Nvidia's Q1 (Feb/Mar/Apr) once again beat across the board, with equally strong Q2 guidance (Q2 revenue guide $28B vs. consensus $26.6B). Nvidia has again become the world's highest-revenue and highest-net-income semiconductor company, and the first in semiconductor history to achieve an annualized revenue run rate above $100B and quarterly net income above $10B.
Data center Q1 revenue was $22.563B, up 23% sequentially, accounting for 87% of Nvidia's total revenue, a scale that now exceeds Azure and is approaching AWS. This quarter's data center growth was driven by Hopper training and inference cards and InfiniBand products, with strong demand across verticals such as automotive and internet. Cloud CSPs accounted for mid-40s% of data center revenue; Tesla expanded to 35K H100s, and automotive is expected to be the highest-revenue vertical this year, contributing billions; inference revenue accounted for ~40% of data center revenue (TTM $26.3B).
Because data center revenue accounts for an outsized share, there was prior speculation that the company would change its reporting structure to separately disclose compute and networking businesses. That has now happened.
Compute revenue was $19.392B, up 478% year over year and 29% sequentially. Hopper demand continues to grow; H200 began sampling in Q1 and will ship at volume in Q2. The first H200 units have been delivered to OpenAI. H200 and Blackwell remain supply-constrained into next year. Grace Hopper supercomputer attach rate is 80%, and the latest Green500 list shows it sweeping the top three spots.
Networking revenue was $3.171B, up 242% year over year and down 5% sequentially due to supply constraints; growth was driven primarily by InfiniBand; Spectrum-X began shipping in Q1, with full-year contribution expected in the billions (directly competing with Broadcom); networking is expected to return to sequential growth next quarter, and surpassing Broadcom to become the world's #1 networking chip vendor for the full year is a near certainty.
Per guidance, Q2 revenue of $28B implies 114% year-over-year growth, GAAP net income of $14.3B (up 131%), and non-GAAP net income of $15.4B (up 130%). In the Nvidia Q4 article, I noted "full-year 2024 non-GAAP net income floor of $50B." The current run rate is $61B, with an optimistic scenario challenging $80B, bringing net income scale in line with Apple, Microsoft, and Google, writing a new chapter for semiconductors.
Of course, views on AI sustainability vary, but skepticism, bearishness, and FUD about Nvidia have never ceased. I maintain my prior view: Nvidia's future growth will likely come from two shifts: data center growth expanding from cloud-led to verticals and sovereign AI proliferating, and the product roadmap accelerating from a two-year cadence to annual.
On China, I maintain my prior view: "Given US-China dynamics, AI decoupling is inevitable; better short pain than long pain, especially since short-term pain is minimal — that's the most bullish signal."
FY25Q1 Call Highlights:
GPU super-high ROI: GPUs deliver massive cloud rental revenue to CSPs, with a 4-year ROI of 4x; HGX H200 delivers massive API revenue to LLM companies, with a 4-year ROI of 6x.
Blackwell shipping ahead of schedule: Blackwell launch has 100+ OEMs/ODMs, double the Hopper era; Blackwell volume production begins in Q2, ramps in Q3, customer data centers operational in Q4, expected to contribute significant revenue this year.
Sovereign AI progress exceeding expectations: Japan, Italy, France, Singapore, and others have deployed; sovereign AI revenue is expected to reach high single-digit billions ($6-9B) this year.
Hopper remains supply-constrained despite product transition: During the H100-H200-Blackwell transition, Hopper products remain in short supply, with H200 constraints persisting into next year; for generative AI companies, training time is critical, so waiting for chips is not a viable option.
Response to customer custom silicon: 1. Nvidia products accelerate full-stack workloads, not just single scenarios; 2. Nvidia products deploy both on-prem and across all clouds; 3. AI is not just a chip problem—Nvidia provides full-system solutions to achieve the lowest TCO.
Q1 repurchased $7.7B, paid $98M in dividends, with $14.8B remaining in authorization (expect a massive buyback announcement at Q2 earnings); quarterly dividend increased 150% from $0.04 to $0.10 per share starting in Q2 (pre-split).
Purchase commitments and obligations for inventory and manufacturing capacity were $18.8 billion (QoQ+17%) . Prepaid supply agreements were $5.6 billion (QoQ+60%) .
Overall, on this earnings call, Jensen highlighted three near-term growth engines: Blackwell + Spectrum-X + NIMs. The previously articulated concept of Nvidia AI supercomputers as the AI factories of this industrial revolution has materialized, with the company now delivering clusters ranging from hundreds to hundreds of thousands of GPUs to over 100 customers.
This earnings release also revealed two key takeaways:
1. AI supercomputers are no longer mere "pick-and-shovel" plays but "money-printing machines" that directly generate profits (super-high ROI). This is evident from Microsoft and Amazon earnings, and as noted in the Global Cloud Giants Q1 article, AI hardware supply constraints limited Azure AI revenue growth—perfectly rebutting the concern that unprofitable AI software would slow AI hardware procurement.
2. The surprise 10-for-1 split signals management's expectation of an annualized net income run rate exceeding $100B. Using Q1 as an example, post-split quarterly EPS would drop to $0.60; if quarterly EPS returns to $1.00, that implies $25B quarterly net income, a $100B run rate.
For NVIDIA, short-term stock volatility, media hit pieces, and even fabrications are par for the course — tall trees catch the wind. Best to keep a level head.