Nvidia (NVDA) Earnings Review – February 21, 2024
Intro:
Nvidia designs semiconductors for data center, gaming and other use cases. It’s unanimously considered the technology leader in chips meant for high performance compute (HPC) and GenAI use cases. The following are important acronyms and definitions to know for this company:
- GPU: Graphics Processing Unit. This is an electronic circuit to display screen images.
- CPU: Central Processing Unit. This is a different type of electronic circuit that carries out tasks/assignments and data processing from applications.
- DGX: Nvidia’s full stack platform combining its chipsets and software services.
- Hopper: Nvidia’s modern GPU architecture designed for accelerated compute and Generative AI. Key piece of the DGX platform.
- H100: Its Hopper 100 Chip.
- L40S: Another, more barebones GPU chipset based on Ada Lovelace architecture. This works best for less complex needs.
- Ampere: The GPU architecture that Hopper replaces for a 16x performance boost.
- Grace: Nvidia’s new CPU architecture designed for accelerated compute and Generative AI. Key piece of the DGX platform.
- GH200: Its Grace Hopper 200 Superchip with Nvidia GPUs and ARM Holdings tech.
- InfiniBand: Interconnectivity tech providing an ultra-low latency computing network.
- NeMo: Guided step-functions to build granular Gen AI models for client-specific needs. Its standardized environment for model creation.
- Generative AI Model Training: One of two key layers to model development. This seasons a model by feeding it specific data.
- Generative AI Model Inference: The second key layer to model development. This pushes trained models to create new insights and uncover new, related patterns. It connects data dots that we didn’t realize were related. Training comes first. Inference comes second.
Demand
“Accelerated computing and GenAI have hit their tipping points. Demand is surging worldwide across companies, industries and nations.” – Nvidia Founder/CEO Jensen Huang
Nvidia beat revenue estimates by 7.5% & beat its guidance 10.5%. Its 64.1% 3-year revenue compounded annual growth rate (CAGR) compares to 56.4% as of last quarter & 51.7% 2 quarters ago. For revenue buckets besides data center:
- Gaming was flat Q/Q & rose 56% Y/Y.
- Professional Visualization was up 11% Q/Q & up 105% Y/Y.
- Automotive was up 8% Y/Y & fell 4% Y/Y.
Source: Brad Freeman – SEC Filings, Company Presentations, and Company Press Releases
Profitability
- Beat 75.5% gross profit margin (GPM) estimates & beat same GPM guidance by 120 basis points (bps; 1 bps = 0.01%).
- Beat EBIT estimates by 10.2% & beat EBIT guidance by 26.1%.
- Beat $4.64 earnings per share (EPS) estimates by $0.52 or 11.2%.
- It generated $5.16 in EPS vs. $0.88 Y/Y. Wow. Just wow.
Source: Brad Freeman – SEC Filings, Company Presentations, and Company Press Releases
Balance Sheet
- $26 billion in cash & equivalents. It had $13.3 billion just 12 months ago.
- Accounts receivable fell a bit Y/Y.
- Days sales outstanding improved from 42 to 41 Q/Q.
- $5.3 billion in inventory.
- Days of inventory improved from 92 to 91 Q/Q.
- Purchase commitments for inventory and manufacturing capacity fell Q/Q “due to shortening lead times for certain components.” This is likely their H100 supply bottleneck improving as expected. CFO Colette Kress said on the call that it had nothing to do with demand levels, which are excellent by all accounts.
- Nearly $10 billion in debt.
- Share count is down slightly Y/Y. It bought back $9.5 billion in stock during fiscal 2024.
Guidance & Valuation
Nvidia’s next quarter guidance was 9.1% ahead on revenue, 12.2% on EBIT, 17.1% ahead on net income and 150 bps ahead on GPM. It also guided to $14.8 billion in GAAP EBIT.
Based on current estimates, which are set to rise in the coming days, Nvidia trades for 34x fiscal year 2025 EPS. EPS is expected to grow by 74% Y/Y.
Call & Release Highlights
Data Center Chips/Hardware:
Data Center revenue is where the GenAI demand explosion can be seen. As CFO Colette Kress stated on the call, continued advancements in affordable compute are reaching a “tipping point.” That tipping point will allow a new computing era to accelerate every workload possible for better performance and efficiency. Nvidia’s GenAI-focused chipsets are the vehicle allowing this improvement to be fully realized.
Chipsets within this data center bucket, including H100 and GH200, are used for model training and inference, recommendation engines and other GenAI apps. These chipsets delivered rapid growth at a sky-high gross margin; the data center mix shift was explicitly credited for gross margin outperformance. Importantly, that sky-high gross margin is direct evidence of its large technological lead. It can sell equipment and software that seasons models more effectively and more affordably than anyone else can. Pricing power is the byproduct, which is depicted in its gross margin trend. The wave of modern data center construction (or what it calls AI factory construction) is underway and there’s no other company on the planet better positioned to capitalize.
During the quarter, Nvidia launched “new optimizations” for Google’s Gemma open language models. It also added Nvidia DGX (defined above) to the AWS marketplace. AWS is one of the mega-caps trying its hardest to build its own chips for model training and inference. It’s still leaning heavily on Nvidia’s chips, and this is another small piece of evidence pointing to the relationship remaining friendly.
- Cloud service providers overall made up more than 50% of this quarter’s data center revenue.
The automotive, financial services and healthcare verticals were all highlighted as category standouts as well. Nvidia boasts a deep series of industry-specific models, apps and tools to cater more precisely to a more granular set of needs. Within financial services, its fraud detection AI model helped American Express raise detection accuracy by 6%. That offers a sizable financial benefit for the large credit network. For automotive, its “NVIDIA DRIVE” service offers software for sharpening autonomous driving programs with more scalable data ingestion. It sees automotive morphing into another compelling growth lever as autonomous programs lean more heavily on its tech.
“In effect, nearly every automotive company working on AI is working with Nvidia.” — CFO Colette Kress
- Consumer internet and enterprise software firms are the second largest revenue contributor here.
- Nvidia took partial credit for Microsoft GitHub’s revenue acceleration, ServiceNow’s record quarter for net new contracts for a new product and Meta’s improving advertising returns.
- 40% of its overall data center revenue for the year was from AI inference.
- Singtel (large telco in Singapore) will use Hopper GPUs in its new data centers.
- Announced a new AI infrastructure maintenance partnership with Cisco.
“Fundamentally, the conditions are excellent for continued growth in calendar ’24 to calendar ’25 and beyond.” — Founder/CEO Jensen Huang
Data Center Chip Supply:
The H100 chips continue to enjoy easing supply constraints. It expects its next generation of GenAI and HPC chips to also be supply constrained as “demand far exceeds supply.” That’s probably because H200 (the new Hopper chipset) nearly doubles inference performance for H100. It plans to ramp H200 shipments in the spring.
Data Center Networking Revenue:
This subsection of data center revenue encompasses technology allowing products to communicate with one another. The segment crossed $10 billion in annual revenue last quarter thanks to customers like Azure using 29,000 miles of its InfiniBand connectivity cabling. Momentum continued as expected this quarter as annualized revenue rose by another 30% Q/Q to $13 billion. Growth should keep humming as it launches products like its Spectrum X” ethernet product and enjoys 400% Y/Y growth in its new Quantum InfiniBand solutions.
For Spectrum X specifically, the product boasts adaptive routing, traffic congestion control and other tools making it what Huang calls the “standard for AI-dedicated systems.” Overall, Spectrum X offers 60% better networking performance for AI communication vs. any other product on the market. That means lower cost for customers like Dell, HP and Lenovo.
Networking revenue rose by more than 200% Y/Y and InfiniBand was cited as a quarterly standout. Hopper and InfiniBand together drove the data center segment’s overall results. These two products, combining unmatched model seasoning with better communication, are like peanut butter and jelly and are “emerging as the de facto standard for HCP and AI.”
Data Center Software:
NeMo is its standardized framework for large language model (LLM) building for GenAI use cases. This quarter, it announced the Nvidia NeMo Retriever as a new micro-service. With this tool, customers can more seamlessly season LLMs with their own enterprise data to sharpen query response accuracy. Many other firms have similar products. Nvidia also introduced its “MONAI Cloud Application Programming Interfaces (APIs).” This makes it easier for 3rd party developers to infuse AI into medical imaging to help with diagnosis. Finally, Amgen will use Nvidia’s DGX SuperPODs for drug discovery. This product combines the Nvidia DGX system, chips, switches and storage needs into a single unit, or SuperPOD, for high performance compute use cases. Integration with DGX ensures Nvidia can easily update these PODs with new software when needed. DGX SuperPODs complement its BioNeMo model service for biotech.
Software and services revenue reached $1 billion annually this quarter.
Gaming & Professional Visualization:
The RTX platform (and its 100 million users) encompasses applications like real-time ray tracing (simulating light’s impact) and Nvidia Reflex (latency minimizer) to bolster gaming performance. Nvidia launched its new GPU series under the RTX platform called the GeForce RTX 40 SUPER Series GPUs. This offers perks such as 4x faster ray tracing.
It also introduced new GenAI tools to offer deeply personalized chatbots seasoned with a customer’s own data. Finally, Nvidia announced its Avatar Cloud Engine to “integrate GenAI models into non-playable characters.” Importantly, revenue growth was helped by a continued recovery in demand following the normalization of supply gluts. The same factor also helped professional visualization revenue growth. Within professional visualization, it debuted a new GPU chip for “compact workstations.”
Expenses:
GAAP & non-GAAP operating expenses (OpEx) rose by a modest 23% Y/Y and 25% Y/Y, respectively. This was entirely related to hiring to support soaring demand levels. It also invested more heavily in compute and infrastructure, which added to the growth. Triple digit revenue growth paired with 23% OpEx growth works just fine.
China:
These impeccable results were despite new licensing restrictions in China that ban the export of chips like the H100. That market represented nearly 25% of its total data center revenue before the new rules went into place. That fell to around 5% as of this quarter. The robust demand growth at the lofty gross margin becomes even more impressive when considering this context. Nvidia has not received licenses needed to ship newly restricted chips to China. As discussed last quarter, it has shifted to shipping less complex technology that doesn’t require a license.
Take
This was another remarkably positive quarter. The guidance shows this demand boom is far from over, with the margin outlook directly showing you how far ahead of the pack this company remains. Some will point to the size of the revenue beats shrinking sequentially. That can be expected when you condition analysts to expect crazy good quarters, especially considering how many analysts cover this name. And for context, I was praising a 6% revenue beat for my largest holding (Meta) this quarter. Nvidia’s beat was over 9%. Show me someone telling you this print wasn’t remarkably positive and I will show you someone with an ulterior motive. Congratulations shareholders.
