Broadcom’s AI Business Is Thriving, Yet the Bigger Picture Is Mixed

Broadcom is riding one of the strongest waves in the artificial-intelligence hardware boom, with its AI chip and networking business growing at triple-digit rates and management now talking about revenue “significantly in excess of $100 billion” from AI chips next year. Yet behind those headline numbers is a more complicated story: a sprawling portfolio that still includes slower-growth legacy semiconductors and a massive infrastructure-software arm, where investors are watching closely to see whether AI is a catalyst or a threat.

AI engine in overdrive

Broadcom has become one of the clearest beneficiaries of the generative AI investment cycle, as hyperscale cloud providers race to build out custom accelerators and high-speed networking to train and deploy large language models. In its latest fiscal first quarter, the company reported artificial-intelligence revenue of about $8.4 billion, up 106% from a year earlier, driven by “robust demand” for custom AI chips and networking solutions. “Our AI revenue growth is accelerating,” President and Chief Executive Hock Tan said, adding that AI-chip revenue alone is expected to reach $10.7 billion in the current fiscal second quarter.

The pace of that expansion has reset expectations for what counts as big in the AI chip world. Just a few years ago, Broadcom was better known on Wall Street for its radio-frequency chips in Apple iPhones; today, AI semiconductors and networking are becoming its main growth engines. The company has guided to AI semiconductor revenue of roughly $5.1 billion in a single quarter in 2025 and forecast roughly 60% annual growth in the AI semiconductor segment through fiscal 2026, underscoring how deeply it has embedded itself in hyperscale data-center buildouts.

That shift is not just about selling more chips but about where those chips go. Broadcom has positioned itself as a behind-the-scenes specialist in custom silicon, networking switches, and Ethernet-based fabrics that tie together large AI clusters, often working hand-in-hand with a small group of very large customers. “We continue to make steady progress in growing our AI revenue,” Tan told investors in earlier commentary, explaining that growth is “largely driven by AI in terms of growth” while other semiconductor businesses expand at a more modest clip.

Big bets on custom AI silicon

A core plank of Broadcom’s AI strategy is custom accelerators designed for what Tan calls a “handful” of hyperscale clients that want chips tuned to their own AI workloads. In an earnings call, he described a “next phase” of custom AI implementation, as the company works with six major customers on chip design names that industry analysts say include giants like Google, Meta and leading AI startups. The appeal for those customers is economics as much as performance: custom silicon promises better cost efficiency, power consumption and workload optimization than off-the-shelf accelerators for massive data centers.

Analysts say that makes Broadcom less of a direct rival to AI-accelerator heavyweight Nvidia and more of a critical supplier in the broader AI plumbing. “Regardless of which software companies win the AI race, they are all investing in Broadcom’s networking chips and custom accelerators,” said Andrew Rocco, a stock strategist at Zacks Investment Research, arguing that the group is “merely entering the early innings of its hypergrowth phase.” A separate Wall Street Journal analysis put it more bluntly: while Nvidia dominates general-purpose AI GPUs, Broadcom’s AI success “won’t all come at Nvidia’s expense,” because its strength lies in custom chips and the network hardware that moves data inside AI supercomputers.

That positioning gives Broadcom a long runway as cloud providers increasingly value differentiation in AI performance and cost. The Futurum Group, a research firm, noted that “hyperscalers are increasingly prioritizing custom silicon to optimize AI workloads for cost efficiency, energy savings, and specialized performance needs,” and concluded that Broadcom is “cementing its role as a critical player in the future of hyperscale AI infrastructure.” By advancing both custom accelerators and AI networking products like its Tomahawk and Jericho families, the company has become a linchpin of Ethernet-based AI clusters used by several large cloud and AI providers.

AI networking: the silent workhorse

If accelerators are the brains of AI systems, networking is the circulatory system and Broadcom has quickly become a dominant supplier of that layer. In its second quarter of fiscal 2025, AI networking revenue surged more than 170% year over year, powered by Ethernet-based switching and routing solutions that connect AI servers at scale. “Broadcom achieved record Q2 revenue on continued momentum in AI semiconductor solutions and VMware,” Tan said at the time, highlighting that AI networking represented roughly 40% of total AI semiconductor revenues for the quarter.

The launch of the company’s next-generation Tomahawk 6 switch, designed specifically for AI cluster performance, has become a focal point for that growth. These chips help large technology companies deploy enormous clusters of accelerators, reducing bottlenecks as they move training data and model parameters across thousands of servers. With hyperscalers doubling down on both training and inference workloads, Broadcom believes this demand will support “sustained annual growth of roughly 60%” in its AI semiconductor business through 2026, including networking and custom accelerators.​

That momentum is also reflected in the company’s backlog and margin profile from AI. An analysis from Investing.com estimated that in a recent quarter Broadcom generated about $8.2 billion in AI revenue at a gross margin close to 78% on the AI side, translating to roughly $6.4 billion in gross profit from AI alone. After accounting for R&D and operating expenses typically running near half of revenue, the AI segment could realistically produce more than $4 billion in operating income per quarter on a run-rate basis, underscoring the profitability of this rapidly growing line of business.​

Software: steady, not spectacular

If the AI story is straightforwardly bullish, the picture in Broadcom’s infrastructure-software division is more nuanced. The unit, built from a series of acquisitions and anchored now by VMware, has become nearly half of the company’s overall revenue, but its growth profile is slower and investor sentiment more cautious. For the latest fiscal first quarter, infrastructure-software net revenue came in at around $6.8 billion, up just 1.4% from the prior year.

Tan has pushed back on fears that the AI boom could cannibalize or disrupt this business. “AI is not disrupting our software business,” he told investors, projecting that infrastructure-software revenue in the current fiscal second quarter would grow 9% year over year to $7.2 billion after the sluggish start to the year. He has argued that VMware’s shift to subscriptions and the embedding of AI into enterprise infrastructure will, over time, support more robust software growth, even if early quarters look uneven as customers adjust to new licensing models.

Outside observers are watching carefully. While AI workloads clearly require more infrastructure software, some CIOs remain wary of price increases and contract changes tied to Broadcom’s VMware acquisition, and there is an open question about how much AI-related software spending will flow through Broadcom versus going to cloud-native or open-source vendors. For now, Wall Street appears willing to tolerate low-to-mid single-digit growth in software as long as AI hardware keeps delivering double-digit or better expansion, but that balance may not hold indefinitely.

Legacy chips and “non‑AI” drag

Complicating the story further is Broadcom’s portfolio of more traditional semiconductors, ranging from networking gear for telecom operators to components for enterprise storage and smartphones. These “non-AI” chip lines still contribute billions in revenue but are growing much more slowly, and in some pockets remain pressured by cyclical downturns or inventory corrections. Tan has acknowledged that demand outside AI is solid rather than spectacular, describing non-AI semiconductors as growing at double digits but “nothing compared to AI, which has grown very strongly.”

Analysts say this creates a tricky optics problem: as AI skyrockets, any softness elsewhere in the portfolio can stand out more clearly. A recent note on Yahoo Finance argued that “Broadcom’s AI accelerator business is booming, but the stock is falling,” pointing to investor profit-taking after blockbuster AI-driven results as well as concerns about how sustainable the broader semiconductor cycle will be outside AI. That analysis highlighted that in a prior quarter Broadcom delivered record revenue of about $18 billion, up 28% year over year, with AI-related revenue up 74%, only to see the shares drop as much as 12% after the report amid valuation worries.

This dynamic helps explain why, as the Wall Street Journal put it, large AI numbers “no longer carry the same weight” with investors when expectations are already high. Some shareholders want to see a clearer path for the rest of the business, traditional chips and software alike to either accelerate or at least maintain healthy, predictable growth that complements AI rather than relying almost entirely on it.

Market reaction and lofty expectations

Broadcom’s rapid AI ascent has already reshaped its market standing. A late-2024 Journal piece noted that the company’s fiscal fourth-quarter outlook for AI helped push its valuation above $1 trillion, making it one of a handful of chip makers in that club and putting it in more direct comparison with Nvidia. That milestone fueled debate about whether Broadcom’s AI business justified a premium valuation over peers, especially given the slower-growing software arm and still-cyclical non-AI semiconductor lines.

The tension between extraordinary AI metrics and cautious investor reactions has persisted. After Broadcom reported a recent quarter that beat expectations and outlined a robust AI outlook, the stock climbed on relief that AI was supporting, not cannibalizing, software, even as analysts parsed the modest 1.4% software growth and the nuances of second-quarter guidance. “Broadcom’s stock climbs after it says AI is not disrupting its software business,” MarketWatch wrote, capturing the idea that narrative reassurance around software can matter as much as AI numbers themselves.

Others argue that the company’s AI backlog and margin profile are underappreciated. Investing.com noted that the AI segment could be generating more than $16 billion in operating income per year on a run-rate basis if current margins and revenue hold, labeling AI “a key driver of overall profitability” that could offset more moderate growth elsewhere. That view echoes commentary from some equity strategists who have called Broadcom “the most underrated AI play right now,” given that much of its contribution comes from infrastructure layers that are less visible to end users but essential to every major AI deployment.

Can software and AI coexist?

For Broadcom, the strategic question is whether its massive software business and its surging AI hardware business will reinforce one another or pull in different directions. On one hand, VMware and other infrastructure-software assets sit at the center of enterprise data centers, which are increasingly weaving AI into virtualized environments, private clouds and hybrid architectures. On the other, many of the most aggressive AI investors are hyperscale cloud providers that build their own control planes and may rely less on traditional enterprise software stacks.

Tan has been adamant that the two sides of the house can grow together. He has repeatedly emphasized that software revenue is adding on a “steady basis,” even if its growth rate looks modest next to AI, and that VMware’s subscription model and AI integration will help. “Software continued to add on a steady basis,” he said in a discussion about the business, while noting that “it’s largely driven by AI in terms of growth” when looking at the company overall.

Still, the contrast is stark. AI semiconductor revenue is growing in the range of 40% to more than 100% year over year, while software is starting from low single digits and aiming for high single-digit growth in the near term. That gap reinforces the perception highlighted in the Journal’s “the rest is complicated” framing that Broadcom is effectively two companies under one roof: a hypergrowth AI hardware supplier and a large, slower-moving software consolidator.

Risks behind the AI boom

Despite the euphoria around AI numbers, Broadcom faces real risks. Heavy concentration among a small set of hyperscale customers means any shift in their capex plans, architectural choices or in-house chip efforts could ripple quickly through Broadcom’s AI order book. Competition is also intensifying, not only from GPU vendors but from other custom-silicon specialists and networking rivals trying to win sockets in the same AI clusters.

There are macro risks, too. The current AI investment boom is happening against a backdrop of tightening tech budgets in some enterprise segments and lingering concerns about the broader economic environment. While most analysts expect hyperscalers to keep spending heavily, some forecasts put their combined capex at more than $600 billion this year. Any moderation in that trajectory could force Broadcom to lean more heavily on its slower-growing software and non-AI chip lines.

And then there is the simple matter of expectations. With the stock already having joined the trillion-dollar club and AI now a significant share of total revenue, Broadcom has far less room for error. As the Wall Street Journal observed, in a world where big AI numbers have become commonplace, even “blowout chip projections” can struggle to impress investors when they are already priced for perfection.

A powerful but uneven AI story

What is clear is that Broadcom has secured a central role in the AI infrastructure buildout, with a combination of custom accelerators and high-performance networking that few rivals can match at scale. Its AI business is not only booming; it is exceptionally profitable and likely to remain a core driver of earnings for years if hyperscale spending stays on its current trajectory. At the same time, the company must continue to prove that its infrastructure-software franchise and legacy semiconductor lines can deliver steady, if less dramatic, growth that supports rather than dilutes the AI narrative.

For now, investors are being asked to live with that complexity. Broadcom is both a poster child for the AI gold rush and a test case for what happens when a company’s fortunes hinge increasingly on a fast-growing but highly concentrated segment, while the rest of its business trudges along at a more conventional pace. As Tan summed up on a recent call, “We expect growth in AI semiconductor revenue to accelerate… as our hyperscale partners continue to invest” a promise that thrills the market even as it leaves open questions about everything else on Broadcom’s sprawling income statement.