Advanced Micro Devices, Inc. (NASDAQ: AMD), initiation of coverage, Nov 6, 2026

AMD’s AI chip sales can grow faster than today’s price assumes

AMD designs server and PC processors and Instinct AI chips, and on our forecast data centres will be 81% of its sales in 2027. The price assumes AI chip sales grow 148% next year; we expect 170%.

Our rating
Buy
12-month target
$457.00
Price, Nov 5, 2026
$404.40
Expected return, with dividend
+13%

By Penmark Team

AMDBuyTarget $457.00+13% expectedFull report

What AMD is now

AMD has turned into mainly a data-centre chip company

AMD designs the chips and TSMC builds them. Data centres buy its EPYC server processors and Instinct AI chips, now sold in Helios racks; it also sells PC, console and adaptive chips.

Data-centre sales rose for five straight quarters to $8.5bn, up 97% on a year. We forecast server processors at $40.6bn and AI chips at $36.6bn in 2027, 81% of sales together.

2027 sales by business, our forecast
2027 sales: $95.3bn$40.6bn$36.6bn
Server processorsData-centre AI chipsPC processorsEmbedded

The server and AI split is a Penmark Research estimate; AMD does not disclose it. Source: Penmark Research estimates.

Data-centre sales per quarter
$4.3bnQ3’25$5.4bnQ4’25$5.8bnQ1’26$6.7bnQ2’26$8.5bnQ3’26

Data Center segment revenue. Source: company filings.

How AMD makes money (page 7)

What you are really buying

Today’s price assumes AI chip sales grow 148% next year

At $404, the market is paying for 148% growth in data-centre AI sales in 2027, or $92.3bn of total sales. That is already a steep ramp.

AMD has guided to more than double its AI sales and to over 80% growth in total sales. We model 170%, which puts our base value at $425 today, 5% above the price.

Data-centre AI sales growth in 2027, four views
Bad case+90%What the price implies+148%Our forecast+170%Good case+220%

Price-implied growth holds the target P/E and other drivers at our base. Source: Penmark Research estimates.

What the price implies (page 14)

Where we disagree: AI growth

Committed customers and shipping racks put sales above analysts’ forecasts

Anthropic, Meta, Microsoft, OpenAI and Oracle are lead customers for Helios. OpenAI and Meta have each signed a 1GW firm commitment, and we expect 2.8GW to ship in 2027.

We forecast 2027 sales of $95.3bn, 5.0% above analysts’ $90.8bn. Analysts sit below AMD’s own guide of more than 80% growth. This gap is worth $31 a share.

Total sales, ours against analysts
$34.6bn2025$52.1bn$51.5bn2026$95.3bn$90.8bn2027
ReportedOur estimateAnalysts’ forecast

Analysts’ forecasts are consensus per InvestingPro, 5 Nov 2026. Source: company filings, Penmark Research estimates.

Why AI sales more than double (page 14)

Where we disagree: margins

Rack sales will pull gross margin a point below analysts’ forecasts

Helios racks carry more bought-in parts than chips alone. We model a 2027 gross margin of 54.5%, against 55.6% in analysts’ numbers, which costs $13 a share.

Profit still grows faster than sales. Operating costs rise 30% against sales up 83%, so the operating margin climbs from 29.0% in 2026 to 35.6% in 2027.

2027 gross margin, four views
Bad case53.0%Our forecast54.5%Good case55.5%Implied by analysts55.6%

Non-GAAP gross margin; analysts’ margin is implied, not published. Source: Penmark Research estimates.

Margins and operating leverage (page 15)

Where we disagree: dilution

Warrants given to customers will add shares that most forecasts ignore

OpenAI and Meta can each buy up to 160m AMD shares at $0.01, vesting as AMD ships them chips. Most models assume about 1.68bn shares and no dilution.

We expect 2GW to vest in 2027, adding 53m shares, and use 1,700m average shares. That costs $7 a share, but each vested share comes with sales we already forecast.

Average diluted shares, millions
1,637m20241,636m20251,660m20261,700m20271,761m2028
ReportedOur estimate

Forecast shares include customer warrants as they vest. Source: company filings, Penmark Research estimates.

How the warrants work (page 16)

The odds

The upside is nearly twice the downside, so we rate it Buy

We weight three cases 25%, 55% and 20%. Rolled forward a year at our 13% required return, they give a $457 target, 13.1% above the price.

The good case is $172 above the price and the bad case $93 below it. Conviction is low: at the lowest valuation among AI chip peers, the rating would be Sell.

  • Good, 20%Helios ramps faster at the five lead customers: AI sales grow 220% and server processors 110%.
  • Middle, 55%AMD delivers its 2027 plan: AI sales grow 170%, server processors double and warrants vest with shipments.
  • Bad, 25%Policy limits the US build, rack margin falls to 53.0% and Arm processors take server sockets.
Where the price could be in 12 months
−$92.40 down+$171.60 upToday $404.40$312Bad case$480Middle case$576Good caseOur target $457

Circle size shows how likely we think each case is. The target weights all three and adds 12 months of expected return. Source: Penmark Research.

Scenarios and target (page 2)

What would change our mind

Data-centre growth below 45% a quarter in February would end the Buy

The February results and annual report settle three of our four debates. We expect data-centre sales near $12.8bn in the fourth quarter.

Growth below 45% on the quarter, or an AI guide of two times or less, would cut us toward the bad case. A confirmed 56% margin would move us toward the good case.

What bears get right. US data-centre policy is a real risk. Permit pauses or a federal moratorium that slow 2027 deployments would cost $74 a share.
Dates to watch
  1. A read on AI demand and rack supply for the whole sector.

  2. We expect $16.96bn of sales against analysts’ $16.31bn, and data-centre sales near $12.8bn.

  3. First warrant vesting terms and the 2027 guide for AI sales and margin.

  4. Helios volume and gigawatts shipped against 2.8GW for the year.

Catalysts and risks (page 22)

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 25 pages
  1. 3Margins widen to 37% EBIT and ROE to 29% as revenue nearly doubles; we sit 3% to 7% above consensus EPS through FY27E
  2. 4Next catalyst is NVIDIA on 17 November; AMD trades 66% above peers on P/E but 11% below on PEG using our EPS
  3. 5Data Center revenue drives the stock, AI is 53% of FY27E growth, and a 24.4x target multiple gives 13% to our target
  4. 6We differ from the market on four debates; the three we can price net to +$11 a share, against a $20 gap to base value
  5. 7AMD sells chips into four end markets, and Data Center is moving from 48% of revenue in FY25A to 65% in FY26E
  6. 8Revenue is a tree with two branches that matter: server CPU and Data Center AI explain $43bn of FY27E growth
  7. 9The market is growing about 55% a year, but a $51bn FY28E AMD AI line rests on customers who are still raising capex
  8. 10AMD has taken about 10 points of x86 server revenue share between Q4’24 and Q1’26, but its AI position still sits behind a 75% margin leader
  9. 11Management beat its revenue guide in Q1 and Q3 of FY26, and buybacks have covered stock pay; warrants are the new dilution
  10. 12AMD trades on guidance and the chip sector, not on the quarter: EPS beats moved the price the wrong way in 3 of 8 prints
  11. 13The drivers we keep are worth +$31 to −$74 a share; rates, peers and EPS beats do not move the price
  12. 14Data Center AI more than doubles in FY27E because lead customers have committed and Helios is shipping
  13. 15Operating leverage lifts the operating margin to 36% even though rack-scale Helios dilutes gross margin
  14. 16Customer warrants dilute EPS by a visible 3% a year, and every vested share is paid for by shipments
  15. 17Revenue has grown 20% a year since FY21A, but cash conversion swings with working capital and non-GAAP excludes $5bn of real costs
  16. 18Our FY27E revenue is 5% above consensus and 1.5% above the guided floor, and EPS is 7% above on two drivers
  17. 19Value moves most with the target multiple and least with margin; free cash flow lags profit until FY27E as inventory builds
  18. 20Forward P/E on a peer PEG gives a $425 base value and a $457 target; the DCF at a 13% discount rate is the cross-check
  19. 21The DCF gives $223 because it uses a 13.0% discount rate; at the 8.7% the price implies it gives $404
  20. 22The February 2027 results and 10-K decide three of four debates, and we are 4% above consensus on revenue into them
  21. 23Five downside risks cost $7 to $74 a share; the thesis ends if Data Center AI growth falls below the +148% in the price
  22. 24Index funds and large managers own 75% of AMD, shorts are 2.3% of float, and the Street is 83% Buy
  23. 25Revenue rises 3.5 times from $34.6bn to $121.3bn by FY28E as non-GAAP EPS rises from $4.17 to $22.12
  24. 26Quarterly revenue rises from $7.7bn to a modelled $17.0bn in Q4’26 as Data Center grows 3.3 times
  25. 27Reference: one primary method, one cross-check, a 5 November 2026 data cut-off and a stated basis of presentation

Penmark Research. This summary draws on our initiation report on Advanced Micro Devices, Inc., 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