory and storage complex also rose with the Roundhill Memory ETF (DRAM) up 4%, while high-flyer SanDisk (SNDK) rose 4%.
Memory maker Micron (MU) jumped 4%, along with rival SK Hynix (SKHY).
What else you need to know: As companies adopt AI, investors have been backing the "picks and shovels" trade, with companies forecasting strong demand for the infrastructure needed to support the technology.
Fundstrat noted Wednesday that, among S&P 500 companies that have reported this earnings season, 87% have beaten estimates, with those that "beat" exceeding expectations by a median of 7%.
Ines Ferre is a senior business reporter for Yahoo Finance. Follow her on X at @ines_ferre.
Click here for in-depth analysis of the latest stock market news and events moving stock prices
Read the latest financial and business news from Yahoo Finance
Four leading AI models discuss this article
"Earnings beats confirm demand but are already largely priced; near-term upside looks limited without fresh hyperscaler capex surprises."
The surge in AI infrastructure names (SMCI +6%, LITE, MU, DRAM ETF +4%) after CoreWeave, Supermicro, and Nebius beats reflects genuine acceleration in GPU-cluster demand, backlog growth, and optical/memory spending. Lumentum’s revenue doubling to $1.01B is particularly concrete. However, the move feels like a momentum extension rather than fresh information; 87% S&P beat rate is broad, not AI-specific. Valuations for SMCI (~45x forward) and several neoclouds already price in multiple years of 50%+ growth. Missing context: actual GPU supply constraints from Nvidia remain tight, and hyperscaler capex guidance has been lumpy.
Most of these beats were already telegraphed in prior channel checks; the real risk is that AI ROI for enterprises stays elusive, causing 2025 capex digestion and order slippage that would hit SMCI, CoreWeave, and memory names hardest.
"The current rally ignores the unsustainable margin pressure caused by the extreme capital intensity required to scale AI hardware production."
The market is conflating 'demand' with 'profitability' in the AI infrastructure space. While SMCI and LITE are posting massive revenue growth, the underlying issue is margin compression due to the high cost of HBM (High Bandwidth Memory) and liquid cooling integration. Investors are cheering top-line beats, but they are ignoring the capital intensity required to maintain this trajectory. The surge in names like APLD and IREN suggests a speculative 'rising tide' mentality rather than a focus on sustainable cash flow. Unless these firms demonstrate operating leverage—meaning revenue growth significantly outpacing operational expenses—this rally is a valuation trap fueled by unsustainable CapEx cycles from hyperscalers.
If the AI infrastructure build-out is truly a generational shift comparable to the electrification of the grid, current valuations may be discounting the massive long-term recurring revenue potential of these data center assets.
"Infrastructure stocks are pricing in multi-year capex cycles and double-digit growth, but historical precedent and valuation (SMCI at 40x forward P/E) suggest re-rating risk if demand normalizes or pulls forward into 2024."
The article conflates earnings beats with sustainable demand. Yes, 87% of S&P 500 reporters beat—but that's survivorship bias; weak players already reported or guided down. CoreWeave's 'surging backlog' is real, but backlogs aren't revenue. Supermicro's 6% pop on a beat is modest; the stock trades 40x forward earnings (vs. 19% consensus growth). Lumentum's revenue doubled, but optical gear is cyclical—data center capex cycles have historically compressed 18-24 months after peaks. Memory (MU, SKHY) up 4% on hope, not fundamentals; DRAM spot pricing remains under pressure. The article ignores: (1) whether these beats reflect pull-forward demand ahead of potential AI capex slowdown, (2) valuation compression risk if growth disappoints, (3) geopolitical chip export restrictions tightening.
If AI adoption accelerates faster than consensus models and capex cycles extend 3+ years instead of the historical 2-year pattern, infrastructure stocks could sustain 25%+ EPS growth and justify current multiples. CoreWeave's backlog could prove sticky.
"Near-term upside hinges on durable demand, backlog conversion, and margin resilience; otherwise the rally risks a sharp pullback."
News confirms a short-term rally in AI infrastructure names, but the punch bowl may not last. The gains look like a relief rally built on one-quarter beats, improving backlog, and optimism about hyperscale capex, not a durable earnings plateau. Valuations across listed suppliers and hosting players remain sensitive to discrete project pipelines, warranty, and margin pressures as components like memory and cooling costs rise. Macro risk persists: AI demand could decelerate if cloud budgets tighten or competitors flood the space with cheaper architectures. A few high-profile beaters can’t erase the risk that this cycle is still highly cyclical and lumpy.
Bull case: AI infrastructure demand is secular and hyperscalers will continue expanding data-center capacity, so a durable capex cycle could sustain higher valuations. If backlog proves durable, the upside may extend beyond current expectations.
"Easing GPU constraint in 2025 could drive margin expansion even if revenue growth moderates."
Claude correctly flags optical cyclicality and backlog-to-revenue translation risk, yet underweights the second-order effect: if GPU supply eases in H2 2025 precisely as CoreWeave and hyperscalers digest 2024 orders, memory and server margins could expand sharply on mix shift. That margin relief, not just top-line, is what SMCI and MU need to avoid multiple compression.
"Hyperscaler adoption of custom silicon poses a structural margin threat to SMCI that cyclical supply easing won't offset."
Grok is overly optimistic about margin expansion. Even if GPU supply eases, the structural shift toward custom silicon (ASICs) by hyperscalers like Google and Amazon creates a massive headwind for SMCI’s standard server margins. You cannot assume a 'mix shift' will save margins when the largest customers are actively trying to bypass the very rack-level integration SMCI sells. The risk isn't just cyclical; it's a fundamental disintermediation of the hardware supply chain that everyone is ignoring.
"ASIC risk is real but the speed of adoption relative to backlog duration determines whether it's a 2025 headwind or a 2027 structural cliff."
Gemini's ASIC disintermediation thesis is the sharpest risk here, but it's incomplete. Google TPUs and Amazon Trainium exist—yet SMCI still grew 200%+ YoY. Why? Because custom silicon takes 2-3 years to scale, and hyperscalers still need standard servers for non-ML workloads. The real question: does ASIC adoption accelerate faster than historical semiconductor cycles? If yes, Gemini wins. If no, SMCI's backlog survives the transition. Nobody's quantified the timeline.
"ASIC disintermediation risk is real and could compress SMCI margins sooner than expected, even as backlog and non-ML workloads cushion near-term revenue."
Gemini's ASIC disintermediation risk is valid but incomplete. If hyperscalers accelerate custom silicon, SMCI's server margins could compress faster than expected, not only from cycles but from a shift to OEM/custom builds. Yet backlog visibility and non-ML workloads cushion near-term revenue. The margin outcome hinges on AI workloads expanding without a full-stack replacement; timing remains uncertain, but 2–3 years of scale feels plausible.
The panel is bearish on the current AI infrastructure rally, citing high valuations, unsustainable capital intensity, cyclical nature of optical gear, and potential disintermediation by hyperscalers' custom silicon. They warn of valuation traps and risk of multiple compression if growth disappoints.
Potential margin relief for SMCI and MU if GPU supply eases in H2 2025, allowing for a mix shift.
Acceleration of custom silicon (ASICs) adoption by hyperscalers, potentially leading to margin compression and disintermediation of hardware supply chain.