Nvidia Earnings Tonight Will Test the $725 Billion AI Capex Bet

Nvidia goes into Wednesday’s print at a $5.38 trillion market value, sitting on a Philadelphia Semiconductor Index that has run 65% year to date and an options chain pricing a roughly 8% one-day move on the close. The company has already told the Street to expect about $78 billion in fiscal first-quarter revenue, plus or minus 2%. Goldman Sachs sits at $80.05 billion. The consensus number is $400 million above Nvidia’s own midpoint, a rare inversion of a company that usually sandbags its own guide.

Read the setup honestly and the print is not really about Nvidia. It is the first hard read on whether the $725 billion in 2026 capital spending that Alphabet, Amazon, Meta and Microsoft committed to on their April calls actually flows through to Jensen Huang’s order book, or whether a meaningful slice gets diverted to Broadcom, custom silicon, and the H2 calendar slips.

The Setup Wall Street Has Already Bought

The chip rally going into tonight is statistically extreme. The PHLX Semiconductor Index (SOX, the benchmark for US-listed chipmakers) closed 17 consecutive sessions higher between March 31 and April 23, the longest unbroken streak in the index’s 32-year history. The SOXX ETF trades roughly 60% above its 200-day moving average. Nvidia itself closed at an all-time high of $235.74 on May 14 before easing to $220.61 by Tuesday’s bell.

The going-in scorecard, drawn from the company’s own February release and the consensus tape:

• $215.9 billion in fiscal 2026 revenue, up 65% year over year
• $68.1 billion reported for Q4 fiscal 2026, up 73% year over year
• $62.3 billion data center segment in the same quarter, up 75%
• ~75% non-GAAP gross margin, guided to hold within 50 basis points

Those are the numbers tonight has to clear without flinching. The reaction function the buyside is watching, per the options-market expected move for the May 22 expiry, prices an 8% gap in either direction, with the 30-day implied volatility running near 45% against realized closer to 30%.

Why the Bar Is Already Above the Guide

The unusual feature of this quarter is the Street has front-run management. Nvidia’s official outlook called for about $78 billion in revenue. Consensus from the published Wall Street estimate tracker sits roughly $400 million higher, with the top end (Goldman) above $80 billion. That gap is small in dollar terms and enormous in signaling terms.

The Sandbagging Pattern Just Broke

Through fiscal 2025 and most of fiscal 2026, the company’s own midpoint was typically several percent below where analysts landed by the print date, and the actual result then beat both. Wall Street learned to model an explicit guidance buffer. Going into Wednesday, that buffer is gone in the opposite direction: the buyside is asking the company to beat a number that exceeds its own confident range.

What Counts as a Real Beat

A merely in-line print, $78 billion delivered, $79 billion guided for the July quarter, is now the bear case. Wedbush analyst Matt Bryson wrote ahead of the close that he expects Nvidia to “again exceed estimates and guide above Street given continued positive data points through Q1 as well as healthy 2026 AI infrastructure spend.” Anything below that script reads as a deceleration signal, even if the absolute growth numbers are still extraordinary by any other company’s standard.

The Asymmetry on the Tape

This is what makes the setup uncomfortable. An $81 billion print with a $90 billion Q2 guide moves the stock maybe 4 to 6 percent higher and validates the SOX. A $77 billion print with an $82 billion guide, technically a small miss against consensus while still landing inside management’s own range, would print as a disappointment and likely take the entire semiconductor complex with it. The options market is paying for the second outcome more than the first.

The $725 Billion Question Behind the Print

The reason this single quarterly report carries so much sector weight is what sits underneath it. Alphabet, Amazon, Meta and Microsoft together guided to roughly $725 billion of 2026 capital expenditure on their first-quarter calls, up 77% from $410 billion in 2025. Add Oracle’s roughly $50 billion target and the four-plus-one group is committed to north of three quarters of a trillion dollars in build-out this calendar year, with Microsoft, Meta and Alphabet between them shouldering more than $400 billion of that.

About 75% of that combined dollar pool, roughly $450 billion by the Futurum Group’s count, is earmarked for AI infrastructure servers, accelerators, networking and physical data center build. That is the addressable pie. Whatever share of it lands in Nvidia’s revenue line is what the model has to defend.

The April 28 selloff, when news that OpenAI had missed internal revenue and user targets sent Oracle down roughly 3% and dragged Nvidia, Broadcom and AMD with it, was the market briefly testing what happens if any link in that capex-to-demand chain wobbles. It snapped back within days. The question Wednesday’s call has to address is whether the snap-back was a buying signal or a setup.

Where Custom Silicon Is Pulling Dollars Away

Of the share of hyperscaler capex that goes to AI accelerators, the consensus assumption has been that Nvidia takes the dominant slice, with analyst estimates clustering around 80% of the accelerator market by revenue. The directionality matters more than the level. Meta is moving more inference workload onto its own MTIA silicon. Microsoft is on its second-generation Maia accelerator. Google’s TPU v5p and v6 are absorbing internal Gemini training. Amazon is scaling Trainium 3.

None of that knocks Nvidia off its perch. All of it shaves a few percentage points off the share assumption that Goldman’s $80 billion number implicitly requires. Broadcom, which designs much of the custom silicon for Google and Meta, has been the second-derivative beneficiary of the same capex wave that lifted Nvidia.

Nvidia’s forward guidance will give investors an idea of whether increased spending from AI companies is going to Nvidia or being scooped up by custom chipmakers, such as Broadcom.

That framing, from the Motley Fool’s earnings preview, captures what the buyside is genuinely uncertain about. The data center segment number tells you the absolute dollars. The Q2 guide tells you whether the share is holding.

The China Line Nvidia Already Wrote Off

One distortion on the tape tonight: management has explicitly told the Street that the $78 billion guide excludes any data center compute revenue from China. That is a clean line, and it matters because the comparable quarter a year ago carried a $4.5 billion charge against H20 (the China-compliant accelerator) inventory, plus another $2.5 billion in H20 revenue the company said it could not ship after the April 2025 export controls hit.

Chief Financial Officer Colette Kress said in May 2025 that without the controls, Q1 fiscal 2026 H20 orders would have run around $8 billion. The Trump administration revoked the Biden-era AI Diffusion framework in early 2025, but the Bureau of Industry and Security (BIS, the Commerce Department arm that controls export licensing) tightened H20 licensing the same week. The result is a structurally zero China contribution in the guide.

If anything leaks back through during the print, from a license grant on the modified H200 variant under the reported 75,000-unit cap and 25% tax framework, it is upside the Street cannot model yet. If the language hardens further, it is the cleanest negative catalyst on the call.

What Wednesday’s Tape Will Show

The four lines on the earnings release that will move the SOX, in order of immediate market sensitivity:

1. Q2 fiscal 2027 revenue guide. Anything below $86 billion at the midpoint reads as a deceleration. Above $90 billion validates the Goldman case and the SOX rally.
2. Data center segment dollar number. The Street will back into accelerator share-of-wallet from the gap between this line and the hyperscaler capex pace.
3. Gross margin. Held at 75% is the bull case. A drift toward 73% as Blackwell yields normalize and Vera Rubin pre-production cost flows in is the early signal that the next architectural transition starts costing money before it starts earning it.
4. Vera Rubin commentary. Jensen Huang confirmed at GTC that the first Vera Rubin rack is running at Microsoft Azure, with full production shipments targeting the back half of calendar 2026. Any slip pushes the AI revenue cliff into a later quarter, which the multiple does not currently price.

Underneath all of that sits the structural read: Nvidia has told the market it sees roughly $1 trillion in cumulative Blackwell-plus-Rubin demand through 2027. That number was double the prior year’s forecast when management put it on the GTC slide. It is also the number that, more than any single quarterly print, dictates whether the chip sector’s 32-year-record run has another leg or just reached the ceiling.

The print closes the books on a quarter the market already paid for. The guide writes the next one.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. Semiconductor equities, options strategies tied to earnings events, and AI-infrastructure exposure carry meaningful capital-loss risk including the possibility of total loss. Readers should consult a licensed financial professional before acting on any view expressed here. All figures are accurate as of publication on May 20, 2026.