Broadcom Inc. is no longer a semiconductor company with a software appendage. The fiscal second-quarter results reveal an enterprise that has restructured itself around two mutually reinforcing infrastructure franchises—custom AI compute and a ruthlessly monetized VMware installed base—each operating on vastly different timescales and risk profiles 14,16,24,35,43,45. The consequence is a business model with a projected 60% compounded annual growth rate over the next three years 25, buttressed by a $73 billion backlog 20 and a declared path to over $100 billion in revenue by fiscal 2027 20,39,42. But this trajectory is not a story of uncomplicated demand. It is a lattice of supply-chain commitments, licensing transitions, and customer concentration risks that will be tested as capacity ramps and renewal cycles mature.
The Supply Chain Underpinning the $100B Ambition
The core of Broadcom’s growth is its custom AI accelerator franchise, a business built on multi-year, multi-gigawatt partnerships with six hyperscale and AI foundational model customers 34,46. These are not spot-market chip sales; they are structural, energy-backed compute capacity deals that trace directly back to wafer starts and high-bandwidth memory (HBM) allocations 34,35,46. The company has disclosed a clear line of sight to more than $100 billion in ASIC revenue by 2027 20,39,42, underpinned by commitments such as 1.3 GW to OpenAI 16, expansion of Anthropic’s TPU capacity from 1 GW to over 3 GW 4,5,16,20, and Meta’s plan for 3 GW through 2028 34. Across these agreements, Broadcom is targeting deployment of nearly 10 gigawatts of compute capacity in 2027 alone 47.
Critically, the company has secured wafer and HBM supply through 2027 and is already planning allocations for 2028–2029 34,35,46. This transforms the binding constraint from component availability to execution on deployment and power infrastructure—a margin that is dangerously thin given the physical realities of data center construction 47. The pipeline is deepening: two unnamed customers have contributed $6 billion in purchase orders with shipments expected to begin in late 2026 16. Counterpoint Research projects Broadcom will retain a 60% share of the custom processor market through 2027 37, while the networking business, with its 60% year-over-year revenue growth 7,44, reinforces this dominance. The Tomahawk 6 100-terabit switch is already in market 34,44; the Tomahawk 7, slated for 2027, will extend the competitive lead in interconnect density 40,44.
But this is not an unencumbered ramp. CEO Hock Tan acknowledged that bookings are for future delivery, not immediate shipment 34,45. The translation from backlog to recognized revenue depends on the timely availability of power, cooling, and construction—every link in the infrastructure chain. The margin for slippage is thin.
The VMware Margin: Pricing Power vs. Churn Risk
The VMware integration, completed in November 2023 for $61 billion 1,2,3,6,8,9,10,11,12,17,18,19,27,30,31,33, is a study in compressed monetization timelines. Broadcom immediately eliminated perpetual licenses and mandated a transition to subscription models 28,32, a move whose contractual exposure is only now being felt across renewal cycles. The restructuring of the go-to-market around VMware Cloud Foundation subscriptions 19 was designed to capture value from a massive installed base 18,34, and software margins reported at 93% 34 demonstrate the leverage inherent in the model.
Yet the pricing mechanism that generates this margin is also generating attrition. Customer anecdotes detail renewal increases that reach 11x, 1,500%, or 2.5x prior charges 32. A small enterprise, previously paying $15,000 annually, faced a renewal quote of $76,000 29; a two-host, six-VM setup was quoted at $10,000 per year 27. The pattern is consistent: Broadcom is actively rationalizing the customer base toward its largest accounts, a strategy prefigured in a 2022 shareholder presentation that outlined retaining only about 600 of the largest clients 32. For those outside this tier, the calculus is stark—defection planning to Nutanix is underway 26, enterprise mandates for removal are multiplying 32, and the risk of non-compliance litigation is rising 26.
Notably, Broadcom does exhibit pricing flexibility when customers threaten non-renewal 27,29, and the decoupling of pricing from core count in certain structures 29 suggests a deliberate calibration: the price increases extracted from retained accounts must outweigh the revenue lost from churn 15. This is a narrow-gauge calculation. It works only if the installed base of locked-in, high-switching-cost enterprises proves sticky enough. The recent five-year VMware Cloud Foundation renewal with LSEG 23 indicates that for the largest, most embedded accounts, this assumption holds. For the mid-market, the margin for error is essentially zero.
Concentration and Capacity: The Thin Margin for Error
Underlying both these franchises is a systemic risk: customer concentration. The top five customers account for 40% of net revenue 21, and the XPU business is concentrated among six partners 34. Any single hyperscaler deciding to shift internal accelerator development—or failing to meet its own demand forecasts—would introduce a cascading effect on Broadcom’s forward commitments 15,34,41. When combined with the physical challenges of deploying nearly 10 GW of capacity while holding 77% semiconductor gross margins 47, the operating envelope becomes tight.
This is not an argument against Broadcom’s strategy. It is a recognition that the company’s valuation premium—a trailing P/E of 81.3 13,38 and a gross profit multiple exceeding 40x 46—prices in a near-perfect execution sequence. Any failure to bring capacity online on schedule, any spike in VMware churn beyond modeled levels, would compress these multiples rapidly. The underlying physics of infrastructure deployment has not changed; what has changed is the scale of the bet Broadcom has placed on its own ability to coordinate a multi-vendor, multi-geography supply chain with no buffers for delay.
A Forward Look: Infrastructure as the Binding Constraint
Broadcom is also extending its silicon logic into adjacent domains. The launch of a comprehensive Wi-Fi 8 product family 36, built across five innovation waves and segmented into seven product categories 22,36, positions the company to capture value at the network edge. Collaborations with Samsung and ODM vendors 23, a push into fixed wireless access hardware 23, and the integration of CPU/NPU capabilities into a broadband Edge AI portfolio 36 all signal an ambition to own the entire connectivity stack—from hyperscale fabric to last-mile orchestration.
What the financial illustrations do not show is the fragility inherent in this expansion. Each Wi-Fi 8 design win depends on OEM integration cycles 23 and carrier trials 23 that are subject to their own supply-chain delays. The $690 billion annual hyperscale spending cycle 46 that powers the custom silicon business is itself contingent on the end-user demand for AI services—a variable that remains unproven at scale. Broadcom’s partnership structure, including the $35 billion first tranche with Apollo and Blackstone for over 20 GW of energy-backed capacity 34, effectively converts hardware sales into infrastructure-as-a-service models, bundling funding risk with delivery risk. The historical parallel is instructive: the first transatlantic telegraph cable, celebrated at its launch, failed within weeks because the engineering overreached the insulation constraints of the era. The ambition was correct; the execution, underappreciated.
Broadcom’s present position is structurally strong, but the margin for error—in wafer allocation, in license renewal acceptance, in power infrastructure availability—is thinner than the headline growth rates imply. The next 18 months will determine whether the company has built a resilient infrastructure machine or a tightly coupled system with multiple single points of failure.
