Nvidia’s $96 Billion Quarter, Converted Into Watts and Job Openings: What the 2025-2035 Projections Say
Updated September 27, 2026. Every figure below was read on that date from the filing, statistical release or manufacturer page named at the end. Nothing here is taken from a press summary of those documents.
Photo: Wikimedia Commons, CC BY-SA 4.0
The quarter, taken from the filing instead of the headline
NVIDIA reported revenue of $96,221 million for the second quarter of fiscal 2027, the three months ended July 26, 2026, in a release dated August 26, 2026 and in the Form 10-Q filed the same day. So the ninety-six-billion figure is correct, and it belongs to a specific quarter with a specific end date — worth pinning down, because the number gets quoted loose.
| Line | Q2 FY2027 (quarter ended 2026-07-26) | Change |
|---|---|---|
| Total revenue | $96,221 M | +106% year over year, +18% sequentially |
| Data Center | $89,023 M | +117% year over year, +18% sequentially |
| Edge Computing | $7,198 M | +27% year over year, +13% sequentially |
| Gross margin | 75.0% GAAP and non-GAAP | — |
| Operating income | $63,734 M GAAP | — |
| Net income | $59,688 M GAAP / $53,954 M non-GAAP | GAAP above non-GAAP, which is unusual |
| Diluted EPS | $2.46 GAAP / $2.22 non-GAAP | — |
| Next-quarter guidance | $108.0 B, plus or minus 2% | gross margin guided to 74.0% |
Two details in that table are the kind that get lost in a summary. First, NVIDIA stopped breaking out Gaming, Professional Visualization and Automotive as separate dollar lines in this release — everything outside Data Center now appears as a single "Edge Computing" figure. Anyone comparing this quarter's gaming revenue to a year ago cannot do it from the company's own release. Second, GAAP net income exceeds non-GAAP net income, the reverse of the usual pattern, because a $7,771 million gain on equity securities sits inside the GAAP figure and is excluded from the adjusted one. That single line is larger than the entire Edge Computing segment.
Correction: the Federal Reserve does not publish a job map
A companion piece on this site was headlined around "the Fed, AI and the 2035 job map." The Federal Reserve publishes no employment projections of any kind — not by occupation, not by year, not under that or any other name. What exists is a Bureau of Labor Statistics product, Employment Projections 2025-2035, released August 27, 2026, which replaced the previous 2024-2034 cycle. A 2035 horizon is real; it belongs to BLS.
The Fed's contribution to this subject is narrower and worth quoting accurately rather than paraphrasing into a forecast. The FOMC statement of September 16, 2026 does not mention artificial intelligence at all; it notes "strong productivity gains and robust business investment." Individual governors have gone further in speeches. Governor Michael Barr, on July 14, 2026: "As of right now, there has been little evidence of economy-wide job displacement from AI," alongside the observation that AI "might disproportionately affect new entrants to the labor market." Governor Lisa Cook, on May 27, 2026, noted that companies "have announced more than $1.5 trillion in data-center plans, only a small portion of which have been realized," and that "AI-related job loss could precede job gains. Although we do not have conclusive evidence of this occurring yet, it may still be on the horizon." Those are hedged observations by individuals, not institutional projections, and treating them as either a promise or a warning overstates them.
One hour on the rack, one hour on your laptop
A $96 billion quarter is not a unit anybody can feel. Electricity is. NVIDIA's own reference-architecture documentation states that a full GB300 NVL72 rack — the current flagship configuration, carrying 20 TB of HBM3E with up to 576 TB/s of aggregate memory bandwidth — draws up to 142 kilowatts, from eight power shelves of 33 kW each. NVIDIA's current consumer flagship, the GeForce RTX 5090, carries 32 GB of GDDR7 and a stated total graphics power of 575 watts.
Price both at the BLS average-price series for US city average electricity in August 2026, 19.6 cents per kilowatt-hour:
| Machine | Rated draw | Electricity for one hour | A 40-hour week | A full year at full load |
|---|---|---|---|---|
| GB300 NVL72 rack | 142,000 W | $27.83 | $1,113 | $243,808 |
| GeForce RTX 5090 | 575 W | $0.113 | $4.51 | $987 |
| MacBook Air, idle | 3.5 W | $0.0007 | $0.027 | $6.01 |
The multiplication is mine; the wattages are the manufacturers' and the rate is the Bureau's. The ratio that matters is the middle column: one rack-hour costs 247 times what one hour on a top-end consumer card costs, in electricity alone, before the hardware, the building or the cooling. That is the actual shape of the thing a student is reading about. It is not a machine you will have. It is a machine whose output you will rent by the hour, which makes "how efficiently can I use an hour of it" a more useful skill than "can I build one."
One honest caveat on the rate: the Energy Information Administration's Electricity Monthly Update, released September 24, 2026, gives 18.31 cents per kilowatt-hour as the US residential average for July 2026. Using that instead moves the rack-hour to $26.00. The two series differ because EIA divides total utility revenue by total sales while BLS samples an urban consumer basket; neither is the price an operator of that rack would pay, since industrial rates are lower than residential ones. The ratio between the machines is unaffected.
What the 2025-2035 projections actually project
BLS projects total employment rising from 170.3 million in 2025 to 176.2 million in 2035 — 5.9 million jobs, or 3.5 percent, which BLS itself contrasts with "the 10.9 percent growth recorded over the 2015-25 decade." Against that backdrop, here is the computing column, with the wage figures as BLS states them on each occupation's page.
| Occupation | 2025 | 2035 | Change | Median wage | Openings per year |
|---|---|---|---|---|---|
| Data scientists | 275,600 | 371,000 | +35% | $120,230 | ~24,800 |
| Computer and information research scientists | 38,600 | 47,000 | +22% | $140,300 | ~2,900 |
| Information security analysts | 192,900 | 233,400 | +21% | $129,180 | ~14,100 |
| Software developers | 1,717,800 | 1,892,600 | +10% | $135,980 | ~106,100 (with QA analysts) |
| Computer systems analysts | 544,400 | 587,200 | +8% | $105,850 | ~32,900 |
| Computer network architects | 181,800 | 195,800 | +8% | $134,050 | ~9,600 |
| Web developers and digital designers | 220,100 | 231,400 | +5% | $99,520 | ~13,600 |
| Database administrators and architects | 144,500 | 150,900 | +4% | $104,620 / $139,500 | ~7,300 |
| Computer support specialists | 903,100 | 878,800 | -3% | $62,890 | ~48,700 |
| Computer programmers | 110,800 | 102,700 | -7% | $100,390 | ~4,400 |
Read the last two rows next to the first one and the "golden age" framing gets more specific. Two computing occupations are projected to shrink over the decade, and the openings BLS still shows for them come from replacement rather than growth — people leaving, not positions being created. Meanwhile data scientists grow 35 percent from a base of 275,600, which is a large percentage on a small number: 95,400 added jobs against the 174,700 added software-developer jobs. Percentage growth and job count point at different answers, and a student choosing a specialisation is choosing between them.
The denominator nobody puts on the slide
Projected openings only mean something next to the number of people arriving to fill them. The National Center for Education Statistics counts 108,503 bachelor's degrees conferred in computer and information sciences in 2021-22, up from 79,597 in 2017-18 — a 36 percent rise in four years, and a 129 percent rise across the decade from 2011-12.
Set that against BLS's roughly 106,100 annual openings for software developers and quality-assurance analysts combined, and the two numbers land within two percent of each other. That comparison is rough and should be read as rough: computer science graduates fan out across all ten occupations above and into jobs outside them, and not every opening is filled by a new graduate. But the direction is not ambiguous. One year of degree production is the same order of magnitude as one year of openings in the largest computing occupation, and the degree line has been rising faster than the projection line.
Two BLS products, one occupation, two different wages
Anyone checking the wage column above against the Bureau's other publication will find a discrepancy worth naming rather than hiding. The Occupational Outlook Handbook and the Occupational Employment and Wage Statistics news release both describe themselves as carrying May 2025 data, and they disagree:
| Occupation | Handbook median | OEWS release median | Gap |
|---|---|---|---|
| Software developers | $135,980 | $148,100 | $12,120 |
| Computer and information research scientists | $140,300 | $153,930 | $13,630 |
| Computer systems analysts | $105,850 | $114,610 | $8,760 |
| Computer support specialists | $62,890 | $67,330 | $4,440 |
| Web developers | $99,520 | $98,770 | -$750 |
The likely explanation is that the Handbook sometimes reports a broader combined occupational grouping than the single detailed code in the OEWS table, which would explain both the direction and why web developers move the other way. That explanation could not be confirmed from either publication's own text, so it stays a flagged disagreement rather than a resolved one. The practical rule: cite which BLS product a wage came from, and never mix the two in the same table.
What BLS says its new AI measure is not
Alongside the projections, BLS published a new product on August 27, 2026 called AI Exposure Categories, sorting all 831 detailed occupations into low, moderate, high and very high relative exposure from three theoretical and two observed-usage data sources. The Bureau attaches an unusually blunt disclaimer to it: "Exposure does not imply job loss, productivity gains, automation probability, or wage effects," and "An exposure category is not a forecast of employment growth or decline." Its own summary of AI's effect on the projections is similarly two-sided: "Although the growing adoption of AI is expected to support demand for some occupations, associated productivity gains may dampen employment demand for others." An exposure ranking is a description of task overlap, and nothing in it tells a student whether a job will exist.
Four things none of these sources publish
NVIDIA does not disclose Gaming, Professional Visualization or Automotive revenue separately this quarter, and does not publish a memory-bandwidth figure for the RTX 5090 on that card's own product page. It publishes relative performance claims for the GB300 generation — a stated 50x leap in AI-factory output and 5x throughput per megawatt against the previous architecture — but no absolute tokens-per-second or output-per-dollar figure. BLS does not publish 10th or 25th percentile wages in the static release tables, which means there is no clean entry-level wage in any of the figures above: every wage here is a median across all experience levels, and a first job will pay less than all of them. And NCES could not be confirmed to have published a computer-science degree count for any year after 2021-22, so the denominator in the previous section is four years old.
What the two sets of numbers, read together, actually suggest
The hardware side and the labour side are telling a consistent story once you stop reading either one alone. Revenue more than doubled year over year and next quarter is guided 12 percent higher again, which is a capital-spending boom in machines that cost $27.83 an hour to keep powered. The occupations that grow fastest in the decade that boom is supposed to define are the ones that sit closest to those machines and their data — data scientists at 35 percent, research scientists at 22, security analysts at 21. The ones that shrink are the ones whose tasks are most routine and most repeatable: programmers at minus 7 percent, support specialists at minus 3. Meanwhile the labour market underneath all of it is projected to grow 3.5 percent over ten years, against 10.9 percent over the last ten, and the Fed's own governors say there is not yet conclusive evidence of AI displacement either way. If someone were planning a course of study today on the published figures alone, the defensible reading would be that the safest position is not "AI" as a subject but proximity to the expensive machine plus something the projections say keeps growing — and that the single most useful habit is checking the openings-per-year column rather than the percentage column, because 35 percent of a small number is still a small number.
Four numbers to check before choosing a specialisation
- Annual openings, not percentage growth. Data scientists grow fastest; software developers have four times as many openings. Both facts are on the same BLS page and they point in different directions.
- Which BLS product your wage figure came from. The gap between the two official medians for software developers is $12,120, which is larger than most salary negotiations.
- Degrees conferred in your field, most recent year. For computer and information sciences it was 108,503 in 2021-22 and rising. That is the queue you are standing in.
- Whether the growth in your target occupation is growth or replacement. BLS states this explicitly for shrinking occupations. Openings that exist only because people are leaving are a different bet from openings that exist because the work is expanding.
Tip: Renting an hour of someone else's accelerator is cheap; moving the dataset to it repeatedly is what wastes the hour. Keeping working copies on a fast external SSD saves more billable minutes than most optimisations. (These are Amazon Associate links — we may earn a small commission on qualifying purchases.)
Related on this site
Filings and releases these figures were read from
- NVIDIA, financial results for the second quarter of fiscal 2027, August 26, 2026, and the Form 10-Q for the period ended July 26, 2026.
- NVIDIA, GB300 NVL72 and its enterprise reference architecture — the 142 kW figure and the memory specifications; and the GeForce RTX 5090 product page.
- Bureau of Labor Statistics, Employment Projections 2025-2035, released August 27, 2026; the Occupational Outlook Handbook computing pages; the Occupational Employment and Wage Statistics national table; and AI Exposure Categories.
- Bureau of Labor Statistics, Employment Situation, August 2026, released September 4 — the 4.1 percent unemployment rate.
- National Center for Education Statistics, Digest of Education Statistics table 325.35 — degrees conferred in computer and information sciences.
- Federal Reserve, FOMC statement of September 16, 2026; Governor Barr's speech of July 14, 2026; Governor Cook's speech of May 27, 2026.
- Bureau of Labor Statistics average price series for electricity, August 2026, and the Energy Information Administration Electricity Monthly Update of September 24, 2026.
This page summarises published filings and statistical releases and is not personalised career or investment advice; it is written by no one holding a financial credential. Projections are projections, company results are historical, and every figure carries the date it was read.
Comments
Post a Comment