Broadcom, Credo and Snowflake: One AI Cycle, Three Different Signals

Three AI-related companies reported earnings in the same week, all with strong AI demand to show for it. The market answered Broadcom, Credo, and Snowflake three completely different ways: Broadcom dropped roughly 5% after hours, Credo fell about 20% following its report, and Snowflake jumped more than 20%. That divergence is not noise. It is the market pricing each company at a different point on the AI commercialization chain — building the machines, wiring them together, and then getting paid for the software that runs on them.

Broadcom: the buildout, priced for even faster

Broadcom lives at the top of the chain, where AI capital is turned into silicon. For its fiscal third quarter it reported revenue of $29.59 billion, up 86% from a year earlier, with adjusted earnings of $3.32 a share against $3.24 expected. The AI segment is the engine: AI semiconductor revenue hit $16.7 billion, up 221% year over year and 54% sequentially.

Then management did something unusual for a company this size. It put multi-year numbers on the table: $58 billion of AI semiconductor revenue for fiscal 2026, $115 billion for fiscal 2027 — roughly a doubling — and $230 billion for fiscal 2028. For the current quarter it guided AI semiconductor revenue to $21.7 billion, up 236% year over year.

And the stock still fell about 5%. This is the expectations scoreboard in its purest form: an excellent result against a hurdle so high it moved. Wall Street's concern was the total fourth-quarter revenue forecast, which came in below estimates, not the AI growth line. The market is not doubting AI demand at Broadcom; it is demanding that these numbers keep accelerating and punished the one quarter that didn't rise fast enough. With a forward earnings multiple near 90 times, the premium has to be earned every quarter.

Credo: the interconnection, priced for perfection

One layer down, Credo makes the high-speed copper and optical interconnect gear that ties AI clusters together. It delivered the fastest growth of the three — record revenue of $479 million, up about 115% year over year, beating both its own and Wall Street's forecasts, and it guided the next quarter to $525–535 million. Adjusted earnings were $1.20 a share.

Credo fell about 20% on the report anyway, and it has been sliding all summer: down roughly 45% from a 52-week high near $309 and about 24% over the past month. This is the mirror image of the Broadcom trap. The growth is real and the demand story is intact, but investors had already paid an enormous price to own it — the stock more than doubled off its 2025 lows, and even after the selloff it still trades at more than 120 times forward earnings. The drop was a valuation and positioning reset, not a demand signal. The risk is that at that multiple, no beat is ever quite enough.

Snowflake: where AI becomes paid revenue

At the bottom of the chain sits the layer that actually gets a check the day a customer runs a workload: software. Snowflake's numbers look modest next to the chipmakers — total revenue of $1.55 billion, up 35% year over year, with product revenue up 37% — but they carry a different meaning. Product revenue accelerated for a third straight quarter, adjusted earnings of $0.62 a share beat the $0.45 expected, net revenue retention held at 126%, and remaining performance obligations grew 30% to $9.0 billion. Management raised full-year product revenue guidance to $6.07 billion and guided the next quarter to 37–38% growth.

The stock jumped more than 20%. This is the layer where AI cuts capex and turns into recurring, usage-based revenue — and investors paid up for evidence of a conversion they could see in the numbers.

The deeper question: what the money is actually proving

Read the three reactions together and the market is telling one coherent story. It rewards confirmed monetization (Snowflake), it second-guesses the price of a scarce bottleneck (Credo), and it expects constant acceleration even from the biggest winner (Broadcom). "AI" is not one trade; it is three layers with different economics and different risks, and this week's reports are a reminder that the demand is real at every layer while the market's willingness to pay varies sharply with where the money lands.

The checkpoints that confirm or break the frame: Broadcom has to actually deliver the $21.7 billion quarter and the fiscal 2027 doubling; Credo has to land its $525–535 million guide without further deceleration; Snowflake has to keep product growth accelerating and net revenue retention above 125%. The single signal that would undercut all three is tightening at the source of the money — if hyperscaler capex permission wavers, the buildout, the interconnection, and the software all lose their floor at once.