NVIDIA Corporation (NASDAQ: NVDA), initiation of coverage, Nov 6, 2026

NVIDIA is priced for a margin collapse we do not forecast, but returns are modest

NVIDIA sells most of what goes into an AI data centre, and its data centre sales grew 2.5 times in six quarters. The shares trade at a discount to peers; we expect a 13.9% return, mostly from time.

Our rating
Buy
12-month target
$205.86
Price, Nov 5, 2026
$181.60
Expected return, with dividend
+14%

By Penmark Team

NVDABuyTarget $205.86+14% expectedFull report

What NVIDIA is now

NVIDIA sells most of the AI factory, and data centres are nearly all its sales

Data Center sales rose from $35.6bn a quarter to $89.0bn in six quarters. Each gigawatt of AI data centre now carries about $40bn of NVIDIA content, against $18bn in the Hopper generation.

We expect $409bn of sales this fiscal year. Of that, $266bn reaches operating profit and $191bn becomes free cash flow.

Who buys, FY27 sales we expect
FY27 sales: $409bn$208bn$172bn
Hyperscale cloudsAI clouds, industrial and enterpriseEdge computing

The customer split applies the latest quarter’s hyperscale share to our estimate. FY is NVIDIA’s fiscal year, which ends in January. Source: company filings, Penmark Research estimates.

Data Center sales per quarter, $bn
$36$89Q4’25Q2’26Q4’26Q2’27

Fiscal quarters. Source: company filings.

How NVIDIA makes money (page 8)

What you are really buying

The price is close to our value today, so the return is mostly time

The shares cost 11.4 times analysts’ FY28 earnings, 33% below peers. Three debates explain the whole $7.25 gap between the price and our $188.85 value today.

We are less hopeful than the market on FY29 capacity and more hopeful on margins. Most of the 13.9% return comes from rolling value forward at our 12.5% required return.

From today’s price to our value, per share
$181.60Today’sprice−$15.59Slower FY29capacity+$20.77Marginsfade slowly+$2.07Workingcapital$188.85Our middlecase today

Each step is the value of one debate at our view against the market’s. Source: Penmark Research estimates.

The three debates (page 7)

Where we disagree: margins

Margins fade slowly, not to the 52.9% the price implies

The price only works if gross margin settles at 52.9% for good. That is below the 69.1% Broadcom earns on the custom AI chips that compete with NVIDIA.

We model a fade from 73.7% this year to 68% by FY32 as custom chips and memory costs bite. Holding 68% is worth $20.77 a share.

Long-run gross margin, by view
What the price implies52.9%Bad case62.0%Our forecast68.0%Broadcom custom AI today69.1%Good case71.0%

Case margins are for FY32 onwards. Source: company filings, Penmark Research estimates.

Why margins fade slowly (page 17)

Where we disagree: FY29

We forecast less FY29 growth than analysts, then content carries it

Analysts’ FY29 sales of $903bn imply 21% more capacity shipped. We model 10% as power and permitting bind, so $824bn.

After that, growth comes from content per gigawatt. Vera Rubin’s $40bn is 60% more than Blackwell, and we grow content 5% a year.

Sales, ours against analysts, $bn
$409.0$411.7FY27$695.2$692.2FY28$824.5$903bnFY29
Our estimateAnalysts’ forecast

Analysts’ forecasts are Yahoo Finance consensus. Source: Penmark Research estimates.

Content per gigawatt (page 16)

Where we disagree: financing

Customer financing is large but covered by one year of cash flow

NVIDIA invests in AI labs that buy its chips, which worries bears. Stakes and commitments total $153.5bn, against $191.5bn of free cash flow we expect this year.

Even a 50% loss on all of it would be $77bn, 0.4 times one year of cash flow. Customers took 61 days to pay at the last quarter.

Financing exposure against cash flow
Equity stakes$93.9bnCustomer support commitments$59.5bnFY27 free cash flow$191.5bn

Source: company filings, Penmark Research estimates.

Why financing is contained (page 18)

The odds

A low-conviction Buy: the bad case sits $59 below the price

We weight three cases 30%, 50% and 20% to $205.86, 13.4% above the price. The good case is $316.73.

The bad case is $122.60, 32% below. With $59 of downside against $24 of upside to the target, conviction is low. A one-point higher required return would make it a Hold.

  • Good, 20%Hyperscaler spending of $1.3tn lifts capacity 40% in FY28 and 25% in FY29, and margins hold near 71%.
  • Middle, 50%Guidance is delivered, capacity growth slows to 10% in FY29 and gross margin fades to 68%.
  • Bad, 30%Permitting curbs and custom chips cut FY29 capacity and memory costs take gross margin to 62%.
Where the value could be, by case
−$59 down+$135.13 upToday $181.60$122.60Bad case$211.47Middle case$316.73Good caseOur target $205.86

Circle size shows how likely we think each case is. Source: Penmark Research.

Scenarios and target (page 22)

What would change our mind

A gross margin below 71% in February would break the thesis

The thesis breaks if gross margin prints below 71% in Q4 FY27, if FY28 capacity additions come in below 20%, or if customers take more than 70 days to pay.

Permitting and custom chips breaking together would make it a Sell.

What bears get right. Permitting and custom chips are the two most likely risks. Together they are worth −$70.3 a share and would take today’s value to $118.52, 35% below the price.
Dates to watch
  1. We model $108.0bn of sales and a 74.0% gross margin, against analysts’ $109.0bn.

  2. A guide well above $115bn would add weight to the custom-chip risk.

  3. We expect a 71.5% margin low point and FY28 sales guided up about 70%.

  4. The first quarter of price increases; we expect gross margin at or above 72%.

Catalysts and risks (page 25)

Inside the full report

What each page covers after the summary on pages 1 and 2. Each opens the full report at that page.

See all 26 pages
  1. 3Returns stay above 50% through FY30E, and the stock trades on 11.4x FY28E consensus earnings against a 17.0x peer median
  2. 4Four dated events and four priced risks: the stock is cheap against its own history and against its peers
  3. 5Data Center revenue is up 2.5x in six quarters; the shares rose 45% to July and are now 12% below that level
  4. 6Our base-case value is $7.25 above the price, and three debates account for all of it
  5. 7Three debates worth $7.25 a share: we differ from the market on FY29 capacity, long-run margin and working capital
  6. 8Of $409bn of FY27E revenue, $266bn reaches operating profit and $191bn becomes free cash flow
  7. 913.8 gigawatts times $27.6bn of content: capacity is the constraint now, and content per gigawatt is what carries growth once it stops
  8. 10A gigawatt of Vera Rubin leaves $29bn of gross profit and $26bn of operating profit: the pricing power sits in the rack, not the chip
  9. 11A $2tn cloud backlog and $1.3tn of 2027 capex against 13.8 gigawatts of supply: the constraint is physical, not financial
  10. 12Our FY27E Data Center revenue is 3.3x Broadcom's guided AI semiconductor revenue, and NVIDIA's operating margin is 11.2pp higher
  11. 13Five years of operating cash flow funded $96bn of buybacks and dividends; the quarterly dividend has just risen 25-fold
  12. 14Results days move the stock 4.5% on average, and earnings have grown far faster than the price: the call is about capacity and margin
  13. 15Capacity is worth $21.64 a share and margin $20.77; the market pays for neither, and pays for 21% FY29 growth we do not forecast
  14. 16Chapter 3. Content per gigawatt carries the growth once capacity stops: $40bn a gigawatt is 60% more than Blackwell
  15. 17Chapter 4. The margin fades slowly: 68% gross margin in FY32E against 52.9% implied by the price
  16. 18Chapter 5. The financing risk is contained: $154bn of commitments against $191bn of FY27E free cash flow
  17. 19Revenue grew 8x from FY22 to FY26 and free cash flow was 80-91% of net income in every year
  18. 20Forecasts. Data Center revenue of $662bn in FY28E and $868bn in FY30E, then content only: the revenue build from the driver up
  19. 21Forecasts. Cost of revenue is 26-30% of sales, so operating margin follows gross margin: 65.5% in FY28E, 57% by FY32E
  20. 22Valuation. A ten-year DCF at a 12.5% discount rate gives $205.86; the 17.0x peer multiple gives $288.86, and we do not use it
  21. 23Valuation. The price implies 2.57% terminal growth and a 54.9% terminal margin, and one point on the discount rate is the difference…
  22. 24Catalysts. Seven dated events in twelve months, and the February 2027 outlook is the one most likely to change the rating
  23. 25Risks. Permitting and custom silicon together are worth −$70 a share and would take the stock to $119
  24. 26Ownership and trading. Short interest is 1.25% of shares and the Street target is 60% above ours: positioning is not the constraint
  25. 27Financial model. Revenue of $909bn by FY30E, a 52% net margin and $572bn of net cash, with no equity issuance
  26. 28Reference. Definitions, the reconciliation from adjusted to reported numbers, the peer set and where every figure comes from

Penmark Research. This summary draws on our initiation report on NVIDIA Corporation, published Nov 6, 2026. It is not investment advice. Analyst certification, rating definitions and important disclosures are in the full report.

Sign in to Penmark