
NVIDIA Q3 earnings again comprehensively beat expectations, with equally strong Q4 guidance (Q4 revenue guidance $20B vs. consensus $17.9B), in line with my personal expectation of $20B. NVIDIA Q3 revenue scale successfully surpassed global foundry leader TSMC ($17.28B), becoming the highest-revenue semiconductor company globally.

Interestingly, NVIDIA and TSMC mutually benefit; TSMC's October revenue unexpectedly hit an all-time high, inevitably attributed to NVIDIA's contribution. (NVIDIA Q3 corresponds to Aug/Sep/Oct)



The core of this quarter's earnings remains data center, as other businesses' share has dropped significantly. Data center Q3 revenue $14.514B, up 41% sequentially, accounting for over 80% of NVIDIA total revenue. Within data center, networking business exploded on the back of compute, growing 5x year over year.

Per guidance, Q4 revenue $20B, up 231% year over year; GAAP net income $10.1B, up 617% year over year; Non-GAAP net income $11.1B, up 411% year over year. Management expects data center product supply growth to continue through 2025. In the prior article 'NVIDIA Q2 Officially Surpasses Intel and Samsung, Global #1 in Semiconductor Chip Revenue,' we predicted '2023 full-year Non-GAAP net income floor $26B, applying NVIDIA's long-term valuation center of 50-70x implies $1.3-1.8T market cap.' Now 2023 full-year Non-GAAP net income floor is $30.6B, making the $1.3-1.8T valuation look cheap.
Of course, AI sustainability remains to be observed; opinions vary; skepticism and bearish calls on NVIDIA have never ceased. Even with Jensen stating on the call that data center growth is sustainable through 2025, the stock showed no after-hours pop.
I believe the biggest surprise in this earnings is that despite a sharp revenue decline in China and other export-controlled regions, Q4 revenue still breaks $20B. The historical '20-25%' data center revenue share from China and other export-controlled regions is now less meaningful. Given US-China dynamics, AI decoupling is inevitable; better short pain than long pain, especially since the short term 'doesn't hurt' yet — this is the most positive news.
FY24 Q3 Call Highlights:
Data center compute revenue up 324% year over year, up 38% sequentially, driven by sustained demand for HGX platform from CSPs, large consumer internet companies, and enterprises; in Q3, large consumer internet and enterprise growth exceeded overall growth; CSPs account for nearly half of data center revenue; Ampere revenue declined sequentially; Q3 shipments primarily Hopper; L40S and GH200 began recognizing revenue; AWS, Azure, Google Cloud, Oracle Cloud to deploy H200 starting next year.
Grace Hopper Q3 shipments to Los Alamos and Swiss National supercomputers; UK Isambard supercomputer to adopt 5,500 Grace Hopper; Germany Julich supercomputer to adopt 24,000 Grace Hopper (90 exaflops); expected next year US, Europe, Japan Grace Hopper-based supercomputer compute to exceed 200 exaflops; Grace CPU targeting multi-billion dollar market, continuing to ramp; Europe prefers Arm ecosystem; CSPs also favor it because their custom chips are Arm-based.
Data center networking revenue up 155% year over year, up 52% sequentially; annualized revenue run rate over $10B; almost entirely from InfiniBand shipped with HGX; InfiniBand revenue up 5x year over year; Azure uses over 29,000 miles of InfiniBand cabling, enough to circle the Earth multiple times; Spectrum-X for traditional Ethernet users shipping Q1 next year (challenging Broadcom); Q4 data center expected to grow significantly sequentially; data center demand visibility extends to 2025; capacity will continue to grow over the next several quarters.
Software revenue continues to grow; year-end annualized run rate over $1B, primarily DGX Cloud + AI Enterprise; AI Enterprise $4,500/GPU/year; Azure AI Foundry initial customers include SAP, Amdocs; NVIDIA's own supercomputer now has over 10,000 H100s.
Q3 share repurchase $3.81B, dividends $99M; previous repurchase authorization had $25.2B remaining.
On this earnings call, Jensen outlined NVIDIA's three future growth engines: CPU, networking, software. He also introduced the concept of Sovereign AI Clouds for the first time, which underpins sustained data center growth.

Short-term stock volatility is inevitable; maintain a level-headed perspective.
