One Press Release, Half a JPMorgan
This Wednesday, May 20, a few minutes after the market closes, NVIDIA will publish its first-quarter results. The conference call starts at 5pm Eastern. By the time it ends, every AI-adjacent asset on the planet will have repriced.
That is not a figure of speech. The options market is pricing roughly an 8% move in NVIDIA stock off this print. On a company worth about $5.4 trillion, that's more than $430 billion of market value swinging on a single report - over half of JPMorgan, America's largest bank, appearing or vanishing in an afternoon.
The stock comes into the week hot and nervous at the same time. Last Thursday it closed at a record $235.74 - a market cap that briefly touched $5.7 trillion, which, as Euronews pointed out the same week, is larger than Germany's entire GDP. Then two straight red sessions took it to Monday's $222.32 close: 5.7% off the record, yet still up 12% in May alone.
So we did what we always do before a print this size. We used Barebone to pull every NVIDIA earnings report since May 2023 - the estimate, the result, and the next-day tape - alongside two years of margin history and the capex guidance of its four largest customers.
Here's the headline finding. NVIDIA has beaten earnings estimates twelve quarters in a row. The stock fell the next day after six of them.
Hold that thought. It's the key to what Wednesday is actually about - and it isn't whether NVIDIA beats.
Not Really About NVIDIA
This year, the four companies funding most of the world's AI buildout guided to roughly $700 billion of combined 2026 capital spending - the money committed to data centers, servers, and chips. We read all four earnings calls when they landed on April 29:
| Company |
2026 capex guidance |
| Amazon |
≈$200B |
| Microsoft |
≈$190B |
| Alphabet |
$180 - 190B |
| Meta |
$125 - 145B |
Sum the midpoints and you get about $710 billion. Four companies. One year.
Now look at where those dollars land. NVIDIA's data center segment did $62.3 billion of revenue in the January quarter - roughly $250 billion a year at the current run rate. To a first approximation, Big Tech's capex line and NVIDIA's revenue line are the same dollars on two income statements. Not all of it - every one of those four buyers is building its own chips, and we'll get to that - but no other company converts more of the AI buildout into reported, audited revenue.
That last word is the point. The AI economy's other tentpoles don't report. Anthropic is in talks to raise $30 billion at a $900 billion valuation, per Bloomberg's May 12 story. OpenAI closed a $122 billion round at $852 billion in March - a private company now valued above JPMorgan, at $817 billion. Those are remarkable numbers resting on private projections nobody outside the room has audited.
NVIDIA, by contrast, publishes GAAP financials every 90 days. Between hyperscaler earnings seasons, its report is the only hard, audited reading the entire AI trade produces.
So when NVIDIA opens its books on Wednesday, it isn't really opening its own books. It's opening everyone's.
Wall Street understands this, which is why the professionals aren't focused on the headline numbers at all. They're watching two lines deeper in the report.
Watch Line 1: The 75-Cent Question
Gross margin is the fraction of every revenue dollar left over after paying the direct cost of building the product. NVIDIA's, last quarter: 75.0% (75.2% on an adjusted basis). For every dollar of AI hardware sold, NVIDIA keeps 75 cents before operating costs. That is what a monopoly on a shortage looks like, expressed as a percentage.
The company's own guidance for the quarter reporting Wednesday: 74.9%, give or take half a point.
Why would half a point of margin matter more than billions of revenue? Because gross margin is pricing power, measured. The same four companies writing the $700 billion of checks are all funding escape routes - Google's TPUs, Amazon's Trainium, Meta's MTIA program with Broadcom, and OpenAI's own deal with Broadcom for 10 gigawatts of custom accelerators. Every one of those programs is a negotiating lever. If NVIDIA's margin starts sliding, it means the biggest buyers on earth are finally winning the argument over price.
That wouldn't just signal a weaker NVIDIA. It would be the first measurable crack in the assumption underneath every AI valuation: that demand exceeds supply at any price.
The chart has one crater, and it's instructive. In the April 2025 quarter, gross margin collapsed to 60.5% - not from competition, but from a $4.5 billion writedown of China-bound H20 chips when U.S. export rules changed overnight. Here's the part worth remembering: the stock rose 3.3% the day after that report. The market forgave a twelve-point margin crater because it came with a geopolitical excuse and the demand story stayed intact.
That's the asymmetry to carry into Wednesday. A margin dip with a one-off attached is survivable. A margin guide that slips quietly, with nothing to blame - that's the print where the pricing-power debate stops being theoretical.
Watch Line 2: The Guide Is the Product
A stock price is, mechanically, the market's estimate of all future cash flows, discounted to today. The quarter NVIDIA reports Wednesday ended in late April - the market has spent months assuming it. What nobody has is management's forecast for the July quarter. That single number, the guide, is what every model on Wall Street re-anchors to within minutes of the release.
Here's the bar as it stands:
| Line |
NVIDIA's own guide (Feb 25) |
Street consensus (mid-May) |
| Revenue |
$78.0B ±2% |
$79.17B |
| Adjusted EPS |
- |
~$1.78 |
| Gross margin |
74.9% ±50bp |
holding ~75% |
Notice the trap built into that table. Consensus sits $1.2 billion above the midpoint of NVIDIA's own guidance. The Street's official forecast already assumes NVIDIA lands in the top half of its stated range - "meeting guidance" is, by construction, a disappointment.
Two honesty notes on the growth rates attached to those estimates. The +80% revenue and +120% EPS growth implied by consensus screen spectacularly - but the comparison quarter is the writedown quarter, so the year-over-year math is flattered by a depressed base. And the guide carries a quiet asterisk: NVIDIA assumed zero data-center compute revenue from China in its outlook. Every sentence about China on Wednesday's call is pure optionality - upside that sits in no model, downside already written to zero.
The call matters as much as the tables. Here's Jensen Huang in February, setting up this quarter:
"Computing demand is growing exponentially - the agentic AI inflection point has arrived. Grace Blackwell with NVLink is the king of inference today - delivering an order-of-magnitude lower cost per token - and Vera Rubin will extend that leadership even further."
We've tracked Jensen's language quantitatively since GTC 2025, and the pattern holds: what he repeats gets funded; what he drops is priced in. Wednesday, listen for whether the "agentic inflection" framing survives an analyst Q&A about customer returns - and whether Vera Rubin's ramp gets a date or just an adjective.
The Trap: A Beat Settles Nothing
On May 11, Polymarket opened its usual market on the print: will NVIDIA report adjusted EPS above the Street's $1.77 (the threshold fixed when the market launched)? Traders price the beat as a near-certainty - just as they did in February, when the equivalent contract sat at 95%.
In February, the 95% crowd was right. NVIDIA beat by about 5%. The stock fell 5.5% the next day.
That isn't a paradox. That's the definition of priced in.
Look at the shape of that chart. In 2023, NVIDIA was beating estimates by 18%, 32%, 19% - Wall Street simply couldn't model a revenue ramp with no modern precedent, and the stock gapped up 24% in a day. Three years later, the last four beats average about 5%. The Street has learned to forecast NVIDIA. And as the beats compressed, the reactions inverted: the stock has now dropped the day after four of the last five reports - every single one of them a beat.
Most people think "beating earnings" is one number. The market grades at least three - the reported quarter, the margin, and the guide - and then grades itself on what was already positioned before the print. Twelve consecutive beats with six next-day drops is the tape telling you the bar isn't the analyst estimate. The bar is perfection, plus a surprise.
The Other Side of the Print
Everything above could read as bull-case homework, so here is the short side of the ledger, with names attached.
| The bear |
The position |
The argument |
| Michael Burry (Scion) |
Put options against NVDA and Palantir, disclosed in late-2025 filings, hundreds of millions in notional |
Big Tech stretches server depreciation, overstating profits; returns on AI capex are falling; expects a 2026 - 27 reckoning |
| Peter Thiel (Thiel Macro) |
Sold all 537,742 NVDA shares in Q3 2025 - roughly 40% of the fund's book |
Rotated to Apple and Microsoft; sees better risk in AI's landlords than its arms dealer |
| The circularity critics |
- |
NVIDIA has committed $40+ billion of equity to AI companies - about seven multibillion-dollar stakes in public companies and roughly two dozen private rounds, many of them customers |
That last row deserves a sentence. CNBC tallied NVIDIA's equity bets at more than $40 billion as of May 9 - and when a supplier invests in customers who use the money to buy its chips, skeptics hear an echo of dot-com vendor financing. Published bear cases cluster near $150 a share, which at today's price is a bet on multiple compression, not on revenue decline.
And the structural caveat we'd flag even if every bear capitulated: a beat on Wednesday cannot prove the AI trade right. Seven hundred billion dollars of capex pays back over years; one quarter can only prove the trade wrong faster. Capex guides are intentions, not contracts - Alphabet raised its guide three times in 2025 alone, and guides can be cut at exactly the same speed. A clean beat-and-rally doesn't settle the bubble argument. It rolls the question forward 90 days, with the stakes marked up.
What We're Reading Wednesday Night
The framework, in the order we'll check it:
- The margin guide, before anything else. Near 75% means pricing power is intact. A slip with a one-off attached is survivable. A slip with no excuse is the first crack - and the only number on the page that reprices the entire supply chain behind NVIDIA.
- The July-quarter guide, against inflated expectations. Consensus already sits above NVIDIA's own midpoint for the reported quarter, and the new guide faces the same arithmetic. The exact number matters less than whether the staircase keeps steepening.
- The China sentence. Zero is in the model. Anything above zero is free.
- The tape, not the print. After twelve beats and six drops, the first hour of reaction is the cleanest read of positioning you'll get all quarter. If a beat sells off again, that isn't noise - that's the market showing you what was already owned.
Polymarket has settled the boring question; NVIDIA will very probably beat. The interesting question is the one every AI valuation quietly assumes an answer to: whether the only company that converts the AI dream into audited revenue - at 75 cents of gross profit on the dollar - can keep doing it at a pace that justifies everyone else's price.
The bubble debate will not end on Wednesday. But this is its best-attended referendum for the next 90 days, and the side that's wrong is the side that's positioned for the old answer.
Data: Barebone | Sources: NVIDIA Q4 FY2026 earnings release and Q1 FY2027 outlook (Feb 25, 2026), Microsoft, Alphabet, Meta & Amazon Q1 2026 earnings calls (Apr 29, 2026), Bloomberg (May 12, 2026), CNBC (May 9, 2026), Fortune (Mar 10, 2026), Polymarket | Prices as of May 18, 2026 close
