Approximately three decades ago, the advent and proliferation of the internet completely changed the growth trajectory for corporate America. The internet was a technological advancement that opened new sales and marketing channels, as well as paved the way for the retail investor revolution.
Investors have been waiting decades for the next technological leap forward to rival the rise of the internet. The evolution of artificial intelligence (AI) looks to be this long-awaited innovation, with companies like Nvidia (NVDA 1.53%) spearheading the charge.
Giving software and systems the tools to make autonomous, split-second decisions is a game changer that analysts at PwC foresee adding more than $15 trillion to global gross domestic product by 2030. Nvidia's graphics processing units (GPUs) are the brains of enterprise data centers and power software- and system-driven decisions, including the training of large language models.
Image source: Getty Images.
Nvidia's nearly $4.3 trillion gain in market cap since the beginning of 2023 indicates how central its hardware has become to AI-accelerated data centers.
However, this doesn't mean it's free of competitive pressures. Although direct competitors, such as Broadcom (AVGO 1.63%) and Advanced Micro Devices (AMD 3.46%), are often viewed as the biggest threat to Nvidia, arguably its No. 1 competitive risk is something far more near and dear.
Broadcom and AMD are formidable competitors, but not Nvidia's top concern
Given that enterprise demand for GPUs has been virtually insatiable, Nvidia, Broadcom, and Advanced Micro Devices (better known as "AMD") are all having success in AI data centers.
Nvidia controls the lion's share of GPUs deployed due to its compute advantages. Multiple generations of its AI-GPUs, including Hopper (H100), Blackwell, and Blackwell Ultra, have performed head and shoulders above their competition.
Furthermore, Nvidia CEO Jensen Huang is sparing no expense to ensure his company maintains its competitive edge. Huang is overseeing the launch of a new advanced chip each year, with the Vera Rubin GPU expected to succeed Blackwell Ultra in the latter half of this year. With competitors struggling to match the compute abilities of Nvidia's prior-generation GPUs, Huang's accelerated innovation timeline practically ensures his company will retain its spot atop the pedestal.
NASDAQ: AMD
Key Data Points
But there are still paths for Broadcom and AMD to (pardon the necessary pun) chip away at Nvidia's near-monopoly share in AI data centers.
AMD's path to relevance is its brand and value proposition. Consumers and businesses are familiar with AMD as a trusted provider of central processing units. However, its Instinct AI-accelerating chips offer a generally less costly and more readily available alternative to Nvidia's GPUs. With demand for GPUs overwhelming supply, Nvidia's orders are backlogged. This isn't necessarily the case with AMD's AI hardware.
Meanwhile, Broadcom can make waves through its application-specific integrated circuits (ASICs). Though Broadcom is best known for its networking solutions, its customized ASICs may be responsible for $60 billion to $90 billion in sales from a select group of hyperscalers over the next few years.
Broadcom's specialization and AMD's value proposition make them formidable competitors to Nvidia, but not the greatest threat to Wall Street's largest publicly traded company.
Image source: Getty Images.
The No. 1 threat to Nvidia's AI data center dominance comes from within
Arguably, the top risk that's fully capable of pulling the rug out from beneath America's largest publicly traded company is internal competition.
Though Nvidia has a broad customer base, many of its leading indirect customers -- Nvidia sells its chips to original equipment manufacturers that then build AI data center servers for businesses -- are members of the "Magnificent Seven." Industry giants, including Meta Platforms, Microsoft, and Amazon, have spent tens of billions on Nvidia's GPUs to fuel their AI ambitions.
But the one characteristic most members of the Magnificent Seven also share is the internal development of AI chips or solutions for use in their data centers. While far from a comprehensive list:
- Meta has developed more than one generation of its Meta Training and Inference Accelerator chip to support evolving AI workloads.
- Microsoft recently unveiled its next-generation, in-house Azure Maia 200 AI-accelerating chip for inference workloads.
- Amazon has developed two in-house AI chips, Inferentia2 and Trainium. The latter is a custom-designed AI-accelerating chip used in the training and inference of complex generative AI models.
- Alphabet's Google Cloud tensor processing units are also custom-designed for the training and inference of AI models.
While these internally developed chips and hardware aren't necessarily going to be faster or more efficient than Nvidia's GPUs, they are notably cheaper and more readily available. Nvidia's top customers by net sales deploying in-house hardware in their data centers pose several problems.
NASDAQ: NVDA
Key Data Points
To begin with, Nvidia's otherworldly pricing power and generally accepted accounting principles (GAAP) gross margin in the mid-70% range stem from the persistent GPU supply shortage. Internal chip development can quickly put an end to this supply shortage for the market's biggest buyers, thereby hurting Nvidia's pricing power and gross margin.
Secondly, having the members of the Magnificent Seven develop their own AI chips and solutions risks delaying future upgrade cycles. Although we're still about three years away from what would be the first real wave of data center upgrades, this internally developed hardware may lessen the desire for Wall Street's most influential businesses to buy the latest GPUs from Nvidia.
Lastly, there's the potential for Nvidia's aggressive innovation cycle (debuting an advanced chip each year) to rapidly depreciate prior-generation GPUs. If these prior-gen chips devalue considerably faster than anticipated, buyers of this hardware may be incentivized to use it even longer. If Meta, Microsoft, Amazon, and so on, can complement these prior-gen chips with their latest in-house GPUs, there may be little need to rely on next-generation GPUs from Nvidia.
Nothing is a bigger competitive risk for Nvidia than internal competition.
