Nvidia Crosses $75 Billion in Quarterly Data Center Sales as Jensen Huang Declares the Agentic AI Era Has Arrived

Level 3 — Intermediate

Nvidia closed Wednesday's regular trading session in New York and then released its fiscal first quarter 2027 results, the most closely watched earnings report of the season. Revenue rose to $81.5 billion in the three months ended April 27, with the data center segment, which sells GPUs and networking gear into AI training and inference workloads, contributing $75.2 billion of that total, more than ninety percent of company wide sales and roughly double the year earlier figure. Gross margins held above seventy percent and operating income climbed in tandem.

The company guided to $91 billion of revenue in the current quarter, well above the $86.84 billion consensus tracked by FactSet, and pointedly excluded all China data center compute revenue from that figure, an acknowledgement that the export licensing dispute with Washington remains unresolved. Chief Financial Officer Colette Kress estimated that the addressable Chinese market would be worth roughly $50 billion in calendar 2026 if those restrictions were lifted, while emphasising that visibility into the H200 license process is still limited.

Alongside the operating results, the board announced an $80 billion share repurchase authorization and raised the quarterly dividend from one cent per share to twenty five cents, a 25 fold increase that signals confidence in sustained free cash flow. The combined commitment to return capital comes at a moment when Nvidia is generating cash at an unprecedented rate, with operating cash flow tracking toward $200 billion on an annualised basis.

CEO Jensen Huang framed the results in narrative rather than numerical terms, declaring that the agentic AI inflection point has arrived and that customers are now buying compute not by the GPU but by entire token factories. He highlighted Grace Blackwell with NVLink as the platform of record for inference and called out planned 800 volt direct current data center deployments that he claimed would push another order of magnitude of efficiency into AI workloads. Shares dipped modestly in after hours trading despite the beat, a familiar pattern when expectations run as high as they have for Nvidia in recent quarters.

segment: A separate business unit within a larger company.

inference: Running a trained AI model to make predictions or answer questions.

gross margin: The percentage of revenue left after subtracting the cost of goods sold.

consensus: The average expectation of professional analysts.

addressable market: The total set of customers a business could realistically serve.

authorization: Official permission granted by a board or governing body.

free cash flow: Cash a company generates after operating expenses and investments.

token factory: A data center configured to produce AI generated outputs at scale.

Level 4 — Advanced

Nvidia closed the regular session on the Nasdaq on May 20, 2026, and then printed a fiscal first quarter 2027 result that consolidated rather than merely extended its dominance over the artificial intelligence accelerator complex. Revenue rose to $81.5 billion, with data center, the segment that monetises the GPU, the InfiniBand and NVLink fabrics, and the increasingly important Spectrum-X Ethernet stack, contributing $75.2 billion. That figure roughly doubled the comparable year earlier print, exceeded the FactSet consensus of $73.13 billion, and lifted segment level operating margins above the prior peak set the previous summer. The forward guide of $91 billion of total revenue for the July quarter, against a Street consensus of $86.84 billion, raised the implied annualised data center run rate above $360 billion and explicitly excluded any contribution from advanced chip shipments to mainland China.

The exclusion is more than an accounting curiosity. Chief Financial Officer Colette Kress, repeating language she has used on each of the last four calls, said that visibility into Bureau of Industry and Security licensing for the H200 and successor products remains limited and that customer specific entity list deltas continue to drive intra quarter forecast volatility. Kress placed the addressable mainland Chinese accelerator market at approximately $50 billion in calendar 2026, a figure Bernstein analyst Stacy Rasgon flagged as consistent with the order book Nvidia would lose in a worst case licensing outcome, and as upside in any outcome that approximates the Trump administration's recent post Beijing summit signalling.

Capital return was the headline that surprised buy side desks more than the operating result. The board approved a fresh $80 billion repurchase authorization, on top of the $50 billion programme announced last August that is approximately ninety percent exhausted, and lifted the quarterly cash dividend from $0.01 per share to $0.25 per share. The 25 fold dividend increase, the largest single step jump by any Magnificent Seven constituent since Meta initiated its first dividend in early 2024, telegraphs a willingness to convert escalating free cash flow into formal capital return rather than reinvest it into balance sheet equity in private AI labs, where Nvidia has accumulated more than $40 billion of positions over the past eighteen months.

On the conference call, CEO Jensen Huang reached for the rhetorical register he has reserved for major platform transitions. He declared that the agentic AI inflection point has arrived, characterised the Grace Blackwell platform as the king of inference with an order of magnitude lower cost per token, and previewed an 800 volt direct current data center reference architecture that he claimed would reduce per token energy intensity by another order of magnitude over the planning horizon. Huang also confirmed that Vera Rubin sampling is on schedule for the second half of calendar 2026, a comment that sell side desks parsed as implying Vera Rubin Ultra and Feynman are progressing on roughly the cadence laid out at the prior year's GTC. Shares were volatile in after hours trading, surrendering early gains and closing the extended session modestly lower in a pattern that quantitative desks have begun calling the post print fade, a reflection of how high the implied bar has climbed.

accelerator: Specialised hardware that speeds up specific computing tasks.

fabric: A high speed interconnect that links many computing nodes together.

entity list: A U.S. government list of foreign parties subject to export restrictions.

buy side: Investment firms that buy securities for their portfolios, such as mutual funds and hedge funds.

telegraph: To signal an intention or future action in advance.

balance sheet: A company's snapshot of assets, liabilities, and shareholder equity at a point in time.

sampling: The early shipment of small quantities of a chip to lead customers.

implied bar: The level of performance that investors already assume in a stock's price.

Level 1 — Absolute Beginner

Level 2 — Elementary

Level 3 — Intermediate

Level 4 — Advanced

Nvidia Crosses $75 Billion in Quarterly Data Center Sales as Jensen Huang Declares the Agentic AI Era Has Arrived

Multiple Choice

True or False

Fill in the Blank