If you want to understand where the real money in the AI boom is flowing, you don’t need to look at chatbots or AI apps. You need to look at the companies building the physical infrastructure underneath all of it — and right now, two companies sit at the center of that story: Nvidia and Broadcom.
Table Of Content
- The Headline Numbers: Both Companies Are Having a Historic Year
- Two Different Business Models, Explained Simply
- Nvidia: One Flexible Platform for (Almost) Everyone
- Broadcom: Custom Chips Built for a Handful of Giants
- Why Nvidia Still Leads, Despite Broadcom’s Explosive Growth
- The Numbers Investors Are Watching Going Forward
- What This Means for the Broader AI Industry
- The Simple Takeaway
Both companies have posted extraordinary earnings this year, both are riding the same wave of AI infrastructure spending, and both are now worth staggering sums of money because of it. But they are winning in fundamentally different ways. Nvidia has built what CEO Jensen Huang calls a platform that runs “in every cloud” and “powers every frontier and open source model.” Broadcom has built its business by becoming the company hyperscalers turn to when they want to design custom chips that reduce their reliance on Nvidia in the first place.
Understanding the difference between these two approaches — and why Nvidia’s broader, more flexible platform still holds the clear lead — tells you a lot about where the AI infrastructure race is headed next.
The Headline Numbers: Both Companies Are Having a Historic Year
Let’s start with the raw numbers, because they’re genuinely staggering.
Nvidia reported revenue of $81.6 billion for its fiscal first quarter of 2027, an increase of roughly 85% year-over-year. Non-GAAP earnings per share jumped about 140% compared to the same period a year earlier. The company’s Data Center segment — the part of the business that sells the GPUs and networking equipment powering AI training and deployment — continues to be the primary engine behind this growth, alongside a fast-growing networking business.
Perhaps just as notable: Nvidia is no longer solely dependent on the handful of giant tech companies (often called “Big Tech” or “hyperscalers”) that first drove the AI boom. The company now earns roughly $37 billion from AI, industrial, and enterprise customers outside that core group — a sign that AI infrastructure demand has broadened well beyond a small circle of buyers.
Broadcom, meanwhile, reported revenue of $22.19 billion for its fiscal second quarter of 2026, up about 48% year-over-year — narrowly beating Wall Street’s expectations. But the more eye-catching number sits inside that total: Broadcom’s AI semiconductor revenue specifically grew 143% year-over-year to reach $10.80 billion for the quarter.
Looking ahead, Broadcom CEO Hock Tan has guided for AI semiconductor revenue of roughly $16 billion in the following quarter — implying growth of more than 200%. Tan has gone even further, projecting that Broadcom’s full fiscal 2026 AI semiconductor revenue will reach approximately $56 billion, nearly tripling the roughly $20 billion the company generated from AI chips in fiscal 2025. Some analysts project Broadcom’s total fiscal 2026 revenue could reach around $104 billion, up roughly 63% year-over-year.
By any normal standard, these are extraordinary results for both companies. But the way each company is achieving these numbers reveals two very different bets on how the AI infrastructure market will evolve.
Two Different Business Models, Explained Simply
To understand why Nvidia is described as the leader in AI infrastructure despite Broadcom’s blistering growth rate, it helps to understand what each company is actually selling.
Nvidia: One Flexible Platform for (Almost) Everyone
Nvidia’s core business is selling standardized, general-purpose GPUs — the powerful chips that can train and run almost any AI model — along with the networking equipment and software that connects thousands of those chips together into massive AI “supercomputers.” The key word here is standardized. Nvidia doesn’t build a different chip for every customer. It builds a small number of extremely powerful chip platforms — like its current Blackwell Ultra chips and its upcoming Vera Rubin platform — and sells them to virtually everyone building AI: Google, Meta, Microsoft, Amazon, OpenAI, Anthropic, smaller cloud providers, research labs, and increasingly, industrial and enterprise customers too.
This is precisely what Jensen Huang was pointing to when he described Nvidia as the only platform that runs in every cloud, powers every major AI model — whether it’s a closed frontier model or an open-source one — and scales anywhere AI work is happening. Nvidia’s advantage isn’t just the chips themselves; it’s the enormous software ecosystem (built around Nvidia’s CUDA platform) that developers and companies have built their AI tools on top of over the past decade. That software layer creates what analysts often call a “moat” — a competitive advantage that’s difficult for competitors to replicate, because switching away from Nvidia doesn’t just mean buying different chips, it means rebuilding software that’s been optimized for Nvidia’s ecosystem for years.
Broadcom: Custom Chips Built for a Handful of Giants
Broadcom’s approach is almost the mirror opposite. Rather than selling one standardized platform to a broad market, Broadcom specializes in designing custom AI accelerator chips, known as ASICs (application-specific integrated circuits) or sometimes XPUs, built to the exact specifications of a small number of enormous customers. Instead of buying an off-the-shelf Nvidia GPU, a company like Google or Meta can work directly with Broadcom to design a chip that’s tailor-made for their own specific AI workloads.
Broadcom’s AI growth is heavily concentrated around a small handful of customers — reportedly around six major companies, including Google, Meta, OpenAI, and Anthropic. Google’s in-house AI chips (called TPUs) and Meta’s custom MTIA chips, which Broadcom recently extended a partnership to co-develop, are examples of this custom-silicon strategy in action. Broadcom also has a strong position in AI networking, particularly high-speed Ethernet switches that compete with parts of Nvidia’s networking business.
This is fundamentally a different bet than Nvidia’s. Hyperscalers pursue custom chips like Broadcom’s for a specific reason: reducing dependence on any single supplier, including Nvidia. If a company like Meta or Google can design its own chip for its own specific workloads, it gains more control over its costs, its supply chain, and its long-term technology roadmap — rather than being fully reliant on whatever Nvidia decides to build and how Nvidia decides to price it.
Why Nvidia Still Leads, Despite Broadcom’s Explosive Growth
Given Broadcom’s staggering 143% AI revenue growth rate, you might assume the two companies are neck-and-neck in the AI infrastructure race. They’re not — and the reason comes down to scale, breadth, and risk.
Nvidia captures a dominant share of overall AI hardware spending. According to one industry analysis, Nvidia captures roughly 41.5 cents of every dollar that major hyperscalers spend on AI hardware — a staggering share of the entire market. Nvidia’s quarterly data center revenue alone ($39.1 billion in one recent quarter, growing 73% year-over-year) rivals or exceeds Broadcom’s entire quarterly revenue across all its business lines combined.
Nvidia’s customer base is far broader. While Broadcom’s AI growth depends heavily on roughly six major customers, Nvidia sells to a much wider swath of the market — every major cloud provider, most AI labs, a growing base of enterprise and industrial customers, and increasingly, smaller AI companies that don’t have the scale or resources to design their own custom chips the way Google or Meta can. This broader customer base gives Nvidia a form of protection that Broadcom’s concentrated model doesn’t have: analysts have explicitly pointed out that Broadcom’s custom-chip strategy trades volume for concentration risk. If Broadcom were to lose even one of its handful of mega-customers, the financial impact would be severe. Nvidia, by contrast, isn’t dependent on any single customer to the same degree.
Nvidia’s roadmap keeps extending its lead. With the Blackwell Ultra platform ramping at what Huang has described as full speed, and the next-generation Vera Rubin platform expected in the second half of 2026, Nvidia continues to push the technical frontier forward at a pace that keeps hyperscalers coming back, even as some of those same companies simultaneously invest in custom silicon from Broadcom as a complementary — not necessarily replacement — strategy.
The market is treating them as complements, not pure rivals. This is a crucial nuance often missed in “Nvidia vs. Broadcom” framing: most major AI companies aren’t choosing one or the other. They’re using both. A hyperscaler might rely on Nvidia’s GPUs for the bulk of its flexible, general-purpose AI training and inference workloads, while simultaneously investing in Broadcom-designed custom chips for specific, high-volume, well-understood workloads where a purpose-built chip offers better cost efficiency. One analysis described Broadcom’s overall profile as looking “more like a complement” to Nvidia rather than a direct head-to-head competitor for the same dollars.
The Numbers Investors Are Watching Going Forward
Both companies have set ambitious targets for the rest of 2026 and beyond, and the coming quarters will test whether those targets hold up.
For Nvidia, analysts project full calendar-year 2026 revenue of approximately $370 billion, growing to roughly $484 billion in 2027 — implying the company’s blistering growth rate continues for at least another full year. Nvidia has also committed to substantial supply-related spending, with commitments reportedly totaling around $119 billion, reflecting the sheer scale of manufacturing capacity Nvidia needs to secure to keep up with demand. One area of ongoing pressure for Nvidia has been China: reduced access to the Chinese market due to export restrictions has effectively erased a chunk of data center compute revenue that would have otherwise flowed from that region.
For Broadcom, CEO Hock Tan has set a public goal of exceeding $100 billion in annual AI-related sales by 2027 — a target that would represent a dramatic scaling up of the custom-chip business in just a couple of years. Reaching that goal depends heavily on continuing to expand and deepen relationships with its core group of hyperscale customers, and successfully landing new customers to diversify beyond its current concentrated base.
Valuation in the market reflects some of this nuance too. Broadcom currently trades at a notably higher earnings multiple than Nvidia — investors are effectively paying a premium for Broadcom’s growth potential and the predictability of its software and custom-silicon business, despite Nvidia’s larger scale and lower relative risk profile. Some analysts view this as evidence the market may be paying a premium for Broadcom’s higher growth rate, while others argue Nvidia’s sheer scale, broader customer base, and deep software moat justify continued market leadership even at a lower forward multiple.
What This Means for the Broader AI Industry
The Nvidia-Broadcom dynamic captures something important about where the AI infrastructure market is heading as a whole: it’s no longer a single-winner story.
For the first couple of years of the generative AI boom, Nvidia effectively had the market to itself — there was no serious alternative for companies that needed massive amounts of AI computing power quickly. That’s changing. Broadcom’s custom silicon business, AMD’s growing data center GPU revenue (which hit $5.8 billion in one recent quarter, up 57% year-over-year), and the continued development of in-house chips by companies like Google all point to a market that is diversifying, even as total spending on AI infrastructure keeps climbing to previously unimaginable levels. One industry estimate put the broader semiconductor market on pace to reach $1.3 trillion in 2026 — up sharply from earlier forecasts made just months prior.
Yet even amid that diversification, Nvidia’s position as the default, flexible platform that works everywhere — across every cloud provider, every major AI model architecture, and an increasingly broad set of customer types — continues to give it a form of leadership that’s difficult for any single competitor, including a fast-growing one like Broadcom, to directly challenge. Broadcom isn’t trying to dethrone Nvidia as the industry’s default platform; it’s building a highly profitable, faster-growing business by serving the specific, high-volume needs of a small number of the world’s largest tech companies who want more control over their own AI infrastructure.
The Simple Takeaway
Nvidia and Broadcom are both having historic years, both riding the same enormous wave of AI infrastructure spending, and both reporting growth rates that would have seemed unbelievable just a few years ago. But they’re winning in different ways, for different reasons, serving different needs.
Nvidia remains the dominant, flexible platform that the vast majority of the AI industry — from the biggest hyperscalers to smaller enterprises — builds on by default, backed by a deep software ecosystem that keeps customers within its orbit. Broadcom has built an extraordinarily fast-growing, highly profitable business by helping a small number of the world’s biggest tech companies build customized alternatives, reducing their dependence on any single supplier — including Nvidia itself.
For now, the two aren’t so much competitors fighting over the same dollars as they are two different bets on how the AI infrastructure buildout will keep evolving — and so far, both bets are paying off handsomely. But when it comes to sheer scale, breadth, and the flexibility to power virtually any AI workload anywhere in the world, Nvidia’s platform remains the one the rest of the industry is still building around.
