KEY POINTS
- Nvidia posted Q1 FY27 revenue of $81.6 billion, up 85% year over year, with data center revenue hitting a record $75.2 billion on 92% annual growth.
- Despite a parabolic beat, NVDA shares slipped roughly 0.9% Thursday because buy-side expectations had crept above the $91 billion Q2 guide and margin guidance went flat.
- Computex next week and the September-quarter Blackwell 300 shipment cadence are the next two catalysts that will reset the multiple.
Nvidia posted $81.6 billion in revenue and a record $75.2 billion data-center quarter on Wednesday evening, then watched its stock slip 0.9% by Thursday's close as Wall Street registered the rarest of Nvidia trades: a beat that did not move the needle. Revenue rose 85% year over year and 20% sequentially. Data-center revenue alone grew 92% from a year ago. GAAP gross margin held at 74.9%. CEO Jensen Huang told investors demand "has gone parabolic" and that "agentic AI has arrived." The numbers were better than almost any bull had been willing to model six months ago. They were not better than what the most aggressive buyers had already paid for.
The Beat That Wasn't Big Enough
The whisper number on Wall Street had crept above $80 billion into earnings, and Nvidia's guide of $91 billion for the current quarter still came in roughly $2 to $3 billion above sell-side consensus. That should have been enough. It wasn't, because the buy side had begun penciling in $93 to $95 billion. The chip-cycle math has moved that fast. Combined capex from Microsoft, Alphabet, Amazon, and Meta is projected at $649 billion in 2026, up from $411 billion last year, with roughly 75% of that earmarked for AI infrastructure. Nvidia sits at the inflow point of nearly every one of those dollars.
The stock reaction reflected positioning more than fundamentals. NVDA was up roughly 8% in the week heading into the print and 15% year to date, modest by Nvidia standards but rich enough that the trade-around crowd was looking for a reason to lock gains. The reason came in the form of margins. The 74.9% GAAP gross margin was excellent, but it was flat sequentially, and guidance implied no further expansion this year as Blackwell 300 ramps and yield improvements get reinvested into pricing flexibility for hyperscalers. The bears latched onto the word "flat." For a stock priced for parabolas, flat is a problem. The full Q1 FY27 press release on the SEC site shows the underlying numbers in detail.
Where The Growth Is Actually Coming From
The most important data point in the release was buried inside the data-center segmentation. Hyperscalers — Microsoft, Amazon, Google, Meta, and Oracle — held steady at roughly 50% of Nvidia's data-center revenue. The other 50% came from what Huang called "diversified customers," including AI cloud specialists like CoreWeave and Lambda, industrial buyers building agentic factories, enterprise customers standing up private inference clusters, and sovereign AI deployments across the Middle East, Europe, and Asia. That split matters because the bear case on Nvidia has long centered on hyperscaler concentration risk. If three or four buyers can rationalize spending, the whole revenue stack wobbles. That risk is dissolving in real time. The Blackwell 300 ramp is being absorbed by a customer base broad enough to muffle any single hyperscaler's pause.
Networking is the second story inside the story. InfiniBand, Spectrum-X Ethernet, and NVLink revenue grew faster than GPU revenue inside the data-center segment, a structural tell about what Nvidia is actually selling. Customers aren't buying chips. They are buying full rack-scale systems where the interconnect determines training throughput. That moat is harder to disrupt than the GPU moat, because competitors would have to displace an entire networking stack to win share. AMD's MI400 series will compete on GPU performance. It will not compete on rack-scale system economics in 2026. The CNBC earnings recap details the segment-level breakdown.
What Comes Next
Two events define the next leg. First is Computex next week in Taipei, where AMD's Lisa Su arrived ahead of Huang on May 20 in a clear signal that AMD intends to use the show to press its case on MI400 and on Instinct rack-scale designs. Huang's keynote will be parsed for any commentary on the Rubin R100 and R200 timeline. Rubin is built on TSMC's 3-nanometer process and specced to carry 288 gigabytes of HBM4 memory per chip across eight stacks. Samsung is in active discussions to supply HBM4 to Nvidia, and SK Hynix has finished mass-production preparations. The memory supply chain is the constraint, not the compute side, and any Rubin commentary that hints at HBM availability will move both Nvidia and the memory names.
Second is the September quarter print, where the buy side will be looking for the first material step-up in Blackwell 300 shipments and the first commentary on Rubin volume timing. The guidance of $91 billion for the July quarter assumes a Blackwell 300 mix shift already in motion. Anything north of $95 billion in the October print would reset the multiple. Anything south of $93 billion would reset positioning.
Nvidia trades around 33 times forward earnings as of Thursday's close, well below its five-year average near 47. That is the cheapest the stock has been on a forward-multiple basis since the start of the AI cycle. The fundamentals support that compression. The expectations problem does not. Watch the $1,400 level on NVDA into Computex. A close above on Huang's keynote keeps the broadening AI trade intact. A failure there hands the leadership baton to AMD, Broadcom, and the memory names for the rest of the quarter.
