Artificial intelligence (AI) leader Nvidia isn’t perfect, which may be a painful realization for its investors.
Three decades ago, the internet began going mainstream and kicked off a chain of events that changed the growth trajectory of corporate America forever.
Wall Street has been waiting, sometimes impatiently, for the next-big-thing trend to come along that would rival what the internet did for businesses. After an extensive wait, the artificial intelligence (AI) revolution appears ready to answer the call.
With AI, software and systems are given autonomy to oversee tasks that humans would normally handle. The key to AI’s long-term success, and the source of its seemingly limitless ceiling, is the capacity for AI-driven software and systems to learn without human intervention. Machine learning gives AI the potential to become more proficient at existing tasks, as well as learn new skills.
No company has benefited more directly from the rise of AI than semiconductor goliath Nvidia (NVDA -0.21%).
Image source: Getty Images.
Until recently, Nvidia’s operating ramp-up had been flawless
At the beginning of 2023, Nvidia was sporting a $360 billion market cap, which put it on the fringe of being one of the most influential technology companies in America. By June 20, 2024, less than two weeks after completing its historic 10-for-1 stock split, Nvidia’s market cap peaked at $3.46 trillion on an intra-day basis. Investors have simply never witnessed a market leader gain more than $3 trillion in value in less than 18 months before.
The catalyst behind this potentially once-in-a-lifetime move is the company’s AI graphics processing units (GPUs), which have become the standard in high-compute enterprise data centers. The semiconductor analysts at TechInsights estimate that Nvidia was responsible for all but 90,000 of the 3.85 million GPUs that were shipped to enterprise data centers in 2023.
With demand for the company’s chips overwhelming supply, Nvidia has been able to dramatically increase the selling price of its superstar AI-accelerating chip, the H100. In a stretch of five quarters, the company’s adjusted gross margin expanded by roughly 13.7 percentage points to 78.4%.
Having its hardware be the center of attention in enterprise data centers has also fueled ongoing innovation. In March, Nvidia introduced its next-generation Blackwell platform as being capable of accelerating compute capacity in a number of areas, including generative AI solutions, while consuming less energy than its predecessor. In June, CEO Jensen Huang teased the debut of its Rubin GPU architecture, which will be powered by a new processor, known as “Vera.” Rubin is set to make its debut in 2026.
The final piece of the puzzle to Nvidia’s, thus far, textbook operating ramp-up has been its suppliers increasing their capacity to accommodate robust demand. For example, world-leading chip-fabrication company Taiwan Semiconductor Manufacturing (TSM 1.56%) has boosted its chip-on-wafer-on-substrate (CoWoS) capacity, which is a necessity for packaging high-bandwidth memory in AI-accelerated data centers.
Nvidia is flawless no longer
This seemingly textbook “recipe” as the leader of Wall Street’s hottest trend briefly helped Nvidia surpass Microsoft and Apple to become the largest publicly traded company. But after a rough couple of weeks for Nvidia and the stock market as a whole, it’s become plainly evident that Nvidia is just as fallible as any other company.
To maintain its historic run-up, Nvidia needed to execute flawlessly. It had to be able to sell through all of its hardware, command a top-notch price for its products and software — the CUDA platform, which helps developers build large language models — and maintain its competitive edge by bringing its next-gen GPU architecture to market on time.
Unfortunately, reports emerged last weekend that Nvidia has informed many of its top customers (all members of the “Magnificent Seven”) that it would be delaying the shipment of its Blackwell chip for at least three months. This would push delivery into the first quarter of 2025 from an expected arrival date later this year.
According to various reports, the delay stems from potential design flaws with Blackwell, as well as capacity constraints from Taiwan Semiconductor (TSMC). Even with TSMC effectively doubling its CoWoS capacity, it’s still nowhere near enough for Nvidia to meet the demand of enterprise clients.
Blackwell’s delay is the first domino to fall that signals Nvidia is less than perfect. It also opens the door for Nvidia’s competition to thrive.
On July 30, Advanced Micro Devices (AMD -1.50%) delivered second-quarter operating results that were welcomed with open arms by Wall Street and investors. AMD’s data center segment sales surged 115% from the prior-year period and 21% on a sequential quarterly basis (i.e., from what was reported for the March-ended quarter). AMD attributed this outperformance to its ramp-up of AI GPUs.
In particular, AMD’s MI300X is considerably cheaper than Nvidia’s H100. Even though the H100 holds a number of compute advantages over the MI300X, backlogged supply for the H100, coupled with the now-delayed Blackwell chip, gives AMD’s hardware a lot more luster.
To add, all four of Nvidia’s top customers have been developing AI chips for use in their data centers. Even if Nvidia retains its compute advantage, it’s going to lose out on valuable data center “real estate,” as its top customers choose to install their internally developed (and cost-effective) chips.
Image source: Getty Images.
History suggests the Nvidia sell-off is going to worsen
To make matters worse, there hasn’t been a single next-big-thing innovation, technology, or trend that’s escaped an early-stage bubble in three decades. Beyond the advent of the internet, investors have watched early bubbles burst in genome decoding, business-to-business commerce, housing, China stocks, nanotechnology, 3D printing, blockchain technology, cryptocurrency, cannabis, and the metaverse.
The issue with game-changing innovations and technologies is that investors overestimate their adoption and utility. Regardless of how large the addressable market is, it takes time for new technologies to alter the growth landscape for corporate America.
For instance, even though most leading businesses are spending big on AI-driven data centers, many lack a clear blueprint as to how artificial intelligence is going to help them grow their sales and make more money. Investors saw the same story just a few years ago with the rise of the metaverse, and blockchain technology before that. All innovations need time to mature — no exceptions!
Over three decades, market leaders for every next-big-thing trend have consistently declined by 80% or more in value. On a peak-to-trough basis, the leading businesses behind the internet/networking, 3D printing, genome decoding, cannabis, cryptocurrency, and blockchain technology all plummeted by 90% or more before bottoming out.
The silver lining for Nvidia is that it has multiple established segments beyond its AI GPU operations that can provide a loftier foundation than what other market leaders dealt with when their respective bubbles burst. Nvidia’s GPUs used in gaming and crypto mining, coupled with its virtualization software, should prevent a total wipeout.
Nevertheless, history is crystal clear that a big pullback is in order once the euphoria surrounding a next-big-thing trend fades. Blackwell’s delay is the first domino to fall, and it strongly suggests the Nvidia sell-off will worsen in the coming weeks, months, or quarters.