After the U.S. market closed on August 26, NVIDIA reported its Q2 FY27 results: revenue of $96.22 billion, up 106% year-over-year and 18% quarter-over-quarter, marking a 13th consecutive record quarter. Data center revenue came in at $89 billion, up 117% year-over-year. Earnings per share were $2.22, with a gross margin of 75.0%. Guidance for next quarter came in at $108 billion, plus or minus 2%, well above the roughly $104 billion analysts had expected.
On the earnings call, CEO Jensen Huang also laid out revenue expectations for fiscal 2028: growth of about 70% year-over-year. He added two lines: "Actual demand is far in excess of 70%," and "While demand is far in excess of 70%, given our visibility into our supply, we're confident we can achieve 70% growth." The figure, he said, is capped by supply, not by orders.
The buyer list is shifting
One ratio in the earnings report is easy to miss. Hyperscaler revenue grew 13% quarter-over-quarter, and their share of data center revenue slipped from 59% to 55%. Revenue from the bucket NVIDIA labels ACIE — other clouds, industry and enterprise customers — grew 25% quarter-over-quarter and 138% year-over-year.
The four major cloud providers remain NVIDIA's biggest customers, but the incremental growth is coming from elsewhere: sovereign AI programs, industry customers, regional clouds, and a crop of venture-funded AI companies building their own clusters. Combined, these buyers are growing more than twice as fast as the giants. Huang summed up where this demand is coming from with a single line: "Compute is revenue."
Payment terms are this quarter's subplot
In the same report, accounts receivable climbed from $38.5 billion six months ago to $63 billion, up 64% quarter-over-quarter. Days sales outstanding — the average time from invoice to payment — stretched from 45 days to 60 days.
Rough math on what those extra 15 days mean: spread $96.2 billion of quarterly revenue across 91 days and you get about $1.06 billion a day, so 15 days works out to roughly $16 billion in payments still sitting on the books rather than in the bank. That's not a bad-debt problem, but it does show NVIDIA extending more time for customers to pay up.
Who's getting that extra time lines up with the buyer shift described above. The four hyperscalers have never been a payment risk; the newer buyers inside the ACIE bucket are different — a number of AI startups and regional clouds have cash flow that depends on their next funding round. NVIDIA, meanwhile, is carrying $581 billion in supply commitments, power guarantees, leases and equity investments in AI startups — a pile bigger than a full year of its own revenue.
Supply is still the ceiling
What Huang kept coming back to on the call was that supply, not demand, is the bottleneck — there's no shortage of buyers waiting in line. AWS's new order for 2 million additional GPUs is scheduled for delivery in 2027 and 2028, and its payment terms were likewise extended from 45 days to 60.
Annualized at the current run rate, NVIDIA's revenue works out to roughly $432 billion, which would rank sixth among the world's companies; annualized gross profit comes to about $324 billion, good for second place. For a hardware company operating at this scale, the constraint on growth has shifted from "are there buyers" to "can we build enough, and when does the money actually arrive." This quarter's report is the first time the second half of that sentence has been put on the table.
Sources: NVIDIA earnings report and earnings call, ITHome, CocoLoop, Tomasz Tunguz's earnings breakdown; figures cross-checked against the original earnings release for revenue, gross margin, guidance range and receivables.