When investors think about the artificial intelligence (AI) infrastructure build-out, Nvidia (NVDA +2.57%) is usually the first name that comes to mind. The tech company’s graphics processing units (GPUs) have powered the initial wave of the generative AI boom, rewarding shareholders handsomely along the way.
But as the market matures, a new dynamic is taking shape.
Hyperscalers and tech giants are scrutinizing the massive costs of building data centers and actively seeking ways to reduce their reliance on pricey, power-hungry GPUs.
This shift plays perfectly into Broadcom’s (AVGO +4.69%) hands. As the go-to partner for custom AI chips, Broadcom is offering hyperscalers a cheaper, more energy-efficient alternative.
Image source: Getty Images.
The double-edged sword of Nvidia’s margins
To understand the risk hovering over Nvidia, you have to look at its profitability. Nvidia’s margins have done incredibly well recently, serving as the ultimate proof of its economic moat.
In its fiscal fourth quarter of 2026 (the period ended Jan. 25, 2026), Nvidia generated a staggering $68.1 billion in revenue — up 73% year over year and 20% sequentially. Even more impressive, however, the company’s non-GAAP (adjusted) gross margin expanded to 75.2% — up 170 basis points from the already impressive 73.5% it reported in the year-ago quarter.
These astronomical margins are a clear sign of Nvidia’s seemingly unassailable pricing power. When you sell the essential hardware for a technological revolution, and you are the best comprehensive solution, you can essentially name your price.
But here is the real issue: margins this high paint a massive target on your back.
When a technology hardware provider captures a non-GAAP gross margin north of 75%, the biggest tech customers in the world — companies like Alphabet (GOOG 0.20%)(GOOGL 0.41%), Meta Platforms, and Amazon — are highly incentivized to engineer their way around those costs.
And if these tech giants’ custom AI silicon efforts pay off, Nvidia’s margins could start narrowing as competition intensifies or hyperscalers pull back on Nvidia GPU spending.
Broadcom’s accelerating custom silicon
This is where Broadcom’s strategy shines. The company designs application-specific integrated circuits (ASICs) for tech giants.
Unlike general-purpose GPUs, these custom chips are purpose-built to run specific AI workloads. Because they are optimized for a single task, they are incredibly cost-efficient. For example, Google’s custom Tensor Processing Units (TPUs), which are developed alongside Broadcom, offer a cheaper, more energy-efficient way to run inference workloads compared to premium GPUs.
And other tech giants are piling in, too. The company has also disclosed customer relationships with Meta, Anthropic, and OpenAI.
In its fiscal first quarter of 2026, Broadcom’s total revenue grew 29% year over year to a record $19.3 billion. But its AI momentum, specifically, is staggering. The company’s AI semiconductor revenue skyrocketed 106% year over year to $8.4 billion.
The driver? Robust demand for custom AI accelerators and AI networking, including projects like Alphabet’s TPU.
Broadcom’s “custom accelerator business grew 140% year on year in Q1,” CEO Hock Tan noted during the recent earnings call. “This momentum continues in Q2.”
In addition, since Broadcom works so closely with its customers, its customers have high switching costs. Its custom chip business leverages deep, multi-year co-design partnerships. This embeds Broadcom directly within its customers’ technology roadmaps, making the revenue highly sticky and predictable.
Further, while Broadcom may not have the margins that Nvidia does, its profit margins are still impressive. In fiscal Q1, the company posted an earnings before interest, taxes, depreciation, and amortization (EBITDA) of $13.1 billion, representing an incredible 68% of total revenue. It also generated $8.0 billion in free cash flow during the quarter — converting 41% of its top line straight into cash.
Today’s Change
(4.69%) $16.63
Current Price
$371.54
Key Data Points
Market Cap
$1.8T
Day’s Range
$360.82 – $376.55
52wk Range
$161.61 – $414.61
Volume
1.5M
Avg Vol
27M
Gross Margin
64.96%
Dividend Yield
0.67%
The better buy
Ultimately, both Nvidia and Broadcom are exceptional businesses. But they present very different risk-reward profiles today.
Nvidia’s astronomical non-GAAP gross margins are a sign of its dominance, but they also represent a structural risk, as companies aggressively invest in custom silicon to reduce their dependence on Nvidia’s pricey GPUs.
Broadcom, on the other hand, is the toll booth sitting squarely in the middle of that cost-saving transition. The company acts as the co-pilot for tech giants looking to build their own silicon. And its custom accelerator business is growing at an explosive rate that dwarfs the growth Nvidia is seeing in its data center business.
Of course, like Nvidia, it has its own risks. Its reliance on just a handful of large customers in its custom AI accelerator business, for instance, means that losing jue one would be material. In addition, it’s always possible that Nvidia finds ways to not just dominate general-purpose GPUs but also ASICs.
Ultimately, with an accelerating custom silicon business, a deeply entrenched customer base, and substantial free cash flow, I believe Broadcom is the superior stock to buy (compared to Nvidia) for the next decade of AI.