Broadcom at Goldman Sachs conference: ai growth meets supply limits

Broadcom outlined a supply-constrained AI-revenue path of $115 billion in 2027 and $230 billion in 2028, alongside expected free cash flow in the mid-$40 billions in 2027 after generating $39.4 billion over the past 12 months. CEO Hock Tan said power availability, grid readiness, permitting and data-center construction—not chip manufacturing—are the primary constraints on AI infrastructure deployment. Broadcom is expanding custom-XPU partnerships with six frontier-model developers, including OpenAI, Google and Anthropic, while a $35 billion Apollo-Blackstone-backed vehicle supports more than 20GW of compute capacity. Rising cash generation could support increased dividends and buybacks, while the company views repayment of its roughly $56 billion debt as less attractive given its low legacy borrowing costs.

Analysis

The investable read-through is a rotation within AI capex from compute vendors toward the physical enablement stack. If grid interconnection and site commissioning, rather than silicon availability, set deployment cadence, the highest near-term earnings visibility sits with power-management and electrical-distribution vendors (ETN, VRT), engineering/construction firms (PWR, EME), and gas-turbine/grid-equipment suppliers (GEV), where booked work can convert before an accelerator is energized. This also caps the upside surprise embedded in semiconductor estimates over the next 12-18 months: demand remains intact, but revenue recognition shifts from a chip-order question to a construction-completion question.

AVGO’s custom-silicon position is strategically stronger than a simple GPU substitute thesis because co-design and networking content raise switching costs; however, its incremental AI revenue is unusually exposed to a handful of customers whose infrastructure financing and monetization remain unproven. Residual-value guarantees attached to externally financed capacity are the key underappreciated risk: they create contingent credit/equipment-remarketing exposure precisely if a frontier lab fails to fund its commitments. The company’s long-dated outlook should therefore not receive full credit until investors see disclosed backlog, customer prepayments, guarantee limits, and cash conversion rather than management targets.

Near term, December capital-return guidance is a valuation support catalyst for AVGO, especially if repurchases are meaningful relative to the free-float. Over 1-3 months, permitting delays, transformer lead-time commentary, or evidence of frontier-lab financing stress would favor infrastructure names over AVGO; over 6-18 months, successful site energization would reaccelerate AVGO and pressure NVIDIA’s (NVDA) share of hyperscaler accelerator spend. The contrarian view is that power scarcity is not uniformly bullish for utilities: regulated utilities may earn only allowed returns while carrying political and load-concentration risk; equipment and EPC vendors have more direct operating leverage.

Key Decisions for Investors

• Initiate a 6-12 month basket long ETN, VRT, PWR and GEV versus a smaller short SMH hedge. Target the trade around confirmed 2027 data-center project awards; the catalyst is rising backlog/pricing from grid and electrical equipment, while falsification is order growth decelerating below management guidance or broad cancellation of hyperscaler projects.
• Maintain AVGO as a core long but do not chase on long-range revenue claims alone; add following December capital-allocation disclosure if buybacks/dividend growth exceed the market’s implied cash-return expectation. Risk/reward improves if AVGO holds relative performance versus SOX through a risk-off tape; exit or trim on disclosed expansion of residual guarantees, weaker customer prepayments, or a material reduction in 2027 AI outlook.
• Express custom-ASIC share-gain through a 9-15 month AVGO long / NVDA short pair only at a modest 1:1 beta-adjusted size. The thesis requires hyperscaler ASIC deployments to displace incremental general-purpose accelerators; falsify if NVDA data-center backlog and gross-margin guidance continue accelerating or AVGO reveals deployment delays tied to site readiness.
• Avoid treating APO and BX as clean AI beneficiaries until transaction-level economics are disclosed. Set an alert for financing spreads, collateral terms, and guarantee allocation in AI infrastructure vehicles; wider private-credit spreads or any equipment residual impairment would be a negative second-order catalyst for both alternative-asset managers.