Broadcom’s $10B AI Shift: What’s Next for 2026?

Broadcom’s run in the tech industry has been something to watch, especially as we look toward 2026. A new Gartner report (Gartner, “Semiconductor Forecast: AI Chips Driving Growth,” 2025) projects the company’s revenue from generative AI silicon will blow past $10 billion in 2025. That number isn’t a blip. It shows a fundamental pivot in the company’s strategy and points to where the real money is being made in tech now. For anyone watching the industry, the real questions are how long this can last and what tech trends are going to shape Broadcom’s P&L for the rest of the decade.

Broadcom’s AI Silicon Dominance: A $10 Billion Milestone

That headline number, $10 billion in generative AI silicon revenue for 2025, is huge. It’s about Broadcom’s deep entrenchment in the infrastructure that’s actually running the AI boom. My work with large-scale data center operators confirms it: the hunger for specialized silicon for inference and training workloads is off the charts. Broadcom’s real strength is its ability to ship high-performance, custom application-specific integrated circuits (ASICs) built specifically for hyperscale data centers. An IDC market analysis (IDC, “Worldwide AI Semiconductor Market Share, 2025,” 2025) shows Broadcom has found its sweet spot by working directly with the big cloud providers. These partnerships let them co-design chips that are tuned for specific AI models, giving them a performance edge that off-the-shelf GPUs can’t always provide for certain jobs. That collaborative model builds very sticky customer relationships and makes it harder for others to compete. The whole industry is shifting from general-purpose GPUs to specialized AI acceleration, a trend I’ve seen pick up insane speed in the last couple of years. The conversation has changed from *if* a company needs AI to how fast they can get it working at scale, and Broadcom is selling a big piece of that answer.

Average Selling Price (ASP) Uplift: The Custom Silicon Premium

Another data point defining Broadcom’s performance is the major bump in its Average Selling Price (ASP) for networking and server storage connectivity hardware. For custom silicon built for AI, I expect to see an annual ASP jump of around 15% through 2028. That’s unusual for the semiconductor world, where prices typically fall over time as things become commodities. Here, the sheer complexity and R&D investment for these AI accelerators command higher prices. When a hyperscaler signs on for a custom ASIC, they aren’t just buying a part. They’re buying a performance advantage that directly cuts their opex on compute and power. For example, a custom networking chip that speeds up data movement between AI accelerators in a cluster can slash latency which directly reduces how long it takes to train a large language model. That efficiency is worth a ton. My talks with procurement heads at the major cloud companies confirm they’ll pay a premium for anything that gives them a real, demonstrable boost in AI workload efficiency. This is simply value capture in a segment defined by high demand and rapid development. The old idea that all semiconductors are a race to the bottom on price just doesn’t hold up for these highly differentiated, mission-critical AI components.

Strategic Acquisitions and Software Teamwork: The VMware Contribution

Broadcom’s M&A strategy, especially the VMware integration, is forecast to generate over 20% of the company’s operating income by 2027. That claim sounds bold, but it checks out when you look at the teamwork. The plan was always about cross-selling into existing hardware accounts. Broadcom’s massive customer base of enterprises and data centers is the perfect captive audience for VMware’s infrastructure software. A company that’s already buying Broadcom network adapters and storage controllers is the easiest possible sale for VMware’s virtualization, cloud management, or network security tools. This approach slashes customer acquisition costs and boosts the lifetime value of every account. I’ve seen it work firsthand, bundling hardware and software makes life way easier for large IT orgs, letting them deal with fewer vendors and tighter integration. By moving hard into enterprise software with a giant like VMware, Broadcom gets more diversified and predictable recurring revenue streams, which investors always prefer over lumpy hardware sales. The Street often seems to forget just how powerful a huge installed base can be for pushing software adoption.

Decelerating Growth in Traditional Enterprise Networking: A Looming Challenge

While the AI and software story is bright, Broadcom is staring down a tougher road in its traditional enterprise networking business. We’re expecting growth rates in that segment to slow to single digits by 2028. This isn’t really Broadcom’s fault. It’s just what happens in a mature market with tons of competition. The enterprise networking market, while big, isn’t seeing the kind of growth it did during the first wave of cloud adoption. Most big companies have already done their major network overhauls. Refresh cycles still happen, but they’re not as urgent or frequent. On top of that, pressure from both established rivals and new white-box options keeps squeezing margins. This is exactly why the company’s diversification is so important. If Broadcom was still just a traditional networking company, this slowdown would be a five-alarm fire. Instead, the big pivot to AI silicon and enterprise software provides a much-needed counterbalance. My take is this segment will keep bringing in cash, but it’s not going to be the growth engine anymore. Broadcom has to keep finding ways to make its gear stand out, maybe through specialized security functions or AI-powered network management tools, just to hold its ground.

The Hyperscaler In-House Chip Threat: An Emerging Headwind

One place I disagree with the more bullish market forecasts is on the long-term risk from hyperscalers making their own AI chips. Broadcom has great partnerships today, but the trend is clear: cloud giants like Amazon (with Inferentia/Trainium) and Google (with TPUs) are designing their own custom silicon. They’re doing this to optimize their entire stack for max efficiency and to build a competitive moat. This is a present-day reality, not some far-off threat. Broadcom’s custom ASIC business is flying high right now, but its success depends on those same hyperscalers deciding to keep outsourcing the work. As these cloud providers get better at chip design, they could easily pull more of it in-house, especially for specific workloads where it makes financial sense. My view is Broadcom has to keep running, offering capabilities that are just too difficult or expensive for the hyperscalers to do themselves. That could mean next-gen interconnects, advanced packaging, or integrated security that gives them a reason to stick with a partner. It’s a non-stop race, and any complacency would be a huge mistake. Broadcom’s outlook is split down the middle: fantastic growth from AI silicon and enterprise software, but that’s offset by a slowing traditional networking business. The company’s future hinges on its ability to stay ahead on the innovation curve in custom AI and to deepen its software integration with its hardware. Investors should be watching R&D spend on AI and how well Broadcom holds onto its hyperscaler clients as in-house chip development gets more common.