“Demand has gone parabolic. The reason is simple: agentic AI has arrived.” — Jensen Huang. A clean beat on every metric, but the more important story is structural: five compounding shifts are extending the multi-year runway.
The Bottom Line
NVIDIA delivered a clean operational beat on every line — revenue of $81.6B beat consensus of $78.8B by $2.8B, Q2 FY27 revenue guidance of $91.0B ± 2% came in roughly $5B above the Street, and the company announced a 25× dividend hike alongside an additional $80B share repurchase authorization. The headline numbers, however, are no longer the most interesting part of the story.
Five structural shifts in this print materially strengthen the long-term thesis. First, Blackwell continues to ramp without cannibalising Hopper, and this directly de-risks the upcoming Rubin transition. Second, Data Center networking stepped up to $14.8B, +199% YoY, a standalone business of generational scale. Third, the Hyperscale / non-Hyperscale split inside Data Center is already at ~50/50 — the diversification de-risk is reported fact, not forecast. Fourth, Vera, NVIDIA’s purpose-built CPU, opens a brand-new $200B TAM with ~$20B of standalone CPU revenue this year alone. Fifth, two supporting data points reinforce the durability of demand: H100 rental rates are up ~20% YTD and A100 ~15%, and total supply commitments jumped to $145B.
Headline Financials vs. Consensus
| Metric | Q1 FY27 | YoY | QoQ | Consensus | Beat |
|---|---|---|---|---|---|
| Revenue | $81.6B | +85% | +20% | $78.8B | +$2.8B |
| Non-GAAP EPS | $1.87 | +140% | +18% | $1.77 | +5.6% |
| GAAP EPS | $2.39 | +214% | +36% | — | — |
| Non-GAAP Gross Margin | 75.0% | +1,420 bps | (10 bps) | ~74.5% | Beat |
| Operating Cash Flow | $50.3B | +84% | +39% | — | — |
| Free Cash Flow | $48.6B | +86% | +39% | — | Record |
| Capital Returned to Shareholders | ~$20.0B | Record | — | — | — |
| Q2 FY27 Revenue Guidance | $91.0B ± 2% | — | +11% | ~$86B | +~6% |
Revenue by Market Platform — new reporting framework
| Platform / Sub-segment | Q1 FY27 | YoY | QoQ | % of Total | Read |
|---|---|---|---|---|---|
| Data Center — Total | $75.2B | +92% | +21% | 92% | Blackwell 300 ramp; InfiniBand, Spectrum-X, NVLink |
| Hyperscale | $37.9B | +115% | +12% | 46% | Top hyperscalers; ~50% of Data Center |
| ACIE (AI Clouds, Industrial, Enterprise) | $37.4B | +74% | +31% | 46% | AI clouds >3× YoY; sovereigns scaling |
| Edge Computing | $6.4B | +29% | +10% | 8% | PCs, workstations, automotive, robotics, AI-RAN |
| DC Networking (prior framework) | $14.8B | +199% | +35% | — | Inflection — see Takeaway 2 |
| DC Compute (prior framework) | $60.4B | +77% | +18% | — | Blackwell majority of revenue mix |
Five Structural Takeaways for Investors
Data Center revenue grew +21% sequentially to a record $75.2B, with Blackwell now “the majority of our revenue” — yet gross margin held essentially flat at 75.0%. Customers deploy Hopper, Blackwell, and Blackwell 300 concurrently across workload tiers rather than switching off prior generations.
This is the most important read-through for the upcoming Rubin transition. A year ago, investors worried that Blackwell’s launch would crater Hopper demand — the classic semiconductor air-pocket. That concern proved unfounded. With Vera Rubin shipping in H2 FY27, the same worry is now being raised about Blackwell. The Hopper-to-Blackwell experience strongly suggests the Blackwell-to-Rubin transition will play out the same way. This is a concern that can now be put to rest, not just deferred.
Under the prior reporting framework, Data Center networking revenue was a record $14.8B this quarter, up 199% YoY and 35% sequentially. That is a roughly $60B annualised run-rate at the current pace — and growing. Spectrum-X Ethernet, NVLink, and InfiniBand were all called out as primary drivers.
On a standalone basis, applying multiples comparable to Arista’s or Broadcom’s networking franchises, this business alone would likely command a valuation in the several hundred billion dollar range. NVLink Fusion — which allows third-party silicon to participate in the NVIDIA fabric — repositions networking as the substrate that endures even if compute share shifts.
This is the most important structural disclosure of the quarter. Of the $75.2B in Data Center revenue, Hyperscale contributed $37.9B (~50%) and ACIE — AI Clouds, Industrial & Enterprise — contributed $37.4B (the other 50%). Management explicitly framed the remaining 50% as “continued diversification of customers, including AI Clouds, industrial, enterprise, and sovereign customers.”
A year ago, hyperscalers were a far larger share. The fact that ACIE has scaled to parity in 12 months — with AI cloud revenue more than tripling YoY — directly addresses the long-standing concern that a hyperscaler CapEx pause could disproportionately impair the business. The diversification story is no longer a thesis; it is reported fact.
CFO Colette Kress disclosed that Vera — NVIDIA’s first custom Arm-based CPU, purpose-built for agentic AI — opens a $200B TAM the company has never addressed before, with visibility to ~$20B of CPU revenue this fiscal year. Jensen explicitly clarified: this $20B figure is for standalone Vera CPU only — it excludes Vera that ships as part of Vera Rubin systems.
The economic profile is structurally attractive: ~1.5× performance per core, 2× per watt, and 4× rack density versus x86. Given NVIDIA’s track record (Grace, BlueField DPUs each became multi-billion-dollar businesses within 2–3 years), capturing meaningful share of that $200B TAM over 3–5 years is realistic — and this revenue stream is almost entirely unmodelled by sell-side estimates today.
Older-generation GPU rental rates are still rising. Management disclosed H100 rental rates up ~20% YTD and A100 up ~15%, stating that “customers are generating profitable revenue beyond the depreciable life of their GPUs.” This answers the useful-life concern directly: these chips are likely productive for materially longer than the 4–5 year depreciation periods many hyperscalers already use. Once fully depreciated, these older GPUs become exceptionally profitable — strengthening customer ROI and supporting continued NVIDIA purchasing.
Supply commitments stepped up materially. Total supply, inclusive of inventory, purchase commitments, and prepaids, rose to $145B (vs. $119B narrower SEC-filing definition). Inventory alone rose to $25.8B (from $21.4B), and multi-year cloud service commitments increased to $30B (from $27B). These are contractual commitments NVIDIA has made in response to confirmed customer orders — a clear forward-looking indication topline is set to scale further.
Implications for the Earnings Trajectory
Pulling these threads together, the FY27 earnings trajectory is being driven by five compounding vectors operating in parallel. The Hopper-to-Blackwell experience materially de-risks the upcoming Rubin transition. Networking has stepped up to a $60B+ annualised run-rate. The 50/50 Hyperscale / ACIE split is reported reality. The Vera CPU launch creates a ~$20B revenue line against a $200B TAM. And legacy GPU rental rates rising 15–20% YTD, alongside total supply commitments stepping up to $145B, are two independent confirmations that demand is durable and contracted volumes are scaling further.
Free cash flow of $48.6B in a single quarter — and roughly $20B already returned to shareholders via buybacks and dividends — confirms the cash conversion engine is structurally intact. The Board’s decision to raise the quarterly dividend 25× and authorise $80B in additional buybacks is the clearest possible signal of confidence in cash durability. For long-term holders, this quarter strengthens rather than challenges the multi-year investment case.
Key Risks to Monitor
AMD’s MI-series accelerators continue to improve, while the largest customers scale their own in-house silicon — Amazon’s Trainium and Google’s TPU programmes — disintermediating NVIDIA in inference and select training workloads at the very accounts that drive Hyperscale revenue.
At this scale, execution is the gating factor. Sustaining growth requires flawless delivery across chip design, rack-scale integration, software, and a vast supply chain (TSMC, HBM, advanced packaging). Slippage on Rubin or networking could disrupt the trajectory.
Hyperscaler CapEx eventually requires AI revenue proof-points to sustain. Diversification has materially de-risked the picture, but Hyperscale still drives 50% of Data Center revenue and CapEx decisions remain correlated.
.png)
.webp)
