Skip to main content
Invest AI
From market noise to cited evidence

Hot-Topic InvestigationsIssue No. 6

Public research

NVIDIA Reports Today. The Bear Case Is Loud — So I Went Back to the Filings.

Independent investigation · Published November 17, 2026 · Evidence as of November 17, 2026 (before the release)

NVIDIA has not reported yet. The Q3 FY2027 numbers — the quarter that ended October 25 — come out today, Tuesday, November 17, after the close. Everything being said about this report today, bullish or bearish, is said about numbers that do not exist yet. So this note is a preview, and it keeps three kinds of figures separate: actual results from last quarter, guidance the company gave in writing, and analyst estimates, labeled as estimates. I hold NVDA, and this is the checklist I am using today. It is built from the company's own filings and earnings call, not from anyone's highlight reel.

NVIDIA Q3 FY2027 earnings preview — the bear case checked against the filings before the numbers
NVIDIA reports Q3 FY2027 today, Tuesday, November 17, 2026, at about 4:20 p.m. ET. This preview was published before the release; every yardstick on it is one the company set itself.

What we actually know: last quarter was a beat

On August 26, NVIDIA reported Q2 FY2027. Revenue was $96.2 billion, up 106% from a year ago — against its own guidance of $91.0 billion, a beat of about $5 billion. Data Center revenue was $89.0 billion, up 117% year over year — roughly 92 cents of every dollar the company takes in. Gross margin was 75.0%, right on the guided 74.9–75.0%. Non-GAAP earnings were $2.22 a share, versus a consensus near $2.09 at the time.

Then it guided Q3 to $108.0 billion in revenue, plus or minus 2%, with gross margin of 74.0%, plus or minus 50 basis points — assuming zero Data Center compute revenue from China. Analyst consensus has since settled just above the guide, at about $108.8–109.0 billion in revenue and $2.47–$2.49 a share (Zacks and S&P Capital IQ estimates, mid-November). In other words: the bar for today is "hit your own number, maybe a little more." Nothing exotic.

  • The guide (company, August 26): Q3 revenue $108.0B ±2%, gross margin 74.0% ±50bp, no China Data Center compute revenue assumed.
  • The consensus bar (estimates): revenue about $108.8–109.0B, EPS $2.47–$2.49 — Zacks and S&P Capital IQ, mid-November.
  • The Q4 bar the market will judge against: revenue about $124.7B (S&P consensus).

The loud claims, checked one by one

The bear case on this print is loud, and parts of it quote the company correctly. So each claim gets the same treatment: what exactly is being said, what the filings and the call transcript actually contain, and what is left over once the two are compared.

#ClaimVerdictBasis
1"Gross margin is rolling over" — the bears' lead argumentTrue — and it was the company's own guidanceCFO Colette Kress on the August 26 call: memory pricing had risen more than expected and was "headed even higher into next year"; the margin path was reset in plain language — 74% in Q3, bottoming at 71–72% in Q4, settling at 72–73% in fiscal 2028.
2"Orders stretching into 2028 may not convert" — demand is crackingNo company evidence behind itCompany documents point the other way: $1T of Blackwell-plus-Rubin revenue visibility through the end of calendar 2027 (May); fiscal 2028 revenue growth guided at roughly 70% and described as capped by supply, not demand (August); supply commitments up from $119B to $279B in one quarter.
3"Wall Street downgraded NVIDIA in a wave last week"A real wave — with wrong labels on two big namesActual downgrades: Aletheia, Baird, Bernstein, Truist, Craig-Hallum, Evercore ISI, Oppenheimer, HSBC. But Morgan Stanley kept its Overweight rating and $300 target, and Citi kept its Buy. The trigger in the cautious notes was the leadership change announced November 12 (Form 8-K), not the operating numbers.
4"Cathie Wood warned NVIDIA's GPU share could fall to 38%"The number is real; the warning label is notYahoo Finance interview, November 12 (via TipRanks): 38% was Wood's self-described highly conservative case for GPU share five years out; her base case was about 53%, her model still produced roughly 25% annual returns even at 38%, and ARK had recently been buying the stock.

1. "Gross margin is rolling over" — the bears' lead argument

True — and it was the company's own guidance

The loudest bear point right now is gross margin, and on the facts, the bears are quoting the company correctly. CFO Colette Kress said on the August call that memory pricing had risen more than expected and was "headed even higher into next year," and she reset the path in plain language: Q3 at 74%, Q4 bottoming at 71–72%, and fiscal 2028 settling at 72–73%, once price increases already put in place take effect.

For context, margin has sat at roughly 75% for three straight quarters (Q4 FY26 through Q2 FY27). So yes, there is a step down coming, and it is not a rumor — it is guidance, three months old, with the cause (memory costs) and the planned offset (higher prices on systems shipping in early 2027) both spelled out.

What that means for today: Q3 printing near 74% would not be a miss or a surprise. It would be the plan. The real test is whether management repeats the same path for Q4 while guiding revenue — the market's working bar for Q4 revenue is around $124.7 billion (S&P consensus). A Q4 guide near that level, with the margin path reaffirmed, is a very different story from the same margin path paired with a soft revenue guide. That pairing is the whole ballgame today.

2. "Orders stretching into 2028 may not convert" — demand is cracking

No company evidence behind it

The second bear point is that orders stretching into 2028 may not convert — that AI cloud customers need financing, land, power, and finished buildings before booked demand becomes NVIDIA revenue. As a question, it is fair. As a claim that orders are already slipping, I could not find a single company statement behind it. Here is what the filings actually say.

In May, the company put its Blackwell-plus-Rubin revenue visibility at $1 trillion through the end of calendar 2027. In August, it did something it had never done: guided a full year ahead, calling for roughly 70% revenue growth in fiscal 2028 — and said plainly that the number is capped by supply, not demand. Kress: customer forecasts point to demand well above what the supply chain can deliver.

What has ballooned is NVIDIA's own commitments. Supply commitments jumped from $119 billion to $279 billion in one quarter, mostly memory purchases for Vera Rubin. Add partner programs and guarantees, and BofA's analyst tallied roughly $500 billion of total commitments on the August call. That is real exposure: if demand ever did soften, those obligations are the risk. It is also, today, the price of locking in scarce memory. Both things are true at once.

There was also a Taiwan media report on November 12 claiming next year's orders could fall 30%. It had no named source, and the company has not confirmed anything like it. I am treating it as a rumor to be answered on the call, not as a fact.

3. "Wall Street downgraded NVIDIA in a wave last week"

A real wave — with wrong labels on two big names

You may have seen that Wall Street "downgraded NVIDIA in a wave" last week. The wave is real. Some of the labels attached to it are not.

Actual downgrades: Aletheia (to Hold, November 9), Baird (to Neutral, target cut from $500 to $200), Bernstein (to Market Perform, target halved to $200), Truist, Craig-Hallum, Evercore ISI, Oppenheimer, and HSBC (to Hold). That is a genuine pile-up.

But two names often headlined as downgrades did not downgrade. Morgan Stanley kept its Overweight rating and $300 target — it only removed NVIDIA from a preferred-stock list. Citi kept its Buy rating, cutting its target from $285 to $225. Bank of America kept a Buy with a $350 target. And the trigger for most of the cautious notes was the leadership change the company announced on November 12 — its 8-K says Jensen Huang resigned on November 11 and CFO Colette Kress is serving as interim CEO — not the operating numbers, which last printed as a beat.

4. "Cathie Wood warned NVIDIA's GPU share could fall to 38%"

The number is real; the warning label is not

One more label worth checking: the "38% share" warning attributed to Cathie Wood. She did say the number — on Yahoo Finance, as her self-described highly conservative case for GPU share five years out. Her base case was about 53%, her model still produced roughly 25% annual returns even at 38%, and ARK had recently been buying the stock. A stress case is a useful thing to know. It is not a forecast.

Flagged, and not used as facts

  • The Data Center consensus of roughly $101 billion for Q3 rests on a single source (Zacks, November 13). It appears in the checklist below labeled as such, and nowhere else on this page.
  • Finnhub's estimate table (revenue $111.2B, EPS $2.52) sits well above the Zacks / S&P Capital IQ / SentiSense cluster. It was not used.
  • The maturity split of the $279B supply commitments (about $92B / $87B / $88B across fiscal 2027–2029) comes from a secondhand reading of the Form 10-Q, not from our own reading of that table. It is not relied on anywhere on this page.
  • The November 12 Taiwan media claim that next year's orders could fall 30% has no named source and no company confirmation. It is recorded as a rumor for the call to answer, not as a fact.

Where the real risk sits

If the loud claims mostly dissolve into guidance and labels, what is genuinely worth a holder's worry? Three things, all visible in the company's own documents.

The commitment book

Supply commitments jumped from $119 billion to $279 billion in one quarter, mostly memory purchases for Vera Rubin. Add partner programs and guarantees, and BofA's analyst tallied roughly $500 billion of total commitments on the August call. That is real exposure: if demand ever did soften, those obligations are the risk. It is also, today, the price of locking in scarce memory. Both things are true at once.

The Q4 pairing

Q3 printing near 74% would not be a miss — it would be the plan. The real test is the pairing: a Q4 revenue guide near the market's roughly $124.7 billion bar, with the 71–72% margin bottom reaffirmed, is a very different story from the same margin path paired with a soft revenue guide.

Whether the guideposts move

In August the company guided a full year ahead for the first time: roughly 70% revenue growth in fiscal 2028, capped by supply. If that guidepost is repeated today in the same words, the story I bought is unchanged. If the guideposts move, that is the story.

My read

The honest risk is on the balance-sheet side — commitments taken on at peak memory prices — not evidence of an order book cracking. Those are different problems, and only one of them is in the filings.

Why I own NVDA — the full holder note (EP12)

The bottom line, before the numbers

  1. Margins: the step down from about 75% to 74%, then to a 71–72% bottom, is real — and it is three-month-old company guidance, with the cause (memory costs) and the offset (price increases) already spelled out.
  2. Orders: no company statement supports orders slipping. The company's own words are $1 trillion of visibility through calendar 2027 and fiscal 2028 growth capped by supply, not demand.
  3. Downgrades: the wave is real; two of its most-headlined names (Morgan Stanley, Citi) did not downgrade, and the trigger was the November leadership announcement, not the operating numbers, which last printed as a beat.
  4. The 38% share figure is a self-described stress case inside a model that still returned about 25% a year — context the headline dropped.

What would change my mind

Four checks, all against yardsticks the company set itself. If the numbers land on the guide and the guideposts do not move, the fundamentals story is the same one I bought — a company growing into a margin reset it announced in advance. If the guideposts move, that is the story, and I will write that one instead.

What I'll be checking today, in order

  • Q3 versus the guide: revenue against $108.0 billion ±2%, margin against 74.0% ±50bp. Data Center against the roughly $101 billion consensus estimate (Zacks, single source).
  • The Q4 guide: revenue against the roughly $124.7 billion market bar, and whether the 71–72% margin bottom is reaffirmed, raised, or quietly dropped.
  • Fiscal 2028: does the roughly 70% growth guide get repeated, in the same words?
  • Commitments: any update on the $279 billion supply commitment figure, and how much of it is memory.

How this note was built

Every figure was checked against at least two sources before publication — company filings and the official earnings-call transcript first. Where sources disagreed, the figure was dropped, or the disagreement is flagged on the page. Analyst figures are labeled as estimates, with the source and date. The circulating video that prompted this check was used as a topic lead only; none of its figures, structure or conclusions are used. Nothing is fetched live, and no figures will be added after publication — a post-release follow-up, if any, will be a separate page.

Sources

Company documents first; estimates and reporting labeled with dates. Every figure on this page was checked against a verification dossier built on November 16–17, 2026, from the filings and the sources below.

  1. NVIDIA Q2 FY2027 results release (Form 8-K Ex. 99.1), August 26, 2026 — revenue $96.221B, Data Center $89.023B, gross margin 75.0%, Q3 guidance $108.0B / 74.0% tickergrove.com
  2. NVIDIA Q2 FY2027 earnings call transcript (investor.nvidia.com), August 26, 2026 — Kress on memory pricing, the 74% → 71–72% → 72–73% margin path, fiscal 2028 ~70% growth as a supply-constrained outlook, supply commitments investor.nvidia.com
  3. NVIDIA Q1 FY2027 results release and CFO Commentary, May 20, 2026 — Q1 actuals, Q2 guidance ($91.0B, 74.9%/75.0% margin), $1T visibility restated investor.nvidia.com · CFO Commentary (PDF)
  4. NVIDIA Q3 FY2026 results release, November 19, 2025 — GAAP gross margin 73.4% (margin trend) investor.nvidia.com
  5. NVIDIA Form 8-K, November 12, 2026 — leadership announcement (resignation effective November 11; Kress interim CEO). Fact only; no reason was given in the filing and none is offered here
  6. Zacks, November 12–13, 2026 — Q3 consensus revenue $108.99B, EPS $2.49, Data Center estimate $101.24B zacks.com · zacks.com
  7. S&P Capital IQ consensus via First Financial (第一财经), November 12, 2026 — Q3 revenue $109.01B; Q4 FY2027 revenue bar ~$124.73B; the Taiwan order-rumor's origin smzdm.com
  8. The Street, November 12, 2026 — Bernstein downgrade details; Morgan Stanley Overweight/$300 maintained; BofA Buy/$350 maintained thestreet.com
  9. Motley Fool, November 12, 2026 — Truist and Craig-Hallum downgrades; DA Davidson Buy maintained fool.com
  10. Longbridge / Benzinga analyst-rating roundups, November 12, 2026 — Evercore ISI, Oppenheimer, Cantor, Citi actions longbridge.com · longbridge.com
  11. Insider Monkey, November 2026 — Aletheia downgrade (November 9) and its $165B commitments figure insidermonkey.com
  12. Supply Chain Dive, August 31, 2026 — supply commitments $119B → $279B supplychaindive.com
  13. TipRanks, November 13, 2026 — Cathie Wood interview summary (83% current share estimate, ~53% base case, 38% conservative case) tipranks.com

Checked and not used

  • Finnhub's Q3 estimate table (revenue $111.203B, EPS $2.523) — an outlier against the Zacks / S&P / SentiSense cluster; not used.
  • The November 12 Taiwan media claim that next year's orders could fall 30% — no named source, no company confirmation; recorded as a rumor, not used as a fact.
  • Any account of why the November leadership change happened — this page records only the Form 8-K fact (announced November 12, effective November 11) and nothing beyond it.
  • A hyperscaler split of Data Center revenue (about $49B of $89B) circulating in commentary — a single secondhand source; not used.

Evidence as of November 17, 2026, before the 4:20 p.m. ET release. Estimates are as of the dates cited and can change.

This is my personal research notebook, not investment advice, not a recommendation, and not a promise of returns. Figures from company filings are actuals; figures labeled consensus or estimates are analysts' numbers as of the dates cited and can change. Holdings change, and prices fall.

Full disclaimer →

Weekly digest: AI-map layer changes and new public 13F filings, each with its source. No score, no grade, no trade signal. Unsubscribe anytime.

Browse stock research →