Broadcom has crossed from a chip-cycle survivor into something closer to a strategic utility for the hyperscale AI buildout, and the second fiscal quarter of 2026 makes the case with raw numbers that almost overshadow the more interesting structural story underneath. The company posted $22.2 billion of total net revenue in the three months ended May 3, 2026, a 48% jump from $15.0 billion a year earlier, while net income reached $9.31 billion, up 87% from $4.97 billion. The first half landed at $41.5 billion in revenue against $16.66 billion in net income, and the balance sheet now carries $179.2 billion of total assets, evidence that the post-VMware digestion phase has not only ended but inverted into a compounding cash engine. This is a business that prints money, and the question for the next twelve months is whether the prints keep getting bigger at the pace the market now assumes, or whether the law of large numbers and a softening non-AI franchise start to bite.
The product-side ramp is the headline, and it is genuinely historic. Q2 products revenue of $16.9 billion does not happen in mature semiconductor markets, and the bulk of that surge traces to custom AI accelerators and the networking silicon that surrounds them. Broadcom sits at the architectural center of three of the largest custom-ASIC programs in the world, and the per-rack dollar content of those designs is climbing as training and inference workloads push toward tighter, faster fabrics. The company pairs those accelerators with Tomahawk and Jericho switch silicon, optical DSPs, and PCI Express retimers, and the result is a basket of parts that travels together inside every hyperscale order. Customers do not bolt on Broadcom silicon as a feature; they design their AI clouds around it, and that level of architectural lock-in is what turns a chip company into infrastructure.
Subscriptions and services revenue of $5.3 billion in the quarter, with the full H1 now exceeding the run-rate most analysts expected from the VMware acquisition on day one, is the second story and the more fragile one. The software franchise carries inflated near-term growth from license true-ups and subscription conversions that were pulled forward at the close of the deal, and those non-recurring dynamics are now rolling off. The run-rate, however, is real: tens of thousands of VMware customers paying for virtualization, software-defined networking, and a growing bundle of private-AI and security offerings, with a renewal curve that has held up better than skeptics projected. Broadcom has rebuilt the unit around fewer, larger, deeper contracts, and the cross-sell of AI-ready software into the existing base is becoming a measurable tailwind rather than a talking point.
The mix of those two engines is what makes the story durable. A pure-play AI silicon name is hostage to a single customer's roadmap and to the order timing of three or four hyperscalers. A pure-play software roll-up is hostage to seat counts and IT budgets. Broadcom now has a $16.9 billion silicon quarter anchored by multi-year accelerator commitments, paired with a recurring software base that produces predictable renewals and high incremental margins. Free cash flow conversion has been north of 100% of net income for several consecutive quarters, and management has demonstrated capital discipline by directing most of that cash to debt paydown rather than buybacks, with the deleveraging program now in its final stages.
The valuation, however, embeds the assumption that nothing goes wrong. The stock trades at a premium multiple to large-cap semiconductor peers and to legacy infrastructure software comps, and that premium is justified only if the AI revenue ramp extends into 2027 and 2028 at the slope the buy-side currently models. Any visible deceleration, whether from a hyperscaler pushing out a training build, a custom-silicon program slipping a quarter, or VMware renewals softening, would compress the multiple quickly because the upside in the base case has already been paid for. The risk-reward is asymmetric: a clean print in the next two quarters keeps the path of least resistance higher, while a stumble invites a brutal re-rating.
The watch list for the next six months is concrete. Track hyperscaler capex commentary, particularly from the three customers that anchor Broadcom's AI backlog, and watch the timing of the next-generation AI accelerator tape-outs, since each generation widens the addressable rack content. Monitor VMware renewal commentary for any sign that customers are paring rather than expanding subscriptions as their own cloud bills rise. Follow the broadband and mainframe franchises, where revenue is intentionally being harvested rather than grown, for the slope of the decline, since a steeper drop than guided would pressure the consolidated growth rate. And watch the balance sheet: every meaningful debt repayment changes the equity story, because lower interest expense translates almost dollar-for-dollar into net income on a company this large. Broadcom enters the second half of fiscal 2026 as one of the cleanest expressions of the AI infrastructure trade, but the trade is no longer mispriced, and execution from here has to be flawless.