What Management Said
Read the full Q2 2026 transcript ↗We delivered another record quarter while navigating what continues to be a dynamic external environment. Total revenue was $46.7 billion, exceeded our outlook as we grew sequentially across all market platforms. Data center revenue also grew sequentially despite the $4 billion decline in H20 revenue. Our full-stack AI solutions for cloud service providers, NeoClouds, enterprises, and sovereigns are all contributing to our growth.
The scale and scope of these buildouts present significant long-term growth opportunities for NVIDIA. The new Blackwell Ultra platform has also had a strong quarter, generating tens of billions in revenue. This keeps us on track with our pace of an annual product cadence and continuous innovation across compute, networking, systems, and software. USG officials have expressed an expectation that the USG will receive 15% of the revenue generated from licensed H20 sales, but to date, the USG has not published a regulation codifying such requirement.
We have not included H20 in our Q3 outlook as we continue to work through geopolitical issues. If geopolitical issues reside, we should ship $2 billion-$5 billion in H20 revenue in Q3. Notably in the quarter was an increase in H100 and H200 shipments. As we continue to deliver both Hopper and Blackwell GPUs, we are focusing on meeting the soaring global demand.
What went well
- Record total revenue of $46.7 billion, exceeding outlook, with sequential growth across all market platforms.
- Data center revenue grew 56% year-over-year and grew sequentially despite a $4 billion decline in H20 revenue.
- Blackwell platform reached record levels, growing 17% sequentially; GB300 (Blackwell Ultra) entered production shipments and generated tens of billions in revenue.
- GB300 rack transition was seamless; production run rate returned to full speed at approximately 1,000 racks per week, expected to accelerate through Q3.
- Record networking revenue of $7.3 billion, up 46% sequentially and 98% year-over-year, with strength across Spectrum-X Ethernet, InfiniBand and NVLink.
- Spectrum-X Ethernet annualized revenue exceeded $10 billion and InfiniBand revenue nearly doubled sequentially on XDR adoption.
- On track for over $20 billion in sovereign AI revenue this year, more than double last year, with major EU and UK initiatives.
- RTX Pro servers in full production with nearly 90 companies adopting; Rubin platform's six chips are in fab and on schedule for volume production next year.
What went wrong
- Data center revenue absorbed a $4 billion sequential decline in H20 revenue tied to China export restrictions.
- No H20 was shipped against the licenses received in late July, and H20 was excluded from the Q3 outlook due to unresolved geopolitical issues.
- USG officials expressed an expectation of receiving 15% of revenue from licensed H20 sales, though no codifying regulation had been published.
- China data center opportunity (estimated ~$50 billion this year) remained largely unaddressed pending approval to ship competitive products such as Blackwell.
Guidance Changes
| Metric | Period | Current guidance |
|---|---|---|
| China H20 revenue | Q3 FY2026 | $2B-$5B possible if geopolitical issues resolve; can build/ship more if more orders/licenses arrive |
| Rubin platform volume production | 2H FY2027 (next year) | on schedule for volume production next year with a mature, full-scale supply chain |
| Data center growth driver (Blackwell vs Hopper) | Q3 FY2026 | Blackwell expected to remain the lion's share and the driver of growth into Q3 |
Performance Breakdown
| Metric | YoY | Note |
|---|---|---|
| Total revenue | n/a (record $46.7B, sequential growth across all platforms) | Exceeded outlook on broad-based strength led by data center and networking. |
| Data center revenue | +56% | Blackwell record levels (+17% sequential) and GB300 production shipments, overcoming a $4B sequential H20 decline. |
| Networking revenue | +98% | Record $7.3B (+46% sequential) on Spectrum-X Ethernet, InfiniBand and NVLink demand for AI compute clusters. |
Earnings Call Themes & Trends
| Topic | Previous mention | Current period | Trend |
|---|---|---|---|
| AI infrastructure TAM | ~$1 trillion path outlined at prior GTC | $3-4 trillion AI infrastructure spend by the end of the decade; ~$600 billion hyperscaler CapEx this calendar year | — |
| Blackwell / GB300 ramp | GB200 NVL widespread adoption | GB300 in production, tens of billions in Blackwell Ultra revenue, ~1,000 racks/week run rate | — |
| China H20 | $4.6B recognized in Q1, market closed | select licenses received but nothing shipped; $2-5B possible in Q3; advocating for Blackwell in China | — |
| Networking scale-up/out/across | SpectrumX ~$8B annualized in Q1 | Spectrum-X over $10B annualized; new Spectrum-XGS scale-across for gigascale AI superfactories | — |
| Merchant GPU vs ASIC | not a focus | Jensen argued few ASIC projects reach production; NVIDIA's full-stack, every-cloud, best perf/watt and perf/dollar position wins | — |
Q&A Summary
What is your vision for growth into 2026 across networking and data center, given 12-month wafer-to-rack lead times? (CJ Muse, Cantor)
Jensen cited reasoning/agentic AI needing 100x-1000x more compute, the Blackwell NVLink 72 rack-scale system delivering order-of-magnitude gains, and scaling with Blackwell and Rubin into a $3-4 trillion AI infrastructure opportunity over five years as top-four CSP CapEx doubled to ~$600 billion.
What must happen for the $2-5B China H20 shipments, and could the market shift toward ASICs? (Vivek Arya, BofA)
Colette said there is interest and initial licenses, plus ready supply, but geopolitical issues remain open. Jensen said ASICs rarely reach production because accelerated computing is a full-stack co-design problem; NVIDIA's every-cloud availability, pipeline-wide acceleration, best perf/watt and perf/dollar, and gross-margin advantage make it the choice.
Is $3-4 trillion by end of decade for total infrastructure, and what is your share plus bottlenecks like power? (Ben Reitzes, Barclays)
Jensen said top-four hyperscaler CapEx doubled to ~$600 billion; the US is ~60% of world compute with enterprise and global CSPs on top. For a ~$50-60 billion gigawatt AI factory, NVIDIA represents about 35%, as it is now an AI infrastructure company (six chip types per Rubin supercomputer); power is the key limiter, making perf/watt central.
How should we parse the ~$7B sequential data center increase across Blackwell, Hopper and networking? (Stacy Rasgon, Bernstein)
Colette said Blackwell remains the lion's share and the growth driver into Q3, benefiting both compute and networking (NVLink systems); Hopper (H100/H200 HGX) is still selling, but Blackwell drives the growth. No specific split provided.
Is ~50% CAGR a reasonable bogey for data center revenue growth next year? (Timothy Arcuri, UBS)
Jensen pointed to very significant customer forecasts, AI-native startup funding rising from $100B to $180B and their revenue from $2B to $20B, and an 'everything sold out' environment (H100/H200 sold out); growing into ~$600B/year hyperscaler CapEx over the next several years is reasonable.
More on Nvidia Corp
Reported 2025-08-27 · figures from the Nvidia Corp Q2 2026 earnings call.
See how Top Bucket AI works for your firm
Request DemoTop Bucket AI Research
Stay ahead of private markets
Research and market intelligence for private-markets professionals.
✓
You're subscribed.
Thanks for signing up.