NextFin News - Nvidia reported fiscal Q2 2027 revenue of $96.22 billion, beating the consensus of $92 billion by a record $4.06 billion. Adjusted EPS came in at $2.22, up 111% year over year. Data Center revenue hit $89.02 billion, accelerating to 116.6% year-over-year growth, the fastest rate in four quarters. Q3 guidance was set at $108 billion, plus or minus 2%, against the $104.2 billion consensus.
The stock initially dropped to $206 in afterhours trading, extending its seven-day losing streak. Then the earnings call started. Jensen Huang said Nvidia had achieved AGI for many tasks, that demand is running higher than supply through fiscal 2028, and that next year will be "pretty extraordinary." CFO Colette Kress addressed the circular financing criticism directly. The stock reversed, surged to a high of $220.80, and settled around $218.10, up 4.03% afterhours. It was the first post-earnings rally in four quarters, ending the sell-the-news streak simultaneously with the seven-day price decline streak.
The Numbers Behind the Beat
Revenue of $96.22 billion against a $92 billion consensus is the largest dollar beat in Nvidia's earnings history. YoY revenue growth accelerated more than 20 percentage points to 105.9%, a fact that almost no company at this revenue scale has delivered. Data Center at $89.02 billion now represents 92% of total revenue, with the segment's $13.78 billion sequential dollar increase the single largest quarterly revenue increment Nvidia has ever posted.
The gross margin trajectory was the one item that caused the initial selloff. Q2 GAAP gross margin was 75.0%, down slightly from Q1. Kress said on the call that Q3 gross margins are expected at 74%, plus or minus 50 basis points, before bottoming in Q4 at 71% to 72%, and settling at 72% to 73% in fiscal 2028 as Nvidia's own price increases take effect in Q1 of that year. The cause was stated explicitly: memory pricing has become "extreme," with cost increases exceeding prior expectations and heading higher into next year. The magnitude of DRAM price inflation, which TrendForce projected at 260% for 2026 in server DRAM, is compressing Nvidia's unit economics even as its revenue doubles.
Huang addressed this directly: "We've got supply for about 70% growth. Our demand is much higher than that." The unconstrained demand figure, if supply were not the binding constraint, would imply revenue well above the $108 billion Q3 guide.
The Memory Statement That Changes the MU and SKHY Thesis
The most consequential disclosure of the evening for investors in Micron and SK Hynix was not the revenue beat. It was Kress confirming that memory supply "will remain a bottleneck at least through the end of fiscal year 2028," and that pricing is "extreme" and "headed even higher into next year."
This is Nvidia telling the market, with the credibility of a company that spent $145 billion on component supply in a single quarter, that the memory scarcity that has driven Micron to 84.9% gross margins and SK Hynix to 76% operating margins is not a 2026 phenomenon. It is a 2027 and likely 2028 phenomenon. The argument that Micron and SK Hynix are cheap at 5 to 6 times forward earnings was already analytically strange given their contracted revenue. With Nvidia's CFO confirming extreme pricing conditions extending multiple years, the case that memory stocks deserve commodity multiples becomes harder to sustain.
If Nvidia is experiencing extreme memory costs that exceed its own prior expectations, and it is raising prices in response, the pricing power is flowing to the memory suppliers. The revenue opportunity per gigawatt of AI data center has risen from $18 billion for Hopper to $25 billion for Grace Blackwell to $40 billion for Vera Rubin. Each generational step requires more memory content. The Vera Rubin platform will ship HBM4. Micron already has HBM4 in high-volume production. The memory demand the next generation of Nvidia compute requires is already locked into the supply agreements that Micron and SK Hynix have disclosed.
Jensen Huang's Three Biggest Statements
The earnings call contained several remarks that will be analyzed and debated for weeks.
On AGI: "For many tasks, we could say that we've already achieved AGI." This is the most direct statement from a major AI figure about artificial general intelligence. Huang was careful to say "for many tasks," not universally, which aligns with the technical reality that current AI systems match or exceed human performance on specific domains while remaining far below human capability on open-ended reasoning. The commercial implication of even partial AGI is that the value of compute expands further, because capable AI systems generate revenue that justifies more infrastructure investment.
On agentic compute demand: "The amount of compute necessary for an agent versus a human using it is probably 15 to 100 times, depending on the type of problem you're trying to solve." Salesforce just reported AI and Data ARR of $3.9 billion growing 210% year over year, driven by Agentforce. If agent-based AI requires 15 to 100 times more compute than human-directed usage, and enterprise adoption of agents is accelerating at 210% annually, the demand trajectory for AI compute infrastructure is not plateauing. It is compounding on a compounding base.
On the revenue opportunity per gigawatt of AI data center capacity, Huang laid out the most concrete measure of why each hardware generation matters beyond raw compute performance. Hopper generated approximately $18 billion of revenue opportunity per gigawatt. Grace Blackwell raises that to approximately $25 billion. Vera Rubin reaches approximately $40 billion. The total capital required to build one gigawatt of AI data center capacity has simultaneously risen from $30 billion five years ago to around $60 billion today. The implication is that the revenue density of AI infrastructure is outpacing the cost density, and each new platform generation widens that gap further. For Nvidia, this means each gigawatt deployed on Vera Rubin generates more than twice the revenue of a gigawatt deployed on Hopper, without requiring proportionally more chip volume.
On the circular financing critique: Kress said Nvidia has invested nearly $50 billion in frontier AI labs. Some characterize this as circular, where Nvidia finances the labs that buy Nvidia chips. Kress acknowledged the critique directly: "We recognize the scale of this support, and we know some will call this circular financing. We see it differently." Her counterargument was the generational nature of the technology shift. Whether one accepts the circular critique or Nvidia's response to it, the disclosure confirms that Nvidia's relationship with its largest customers is more complex than a simple supplier arrangement.
What This Means for the AI Infrastructure Complex
Nvidia's quarter is a read-through for every company positioned in the AI buildout. The $108 billion Q3 guide, excluding China, is the demand signal that validates the hyperscaler capex announcements that drove earlier earnings reactions. Alphabet's $195 to $205 billion 2026 capex, Meta's AI infrastructure acceleration, and Microsoft's Azure buildout all produce Nvidia revenue. The $108 billion quarter these investments are funding is the proof that the capex is translating to chip orders at scale.
For the memory sector, the read-through is unambiguous. Extreme memory pricing that exceeds Nvidia's prior expectations and is trending higher into next year is a direct margin transfer from Nvidia to its memory suppliers. Micron, SK Hynix, and Samsung are pricing their HBM products at levels that are compressing Nvidia's gross margins even as they contribute to Nvidia's record revenue. The memory stocks that have been recovering from their June-July correction are now benchmarked against a supplier that has publicly confirmed their pricing power extends through fiscal 2028.
For AMD, the competitive question Huang addressed is whether custom silicon from frontier labs changes Nvidia's trajectory. His response was 100% confidence that frontier labs will remain Nvidia customers for a "quite long period of time" regardless of their own chip development efforts. The platform argument, that Nvidia spans the entire AI lifecycle across every cloud while custom XPUs are inference-specific solutions for a single deployment environment, is the commercial moat he articulates. Whether that moat is as wide as he claims will be tested as Google's TPUs, Amazon's Trainium, and OpenAI's rumored silicon reach production scale.
The most precise summary of what the call accomplished came from Jensen Huang himself: "Compute is revenue." The infrastructure buildout has crossed the threshold where AI-generated output is economically valuable enough to justify ongoing expansion of the compute base that produces it. If that statement is accurate, the demand trajectory for the companies positioned at the compute supply layer has not peaked. It has reached the inflection point where self-reinforcing economics take hold.
Explore more exclusive insights at nextfin.ai.
