SMCI Beats by 78%: $65B–$72B Guide Reshapes AI Hardware
Super Micro Computer's fiscal Q4 EPS of $1.70 obliterated the $0.96 consensus. Here's what the $60B order book means for AI hardware traders.
Key Points
- SMCI posted non-GAAP EPS of $1.70 against a $0.96 consensus — a 78% beat — on $11.12 billion in revenue, up 93.2% year over year.
- More than $60 billion in new orders booked during fiscal 2026 drove a guidance reset to $65B–$72B for FY27, obliterating the Street's prior $53.29B estimate by as much as 35%.
- With NVIDIA's own print due in late August and SMCI volume running at nearly double its 65-million-share average today, the next inflection point for the entire AI hardware complex arrives in roughly two weeks.
Super Micro Computer just handed the AI hardware bull case its most powerful data point of the year. Fiscal Q4 non-GAAP EPS came in at $1.70 against a $0.96 consensus — a 78% beat — on revenue of $11.12 billion, up 93.2% year over year, with GAAP gross margin expanding a stunning 800 basis points to 17.5% from 9.5% a year ago. The company booked more than $60 billion in new orders during fiscal 2026 and guided FY27 revenue to $65B–$72B, blowing past the Street's prior estimate of $53.29B by up to $18.71 billion.
The Numbers Behind the Beat
The margin story is what separates this print from a simple top-line beat. SMCI's GAAP gross margin moving from 9.5% to 17.5% in a single year is not a rounding error — it signals that the company is gaining pricing power in AI server configurations at the exact moment hyperscaler demand is accelerating. Revenue of $11.12 billion represents a run rate that, annualized, puts SMCI above $44 billion — yet management's $65B–$72B FY27 guide implies they expect to roughly 50% to 63% that figure again in the next twelve months. That is not conservative sandbagging; that is a company telling you its backlog visibility is genuinely exceptional.
The $60 billion in new orders booked during fiscal 2026 is the number traders should anchor to. Order books of that magnitude — for a company this size — imply customers are committing capital 12 to 18 months forward, which means the hyperscalers and enterprise AI buildout teams have not blinked despite a 10-year Treasury yield sitting at 4.68% and real financing costs that are genuinely restrictive by post-2010 standards. SMCI's customers are paying up because the alternative — falling behind in AI infrastructure — is more expensive than the cost of capital.
Today's price action reflects the tension between euphoria and profit-taking that always follows a parabolic beat. SMCI is currently trading at $39.53, in a session range of $37.80 to $42.30 — a $4.50 intraday spread that tells you this stock is being repriced in real time, not drifting. Volume has hit 122.78 million shares against a 65.04 million average, meaning nearly twice the normal float has changed hands. The -6.5% gap from the session peak to the current print is characteristic of large-cap momentum names absorbing institutional distribution after an overnight surge. The $37.80 intraday low is now the first technical level of interest; a close above it keeps the bull structure intact.
The Sector Read-Through Is Already Priced In — Partially
Wednesday's SMCI-driven rally lifted AMD 1.8% to $483, INTC 3.3% to $101, and NVDA 3.0% to $224. Those moves were real, but they were also fast, and experienced traders know that a one-day sympathy rally and a structural re-rating are very different things. The question now is whether those gains hold or whether they were front-running NVIDIA's own print, due in late August — and whether NVIDIA can deliver a number that justifies the premium multiple the stock already carries.
AMD's position is worth examining separately. Q2 Data Center revenue hit $6.72 billion, up 107% year over year, and now represents 58% of AMD's total revenue. Management guided Q3 to roughly $13 billion. At 27–28x forward earnings, AMD trades at a meaningful discount to Nvidia's premium, which makes it the more asymmetric bet for traders who believe the AI infrastructure cycle has another 12 to 18 months of runway — particularly if SMCI's $65B–$72B guide proves accurate, since AMD MI350 GPUs are a core component of SMCI server configurations. Intel's data point is less clean — $6.26 billion in Data Center and AI revenue, up 59% YoY, is strong on a percentage basis but the absolute dollar figure trails AMD and Nvidia by a wide margin. INTC's recovery story is still a restructuring trade, not a pure AI leverage play.
The TSMC data point that threads all of this together: CEO C.C. Wei's statement that U.S. production capacity is "still not enough, far from enough" is not marketing language — it is a supply constraint admission from the company that controls 72% of the global chip foundry market. TSMC and Amkor Technology announced a long-term advanced packaging partnership in the United States in June, but partnerships don't spin up fabs overnight. The physical bottleneck in advanced packaging and leading-edge node production is the binding constraint on how fast SMCI, AMD, and Nvidia can actually ship against their order books. That constraint supports pricing power across the stack.
What Traders Watch Next
Google's Gemini crossing 1 billion monthly active users — announced August 11, making it the company's 14th product to hit that milestone — adds a demand-side dimension to the AI hardware story that is easy to underweight. Gemini generating 150 million-plus images daily and driving 63% of users to engage voice features means the inference compute load is scaling faster than most supply-side models projected. More MAUs running heavier workloads means more GPU hours purchased from hyperscalers, which flows directly back to SMCI, AMD, and Nvidia server demand. The software adoption curve and the hardware capex cycle are now feeding each other.
Meta's release of Muse Glimmer — a 30-billion-parameter dense multimodal model under Apache 2.0 — matters for a different reason. Running under 20GB on a single consumer GPU with a 3.1x speedup and 100-plus language support, it represents the open-source competitive pressure on proprietary cloud AI services. If enterprise customers can run capable agentic models locally, the inference revenue story for hyperscalers gets more complicated at the margin, even as the training and fine-tuning infrastructure buildout continues.
The single most important date on the calendar right now is NVIDIA's late-August earnings print. NVIDIA's most recent quarter posted Data Center revenue of $75.25 billion, up 92% year over year, and Jensen Huang described the current buildout as "the largest infrastructure expansion in human history." Against that backdrop, Nvidia is also executing a $20 billion multi-tranche bond sale — funding mode for a company that believes its own forward demand is real. If NVIDIA guides FY2027 Data Center revenue above $320 billion — the implied run rate if the 92% YoY growth pace holds even partially — the entire AI hardware complex gets another leg. Watch $224 on NVDA as the near-term technical anchor; a close above that level before the print sets up a pre-earnings drift trade that has worked in four of the last five NVIDIA reporting cycles.
The Weekly Investor
Daily market analysis for active traders. Free.
