There Are 16 Billion Reasons to Pay Attention to Broadcom Earnings on Sept. 2

The market is treating custom AI silicon as a zero-sum share fight, but the more important mechanism is that large hyperscalers are turning accelerator sourcing into a multi-vendor stack. That usually expands total addressable spend on surrounding content — networking, memory interface, packaging, validation — while putting a ceiling on any one supplier's pricing power. For Broadcom, the real risk is not a near-term collapse in revenue; it is eventual mix dilution and weaker scarcity multiple if investors conclude Google is using alternates to negotiate harder.

Into the print, the key variable is not the headline AI revenue number so much as whether management can show the next quarter still has backlog visibility and no margin giveback from a richer partner set. If Broadcom clears the bar, the stock can re-rate because the market will have to accept that Alphabet's broader sourcing is additive to total TPU deployment, not substitutive. If guidance is merely in-line, the stock can still underperform because expectations are now high enough that the burden of proof is on sustained acceleration, not just a one-quarter beat.

The contrarian point is that Marvell's rally may already discount the Google optionality, while Broadcom's setup is more durable if AI infrastructure spend broadens beyond one customer relationship. The bigger tell over 1-3 months will be whether Alphabet's capex keeps rising without a corresponding slowdown in monetization; if it does, both AVGO and MRVL can stay bid, but the winner is likely the one with better AI networking attach and operating leverage. Over 6-18 months, the main falsifier is a guide-down tied to customer concentration or evidence that TPU design wins are fragmenting faster than spend is growing.