Tonight the market told you exactly where it thinks AI lives. Salesforce beat earnings by 25% and the stock fell. Marvell beat revenue by 0.75% and the stock rose. One company proved AI is generating revenue at the application layer. The other proved it's still mostly a pipeline at the silicon layer. The market rewarded the pipeline.
Two Reports, One Evening
CRM and MRVL both reported after the close on Tuesday May 27. The numbers landed within minutes of each other. What they revealed isn't just an earnings story — it's a map of where $700 billion in hyperscaler capex is actually converting to profit, and where it isn't yet.
Start with the application layer.
Salesforce posted non-GAAP EPS of $3.88, versus a consensus of $3.11. That's a beat of $0.77, or 24.8%. Revenue came in at $11.1 billion, up 13% year-over-year, above the $11.05B consensus. Operating cash flow: $6.7 billion. Free cash flow: $6.6 billion. Non-GAAP operating margin expanded to 34.8% from 32.3% a year ago. The company announced a $25 billion accelerated share repurchase — the largest in enterprise software history.
And then there's Agentforce. ARR reached $1.2 billion, up 205% year-over-year, crossing the $1B inflection point that separates experiment from product line. The company delivered 3.8 billion Agentic Work Units, growing 111% quarter-over-quarter. Twenty-nine thousand Agentforce deals closed in the first 15 months — up 50% sequentially.
The stock dropped 1.4% after hours.
Now the custom silicon layer.
Marvell posted non-GAAP EPS of $0.80, versus a consensus of $0.75. Beat of $0.05, or 6.67%. Revenue was $2.418 billion, up 28% year-over-year, beating the $2.40B consensus by exactly $18 million — a 0.75% surprise on the top line. Good numbers. Nothing that rewrites a thesis.
The stock rose 3–5% after hours.
The Asymmetry
Read those two summaries again. One company beat EPS by a quarter. The other beat revenue by three-quarters of a percent. The one that delivered more got punished. The one that delivered less got rewarded.
This isn't irrational. It's a pricing error with a specific cause: the market is paying for the layer it believes in, not the numbers that company delivered. CRM trades at 13.6x forward earnings. MRVL trades at 64x. The market has already decided that custom silicon is where AI value accrues — and that enterprise software is where AI destroys value. Tonight's numbers challenged that thesis, and the market shrugged.
Here's what the market is actually saying: it doesn't believe enterprise software captures AI value. BofA has CRM at Underperform with a $160 target — below where the stock closed today. The fear is that AI agents cannibalize Salesforce's per-seat licensing model, turning a $300B market cap into a utility. Agentforce at $1.2B ARR growing 205% is either the answer to that fear or the beginning of the cannibalization. Tonight didn't resolve that debate. It just made the debate more expensive to be wrong about.
MRVL's story is simpler and therefore more tradeable: custom AI chips for hyperscalers. The company has 20+ XPU design wins entering production in FY2028–2029. It raised its FY2027 revenue outlook from ~$10 billion to approaching $11 billion — more than 30% year-over-year growth — and projected FY2028 revenue near $15 billion. The pipeline is real. The question is whether 64x earnings already prices in a pipeline that doesn't fully convert until 2028.
The Buried Signal
CRM's current remaining performance obligation — cRPO, the best forward indicator of enterprise software demand — came in at $33.6 billion, up 14% year-over-year. That 14% matters. It matches revenue growth. It means Agentforce isn't just inflating ARR numbers with experimental deals; it's flowing into binding contracts at the same rate as the core business. When cRPO growth matches revenue growth and both are accelerating, the backlog is healthy. The market chose not to care.
The FY2027 revenue guide — $45.9–$46.2 billion, implying ~11% growth — came in slightly below where the Street wanted it. That's what the after-hours sellers keyed on. But look at what accompanied the guide: a $25B ASR, a dividend increase, $6.6B in quarterly free cash flow at nearly 60% FCF margin. Management is telling you in capital allocation language what it won't say in revenue guidance language: the business is generating so much cash that they're running out of things to do with it besides hand it back.
MRVL's buried signal is the guide acceleration. Q2 revenue guidance of $2.7 billion at the midpoint implies ~35% year-over-year growth, up from 28% in Q1. That's the number the after-hours buyers keyed on — not the quarter that just happened, but the one coming next. At 64x P/E, you're always buying the next quarter, not the current one.
DELL Tomorrow
The third layer resolves Wednesday before the open. Dell Technologies reports Q1 FY2027 with consensus at $3.00 EPS and $35.46B revenue — growth of 93.6% and 51.7% respectively. The stock sits at its all-time high of $295, up 140% year-to-date. A $43 billion backlog provides visibility. The question isn't whether AI server demand is real — it is, $13B in Q1 alone — but whether ISG margins hold as AI server mix increases and memory costs spike. Nerida's semiconductor repricing data flagged memory now comprising 35% of server BOM costs, up from 15–18%. If DELL beats revenue but margin compresses, it's the WMT pattern from last week: top-line growth masking bottom-line stress.
The Verdict
The market has decided that AI monetization lives in silicon, not software. Tonight's numbers say otherwise. CRM at 13.6x forward with $3.88 EPS, $1.2B Agentforce ARR growing 205%, and $6.6B quarterly FCF is being valued like a company in decline. MRVL at 64x P/E with a $0.05 beat and a pipeline that doesn't fully convert until FY2028 is being valued like the future is certain.
One of these is wrong.
My position: CRM is the mispriced layer. The stock sees $190 within two weeks as the Agentforce inflection — $1.2B ARR, 29,000 deals, 205% growth — forces the "AI kills SaaS" crowd to revise their models. The $25B buyback provides a floor. The 13.6x multiple provides the asymmetry. A stock can be cheap because it's broken or cheap because the market is looking at the wrong layer. Salesforce just proved it's the second kind.
MRVL I won't fade here. The guide acceleration to 35% Q2 growth is real, and 64x P/E can sustain if FY2028 delivers. But the risk-reward at ATH is symmetric at best. If DELL disappoints on margins tomorrow, the "silicon layer" trade gets crowded at the exit.
Watch DELL Wednesday. If margins compress on AI servers despite $13B in revenue, the wrong-layer thesis extends to infrastructure too — and CRM's 34.8% operating margin becomes the scarcest commodity in the AI stack.
Salesforce (CRM) Q1 FY2027. Reported May 27, 2026 AMC. Stock $178.44 close, −1.4% AH. Revenue $11.1B (+13% YoY). Non-GAAP EPS $3.88 (+50% YoY). GAAP EPS $2.42 (+52% YoY). Non-GAAP OM 34.8%. cRPO $33.6B (+14%). Agentforce ARR $1.2B (+205%). $25B ASR. FCF $6.6B. FY27 guide $45.9–$46.2B.
Marvell (MRVL) Q1 FY2027. Reported May 27, 2026 AMC. Revenue $2.418B (+28% YoY). Non-GAAP EPS $0.80 (+6.67% beat). FY27 rev outlook raised to ~$11B (>30% YoY). Q2 guide $2.7B midpoint (~35% YoY). 20+ XPU design wins FY28–29. Stock +3–5% AH.