AI Chips: A Mostly Harmless Investment

The demand for artificial intelligence (AI) chips has, for some time now, been exceeding the supply. This is, of course, a classic example of wanting things. A great many things, in fact. It’s a bit like trying to find a decent cup of tea on a spaceship – surprisingly difficult, and with potentially catastrophic consequences if you fail. (The consequences, in this case, being slightly delayed AI-powered cat videos. Though, one must always consider the existential implications.) This imbalance is largely due to the sudden, and rather enthusiastic, popularity of large language models (LLMs), generative AI applications, and those specialized AI agents that are, presumably, learning to make even more efficient paperclips. (Don’t think about the paperclips.)

Precedence Research suggests the global AI chip market could expand at a CAGR of 27.9% from 2026 to 2035. A significant number, certainly. Though, when you consider the sheer volume of data being generated – everything from cat pictures to complex financial models – it’s almost…inevitable. (And yet, we still find ourselves surprised. Humans. Go figure.) To capitalize on this decidedly secular trend – and, let’s be honest, to potentially earn a return on investment – one might consider a look at three of the most closely followed players in the AI chip arena: Nvidia (NVDA 0.65%), AMD (AMD 3.26%), and Broadcom (AVGO +1.41%).

The Differences: A Matter of Silicon and Strategy

Nvidia and AMD both manufacture discrete GPUs – those graphical processing units that, rather brilliantly, can process a wide range of parallel tasks. Think of it as having a team of highly specialized squirrels, each capable of cracking a single nut, working simultaneously. (CPUs, by comparison, are more like a single, slightly grumpy, but highly efficient, squirrel.) This makes them particularly well-suited for handling graphics applications, LLMs, machine learning, and, naturally, all those AI applications that require a lot of number crunching.

Both Nvidia and AMD initially cut their teeth developing GPUs for gaming PCs. A perfectly reasonable starting point, given the insatiable human desire to see slightly more realistic explosions. However, over the past decade, both companies have ventured into the realm of high-end data center GPUs, specifically designed for processing AI tasks. Nvidia generally positions itself at the higher end of the price spectrum, offering more powerful (and, let’s be honest, more expensive) GPUs. AMD, meanwhile, aims for a more affordable approach. (A bit like choosing between a luxury spaceship and a perfectly functional, slightly dented, one.)

Currently, Nvidia’s H100 data center GPUs command a price tag of around $25,000 each. A substantial investment, certainly. AMD’s comparable MI300X chips, on the other hand, are available for roughly $15,000. Many tech giants utilize a blend of both Nvidia and AMD GPUs, though Nvidia’s chips still tend to be favored for the most demanding applications. (It’s a bit like choosing between a finely crafted Swiss watch and a perfectly reliable digital one. Both tell the time, but one has a certain…je ne sais quoi.)

Nvidia has also established a degree of lock-in with CUDA (Compute Unified Device Architecture), a proprietary programming platform optimized for its own chips. Applications developed for CUDA often require significant rewriting or modification to function on AMD GPUs. This gives Nvidia a considerable advantage in the market, despite the higher price point. (It’s a bit like building a city entirely around a specific type of plumbing. Switching systems is…challenging.)

Broadcom takes a slightly different approach. Rather than developing data center GPUs, they focus on creating custom AI accelerator chips for hyperscalers like Alphabet’s Google and Meta Platforms. These application-specific integrated circuits (ASICs) are precisely tailored to their clients’ workloads, maximizing power efficiency at scale. Broadcom also bundles its chips with networking switches, optical equipment, and infrastructure software. (It’s a bit like designing a bespoke spaceship for a single, very demanding, passenger.) However, these custom chips lack the flexibility of Nvidia’s and AMD’s discrete GPUs.

Can They All Thrive? A Question of Scale and Strategy

Nvidia, AMD, and Broadcom all target many of the same hyperscale customers. One might assume this would lead to intense competition. However, there appears to be ample room for all three to thrive without completely trampling each other. (It’s a bit like three perfectly competent spaceship builders all vying for contracts. As long as the universe continues to expand, there’s enough work for everyone.)

Fortune Business Insights predicts the global AI infrastructure market could expand at a CAGR of 29.1% from 2025 to 2032. A rather optimistic projection, perhaps, but not entirely unreasonable. As this market grows, Nvidia will likely continue to dominate the general-purpose AI space, AMD will cater to cost-conscious customers by bundling CPUs and GPUs, and Broadcom will focus on supporting the custom workloads of the largest hyperscalers. (A neatly divided market, wouldn’t you say?)

Therefore, while all three companies are vying for a larger slice of the data center pie, it would be unwise to assume Nvidia is the only significant player in this AI gold rush. Analysts expect all three chipmakers to experience rapid growth in the foreseeable future.

They project Nvidia’s revenue to grow at a CAGR of 47% from fiscal 2025 to fiscal 2028. AMD’s revenue is expected to increase at a CAGR of 34% from 2024 to 2027. And Broadcom’s top line is forecast to rise at a CAGR of 38% from fiscal 2025 to fiscal 2028.

Instead of obsessing over which of these chipmakers will profit the most from the AI boom, a prudent investor might consider simply acquiring all three. They may experience some volatility in this rather choppy market, but they could potentially deliver impressive long-term gains. (It’s a bit like diversifying your spaceship fleet. You never know what cosmic hazards you might encounter.)

2026-01-27 00:12

The Differences: A Matter of Silicon and Strategy

Can They All Thrive? A Question of Scale and Strategy

Read More