Nvidia reported its third quarter of fiscal 2027 on 17 November 2026: $115.8 billion in revenue for the quarter ended 25 October. I needed that as a per-second rate for a short data video (a balance scale where coins pour in at the company's real revenue rate). The calculation is one division. Getting it right took three corrections.
Disclosure: I build data tools for AI agents as Fast Data on Apify. This post uses only public figures from Nvidia's press releases and the US Census Bureau.
The calculation
from datetime import date
def per_second(revenue_usd: float, start: date, end: date) -> float:
days = (end - start).days + 1 # both days count
return revenue_usd / (days * 86_400)
q2 = per_second(96.2e9, date(2026, 4, 27), date(2026, 7, 26))
q3 = per_second(115.8e9, date(2026, 7, 27), date(2026, 10, 25))
print(f"Q2: ${q2:,.2f}/s Q3: ${q3:,.2f}/s change: {q3 / q2 - 1:+.1%}")
# Q2: $12,235.45/s Q3: $14,728.33/s change: +20.4%
| Q2 FY2027 | Q3 FY2027 | |
|---|---|---|
| Revenue | $96.2 B | $115.8 B |
| Quarter | 27 Apr – 26 Jul 2026 (91 days) | 27 Jul – 25 Oct 2026 (91 days) |
| Per second | $12,235 | $14,728 |
| Per minute | $734,000 | $884,000 |
| Per day | $1.06 B | $1.27 B |
| One US median household year ($87,460) | 7.1 s | 5.9 s |
The household figure is the US real median household income for 2025 as published by the Census Bureau in September 2026: $87,460. At the Q3 rate, Nvidia's revenue covers it in 5.9 seconds; forty years of it ($3.5 million) take 3 minutes 58 seconds.
Mistake 1: assuming a quarter is 90 days (or a quarter of 365)
Nvidia's fiscal quarters are 13 weeks long and end on the last Sunday of a month, so most are exactly 91 days. Dividing by 365 / 4 = 91.25 days is close; dividing by a round 90 days overstates the rate by 1.1 %, which moved my "seconds per household year" from 5.94 to 5.87. Small, but it changes the rounded on-screen number, and a viewer who checks will find it.
Rule: take the period dates from the release ("quarter ended October 25, 2026") and count days, do not assume.
Mistake 2: the number goes stale in days
My first version of the video used the Q2 rate ($12,235/s) and the result "7.1 seconds". The moment Q3 came out, that was 20 % off. Anything "real time" built on a quarterly figure has a shelf life of one quarter.
Rule: store the source figure, period and publication date next to every derived number, and regenerate from the source, never edit the derived value by hand. In our pipeline, a facts.json holds the claim, the calculation and the sources, and every number on screen is rendered from it.
Mistake 3: "earns" vs "sells"
"Nvidia earns $14,728 a second" is wrong. Revenue is sales, before costs and taxes. The press release reports net income separately, and it is a smaller number. Saying "earns" or "makes" in a headline invites exactly the correction you deserve.
Rule: say revenue or sales, and if there is room, add "not profit". We put "OF SALES, NOT PROFIT" directly under the result in the video.
Bonus: rounding that does not lie
- Round only at the very end. Rounding the per-second rate first ($14,700) and then dividing gives 5.95 s instead of 5.94 s.
- Pick the precision your source supports. The revenue figure is given to one decimal in billions ($115.8 B), so "about $14,700 per second" is honest; "$14,728.33" is arithmetic, not precision. In the table above I kept full numbers because readers may want to re-check them.
- Use "about" for anything derived from a rounded input.
Why bother with per-second numbers at all?
Because nobody can picture $115.8 billion, but everybody can picture six seconds. It is a framing device, and like any framing device it can mislead if the comparison is unfair. Comparing one company's sales to one household's income is a scale comparison, not a moral one; it says nothing about what either should earn.
Try it
The function above is the whole tool. For other data stories built from public sources (Eurostat, the MCP Registry, our own social and video analytics), see the snippets in our data-recipes repo.
Question: what is the most misleading "per second" statistic you have seen? I collect them as examples of what not to do.

Top comments (0)