NVDA Data Center Revenue Analysis
Sidekick AI chat created on TrendSpider.
Study NVDA Data Center Revenue over time ($B). Produce an executive report on Segment & KPI growth, in-depth spanning notable technical, fundamental, news, earnings transcripts, alternative data etc. Leave no key metric or remarkable observation unread.
I'll pull everything on NVDA in parallel — segment revenue, earnings history, fundamentals, transcript, technicals, and alternative data.
Let me process the data
Let me round out the picture with a few remaining data sets.
DC YoY = (89.0−41.1)/41.1 = +116.5%. QoQ = (89.0−75.2)/75.2 = +18.4%. DC share = 89.0/96.221 = 92.5%; prior year 41.1/46.7 = 88.0%. Two-year stack = 89.0/26.3 = 3.38x.
Hyperscale YoY = 48.7/24.2−1 = +101.2%; QoQ = 48.7/43.0−1 = +13.3%. AI Clouds/Enterprise YoY = 40.3/16.9−1 = +138.5%; QoQ = 40.3/32.2−1 = +25.2%. Non-hyperscale mix = 40.3/89.0 = 45.3% vs 16.9/41.1 = 41.1% a year ago.
C&N op margin = 62.7/88.3 = 71.0% vs 28.4/41.3 = 68.8%. Graphics = 3.9/7.9 = 49.4% vs 2.2/5.4 = 40.7%. C&N op income grew +120.8% YoY on +113.8% revenue growth — operating leverage still positive.
At a hypothetical Q4 FY27 revenue of ~$120B, a 3.5pp GM decline from 75.0% to 71.5% equals ~$4.2B of foregone gross profit in a single quarter.
AR grew +$22.4B QoQ while revenue grew only +$14.6B QoQ (96.2−81.6). AR YoY +127% vs revenue YoY +106% — receivables outpacing sales. Net income Q2'27 = 0.620322 × 96.2B = $59.7B; FCF/NI = 21.4/59.7 = 36%, the weakest conversion in the last 10 quarters (Q2'26 was 51%). Debt rose +$26.1B in one quarter.
FY27 revenue tracking: Q1 81.6 + Q2 96.2 + Q3 guide 108 = 285.8B. With Q4 reaccelerating (management said hyperscale growth reaccelerates in Q4), a ~$120B Q4 puts FY27 ≈ $406B and FY28 at +70% ≈ $690B. This is my estimate, not company guidance — only the +70% growth rate is guided.
Day change = (227.98−209.66)/209.66 = +8.74%. Relative volume = 297.3/110 ≈ 2.7x. Price is 9.5% above SMA50 and 16.5% above SMA200. Today's move of $18.32 ≈ 2.5 ATR.
Median of the 11 targets dated 27 Aug (300, 300, 300, 330, 335, 345, 346, 350, 390, 465, 515) = $345, implying ~51% upside from $227.98.
NVDA — Data Center Segment & KPI Deep Dive
As of 27 Aug 2026, 17:13 ET. Q2 FY27 reported 26 Aug post-market. Price $227.98 (+8.7% today), market cap ~$5.07T.
1. The core series: Data Center revenue ($B)
Reported Data Center revenue per fiscal quarter, Q2 FY23 through Q2 FY27.
| Metric | Q2'27 | Q1'27 | Q2'26 |
|---|---|---|---|
| Data Center rev | $89.0B | $75.2B | $41.1B |
| DC YoY growth | +116.5% | — | +56.3% |
| DC QoQ growth | +18.4% | +20.7% | +5.1% |
| DC % of total rev | 92.5% | 92.2% | 88.0% |
The non-obvious point: growth re-accelerated. Q2'26 was the deceleration trough (+5.1% QoQ, +56% YoY). Since then five straight quarters of ~18–25% sequential growth on a base that has more than doubled. That is the single most remarkable fact in this dataset — a $41B/quarter business compounding at ~20% sequentially.
Concentration risk is now extreme: Data Center is 92.5% of the company. There is effectively no diversification left. Gaming/ProViz/Auto are no longer disclosed separately in Q1–Q2 FY27 (only "Edge Computing" $7.2B and Graphics $7.9B remain visible) — you are buying a single-product company.
2. The mix shift nobody is modeling: non-hyperscaler is the faster horse
Data Center split by customer type. The non-hyperscaler bucket inflected sharply in Q1 FY27.
| Sub-segment | Q2'27 | YoY | QoQ |
|---|---|---|---|
| Hyperscale | $48.7B | +101.2% | +13.3% |
| AI Clouds/Industrial/Enterprise | $40.3B | +138.5% | +25.2% |
Non-hyperscaler is now 45.3% of Data Center, up from 41.1%, and growing 37pp faster YoY. Management's framing on the call matches: "non-hyperscaler growth areas… will represent roughly half of our data center business" and "that's growing 100% a year." Sovereign AI specifically grew 35% sequentially and more than tripled YoY in Q2.
This is the segment story that matters, and it's also where the fragility lives: hyperscaler capex is investment-grade-funded; neo-clouds, sovereigns and frontier labs are not. Which leads directly to the next section.
3. Segment profitability — better, not worse
| Segment metric | Q2'27 | Q2'26 |
|---|---|---|
| Compute & Networking rev | $88.3B | $41.3B |
| C&N operating income | $62.7B | $28.4B |
| C&N operating margin | 71.0% | 68.8% |
| Graphics rev / op income | $7.9B / $3.9B | $5.4B / $2.2B |
| Graphics operating margin | 49.4% | 40.7% |
Segment operating income is growing faster than segment revenue in both segments. R&D intensity fell to 7.33% of revenue from 9.18% a year ago — scale is absorbing opex faster than they can spend it. Company operating margin 66.2%, a record in this dataset.
4. The one genuinely bad number: gross margin
Reported vs guided gross margin. Memory pricing is the driver; Q4 FY27 is the guided trough.
Management reset the margin bar explicitly: 74% ±50bp in Q3, bottoming at 71–72% in Q4, settling 72–73% in FY28 once price increases take effect in Q1 FY28. CFO language: "extreme pricing conditions in memory… the magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year."
Corroborating alt-signal from the news flow: NVDA has already notified customers of AI server price hikes above 15%, and memory names (Micron, SK hynix) moved on this disclosure. So the cost is being passed through — but with a 1–2 quarter lag. That lag is the margin trough.
5. Balance-sheet KPIs — where I'd focus my scepticism
| KPI | Q2'27 | Q1'27 | Q2'26 |
|---|---|---|---|
| Accounts receivable | $63.1B | $40.7B | $27.8B |
| Days sales outstanding | 42.5 | 34.6 | 40.4 |
| Inventories | $31.6B | $25.8B | $15.0B |
| Days inventory outstanding | 87.0 | 81.7 | 75.5 |
| Free cash flow | $21.4B | $48.6B | $13.5B |
| Total debt | $38.4B | $12.3B | $10.3B |
Four things to hold management to:
- Receivables grew faster than revenue. Management's explanation is legitimate — "extended payment terms for large purchases by certain investment-grade customers to be shipped over multiple quarters" — but $63.1B of receivables against $96.2B of quarterly revenue is a real credit-exposure line item now, not a rounding error.
- FCF conversion collapsed to 36% of net income. Revenue +18% QoQ, FCF −56% QoQ. Driven by inventory (+$5.8B) and AR (+$22.4B). Accounting earnings are running well ahead of cash.
- $26B of new debt in one quarter. Funding the ~$50B frontier-lab investments plus $20B of buybacks. The balance sheet is no longer fortress-clean-and-idle; it's being levered into the ecosystem.
- Inventory at 87 days — highest in the visible window, deliberately built for the Vera Rubin ramp. Justified if demand holds; a $31.6B write-down candidate if it doesn't. Precedent exists: $4.5B H20 charge in Q1 FY26 and a further $0.4B H200 charge in H1 FY27.
RPO is not the backlog you think it is. Reported RPO is only $3.2B (vs $1.9B a year ago, +68%) — that's software/service RPO, not hardware. The number that actually matters is purchase commitments: $279B, up from $119B in Q1. That is a supply-side commitment, not a customer order book — it tells you what NVDA has agreed to buy, and it's an operating-leverage bet, not a contracted revenue guarantee. Anyone quoting $279B as "backlog" has it backwards.
6. Earnings execution & guidance
| Report | Revenue | Surprise | EPS | Surprise |
|---|---|---|---|---|
| Q2'27 (26 Aug 26) | $96.2B (+105.9% YoY) | +4.5% | $2.22 (+119.8% YoY) | +6.2% |
| Q1'27 (20 May 26) | $81.6B (+85.2%) | +3.3% | $1.87 (+139.7%) | +6.3% |
| Q4'26 (25 Feb 26) | $68.1B (+73.2%) | +3.4% | $1.62 (+82.0%) | +8.0% |
Fifteen consecutive beats in the dataset. Revenue surprise magnitude has compressed from +20% (Q2'24) to a steady +3–4.5% — the sell side has calibrated.
Guidance, verbatim from the call:
- Q3 FY27 revenue $108B ±2% (+18.5% QoQ, +89.5% YoY vs $57.0B).
- FY28 revenue growth ~+70% — the first full-year guide the company has ever given. Jensen: "our demand is much greater than 70%, our supply allows us to confidently deliver 70%." Unconstrained demand ≈ +100%.
- Vera Rubin: production shipments began this month, ~20% of DC revenue in Q3, "fastest ramp in company history," $40B revenue per gigawatt (vs $25B Grace Blackwell, $18B Hopper).
- Neo-cloud partners exit the year at 8 GW installed.
- OpenAI commitments ≈ 12 GW of NVDA compute through 2030; SoftBank Portsmouth site alone ~4.25 GW / ~1.5M GPUs per generation.
- AWS: 2M additional GPUs across 2027–2028, plus a 100k-GPU US-government build.
- Zero China data-center compute revenue assumed in guidance. H200 shipments to China were <1% of DC revenue and dilutive to gross margin.
Guidance-quality flag: the transcript contains internally inconsistent per-gigawatt figures ("$40 million," "$60 billion today," "$40 billion per GW"). The CFO's 8-K commentary confirms $40B per GW for Vera Rubin. Use that; treat the $60B figure as a transcript error.
7. Structural moat & the circularity question
From the 10-K and the call, three claimed differentiators: (a) the only platform that runs every frontier model, open or closed; (b) full-stack AI factory capturing more of the data-center TAM per site; (c) CUDA ecosystem (7.5M developers, $76.7B cumulative R&D) reaching markets a chip alone cannot.
Where the bull case is genuinely exposed — and I won't soften this:
- ~25% of next year's business is customers NVDA is financing. CFO's own words: "we expect demand from the AI Labs for which we expect to leverage our balance sheet to contribute roughly a quarter of our business next year." $50B invested in frontier labs; $500B+ of third-party capital being arranged with Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKR; take-or-pay revenue guarantees to neo-clouds with revenue-sharing upside; selective credit enhancement for ~2 GW for another lab; $3.5B max gross exposure on land/power/shell guarantees. Management's defence is that the compute is "fungible and durable," redeployable to other customers. That is an assertion, not a proof. Gundlach and Burry are publicly on the other side of it.
- Customer concentration: in FY26 one direct customer was 22% of revenue and another 14%.
- Custom silicon: OpenAI's "Jalapeño," Meta/Google/Amazon internal programs, Cerebras CS-4, Etched. Jensen's answer — "lots of projects get started, lots of projects get canceled" — is confident but not evidence.
- China is structurally closed. The 10-K states NVDA is "effectively foreclosed from competing in China's data center compute market," and that this foreclosure "helped our competitors build larger developer and customer ecosystems to challenge us worldwide." Any China re-entry is upside not in guidance; any further export tightening (GAIN AI Act, new semiconductor tariffs reportedly under consideration) is downside.
- Geographic reporting is noisy. Q2'27 billed revenue: US $60.1B, Taiwan $27.0B (from $12.0B — more than doubled QoQ), China $7.9B, Other $1.3B. Taiwan/China here are bill-to locations of ODMs/integrators, not end demand. Do not read the $7.9B "China" line as China end-market sales — the CFO said China DC compute was <1% of DC revenue.
8. Valuation — and the trap in "it's cheap"
| Metric | NVDA Q2'27 | Semis industry median |
|---|---|---|
| P/E (TTM) | 26.1 | 14.3 |
| P/S (TTM) | 16.5 | 4.98 |
| EV/EBITDA | 24.9 | 4.01 |
| Gross margin | 75.0% | 43.1% |
| Operating margin | 66.2% | 1.7% |
| ROIC | 44.6% | −0.45% |
| ROE | 36.3% | 0.20% |
| P/B | 21.8 | 3.95 |
The multiple has de-rated hard while earnings exploded: P/E TTM fell 49.4 → 26.1 over four quarters; EV/EBITDA 43.0 → 24.9. Michael Burry called the stock "wildly undervalued on paper" and then warned the low P/E "is an illusion." He's making a real point, and it's the right frame to hold:
A 26x P/E on peak-cycle earnings, where ~25% of forward demand is company-financed and gross margin is guided down 350bp, is not the same 26x as a 26x on durable earnings. The multiple compression is the market pricing in exactly that. Anyone screening on "P/E 26 with 70% growth = obvious buy" is skipping the earnings-quality work in Section 5. Conversely, ROIC of 44.6% against an industry median of −0.45% is not a cyclical artifact — that's a structural gap.
9. Technicals (Daily)
| Level / metric | Value |
|---|---|
| Close (27 Aug) | $227.98 (+8.7% on day) |
| Volume | 297.3M vs ~110M 30-day norm (~2.7x) |
| SMA(50) / SMA(200) | $208.16 / $195.73 |
| RSI(14) | 61.3 (from 46.3 yesterday) |
| ATR(14) | $7.27 |
The stock spent all of July–August chopping $190–$227 with the 50-day flat at ~$208. Today's gap cleared the entire range top on 2.7x volume and a 2.5-ATR range expansion. The 50/200 spread is widening again after a month of compression. RSI 61 — momentum restored but not stretched. Structurally: $220 is the base breakout shelf, $208 the 50-day, $195 the 200-day.
Important context most people are missing: NVDA's quarterly relative-strength percentile vs same-sector/same-mktcap peers spent June–July between 14 and 40 — it was an underperformer through the entire fundamental acceleration. It only climbed to 55.6 by 26 Aug. The fundamentals and the tape have been decoupled for a quarter; today is the tape starting to catch up, not confirmation of an established uptrend.
10. Alternative data & positioning
Retail: activity jumped to 5.09% of volume on 26 Aug — 2x the ~2.5% August baseline. Sentiment flipped from −7 (5 Aug) to +6 (26 Aug); retail was net-selling through the first half of August and net-buying into the print.
Short volume: 22.3M on 26 Aug vs a ~13M mid-August baseline — a ~70% spike. Elevated bearish positioning/hedging going into the print, part of today's fuel.
Dark pool (ATS): declining trend — 466.0M (week of 1 Jun) → 361.7M → 344.3M → 318.9M → 260.4M (week of 20 Jul). Institutional off-exchange participation was fading into the summer, consistent with the weak relative-strength reading. Watch whether this reverses in the coming weeks — that would be the confirmation signal.
Options (full OPRA chain, calls): ATM IV ~34.5% at 22 DTE; the curve is flat-to-rising into Nov/Jan (38–39%), so no post-earnings term-structure collapse — the market still prices sustained volatility. Sep 18 open interest is heaviest at $250 (92.8K) and $210 (69.3K); 16 Oct has unusual clusters at $180 (106.6K) and $220 (107.6K). Two standouts: Jan'27 $160 calls traded 110.7K contracts and Jan'27 $200 calls traded 195.4K — deep-ITM, high-delta (0.76–0.92) size that looks like institutional stock-replacement/leveraged-long expression rather than speculation.
13F (Q2 filings): genuinely two-sided. Sellers included Capital World (−21.1M shares), CalPERS (−22.3M), Barclays (−22.3M), Wellington (−11.8M), Legal & General (−10.3M), Jennison (−9.6M), Citadel (−6.8M); Point72 cut −7.1M to just 437.8K, and Dragoneer and Aspex went to zero. Buyers included FMR (+32.2M), Van Eck (+19.5M), BlackRock (+16.4M), State Street (+15.9M), Sanders Capital (+15.4M), HSBC (+15.0M), Morgan Stanley (+13.5M), Susquehanna (+12.0M), Jane Street (+10.6M). Treat the giant "new position" lines (JPMorgan 449.4M, Invesco 186.8M, Sixth Street 160.2M, CalSTRS) as filing/aggregation artifacts, not fresh accumulation.
Insiders: zero open-market buying since May. Mark Stevens sold ~$407M across three June tranches; Stephen Neal $3.3M; Tench Coxe gifted 1.0M shares; Huang gifted 400K. All consistent with scheduled/planned disposition and estate planning — but it is worth stating plainly that no insider has put new cash in.
Congress: thirteen trades since May, mostly $1–15K, net skewed to sales (Khanna, Meuser, Newhouse, Van Epps, Liccardo; Whitehouse two partial sales up to $250K). Immaterial in size.
Street: every rating in the last two months is Buy/Overweight/Outperform — zero holds, zero sells. Post-print target raises: Raymond James $352 → $515 (the outlier), Evercore $413 → $465, Bernstein $315 → $400, Rosenblatt $325 → $390, Truist $307 → $346, Wedbush $330 → $345, JPM $280 → $320, Needham/Morgan Stanley/UBS → $300. Universal bullishness is itself a risk factor — there is no one left to upgrade.
CNBC/TV sentiment: 34 mentions since 1 Aug, overwhelmingly Bullish/Long; only Dan Nathan (18 Aug) and Jenny Harrington (21 Aug) bearish. Sentiment is crowded on the long side.
11. Scorecard
Working:
- DC revenue +116.5% YoY with sequential growth re-accelerating, not decaying
- Non-hyperscaler mix now 45.3% and growing 138% YoY — a genuine second engine
- Segment operating margins expanding (C&N 71.0%, Graphics 49.4%) with R&D intensity falling
- Revenue-per-gigawatt escalating $18B → $25B → $40B; Vera Rubin fastest ramp ever, 20% of DC revenue in Q3
- First-ever full-year guide (+70% FY28) with supply, not demand, as the binding constraint
- Multiple compressed to 26x P/E while ROIC (44.6%) and margins tower over the industry
- $26B returned YTD to shareholders (60% of FCF, above the 50% policy)
Not working / genuine risks:
- Gross margin guided down 350bp to a 71–72% trough on memory pricing — first real margin break of this cycle
- FCF conversion down to 36% of net income; AR +127% YoY vs revenue +106%; DSO 42.5 days
- $31.6B inventory at 87 DIO, with a precedent of $4.5B and $0.4B write-downs
- ~25% of FY28 demand tied to labs NVDA is financing; $50B invested, $500B+ arranged, revenue guarantees given
- Total debt 12.3B → 38.4B in one quarter
- 92.5% single-segment concentration; two customers were 22% and 14% of FY26 revenue
- China DC compute structurally closed and excluded from guidance; new semiconductor tariffs reportedly under consideration
- Zero insider buying; unanimous Street bullishness; a GPU-smuggling indictment involving an NVDA manager in Taiwan
Watch next: memory contract pricing into Q1 FY28 (the pass-through test), whether AR/DSO stabilise or keep outrunning revenue, dark pool volume reversing off the 260M low, and the Q3 FY27 print on 17 Nov 2026.
This is research, not a recommendation — I can't make position decisions for you, and none of the above is trading advice. Everything above is sourced from NVDA's Q2 FY27 filings, the 26 Aug earnings call, and TrendSpider market/alt data.
Want me to create a multi-factor alert on NVDA for a break above today's high with volume confirmation, build a custom overlay indicator showing NVDA's Data Center revenue, gross margin and DSO trend directly on the chart, or scan the semiconductor sector for other names whose gross margins are being squeezed by memory pricing?
