Same Price as August, a Far Bigger 2027: Upgrading AMD to Outperform
Key Takeaways
- The EPS beat was wider than last quarter's, and every Q3 income-statement target on our August checklist was met. Revenue of $13,377M (+45% YoY, +16% QoQ) beat the ~$13.10B Street by 2.1% and cleared the top of management's own $12.7–13.3B range. Non-GAAP EPS of $2.09 beat $1.95 by 7.2%, against a 2.5% EPS beat in Q2. Data Center reached $8,546M, +97% YoY, and 64% of the company. Server revenue in cloud and enterprise more than doubled, well past the "more than 80%" second-half outlook, and Helios shipped for revenue on schedule.
- The fourth quarter and 2027 both moved well beyond our model. Q4 is guided to ~$17.0B ± 3%, about 5% above the ~$16.2B Street and about 13% above the ~$15B our August note implied. Data Center is guided up ~50% sequentially, and data center AI passes server CPU in dollars for the first time. For 2027, management guided total revenue growth above 80%, server revenue roughly doubling, data center AI growing "significantly more than double", and Data Center above 80% of revenue. On our arithmetic that is more than $93.9B of 2027 revenue, against a Street near $90B.
- The costs of the ramp are now visible in three places. Q4 gross margin is guided down to ~55% from 55.9%, a little under the Street. Free cash flow fell to $554M, a 4% margin, as inventory rose $3.0B to $11.5B. The 10-Q shows AMD supporting its customers' build-out with $8.5B of lease and debt guarantees, $13.3B of data center leases it expects to sublease to partners, and $6.2B of cloud compute backstops. The OpenAI and Meta warrants, for up to 320M shares combined, also reached the GAAP income statement as a $1,097M charge.
- The stock is where we downgraded it, and the 2027 estimate is about 24% higher. AMD closed at $488.78 on the reaction session, against $483.36 when we moved to Hold in August. In between it reached $649.42 and then fell 25.5% into the print, partly on AI-financing worries and Intel's weak guide. Our FY27 non-GAAP EPS estimate rises to about $16.50 from about $13.30, which takes the stock to roughly 30x FY27 from 36x on the same year in August.
- Rating: Upgrading to Outperform from Hold. Raising fair value to $495–610 from $400–500. In August we said we would return to Outperform on evidence that the Helios ramp was running materially ahead of the guided cadence. Management said Helios production is "ramping as planned", but the August guided cadence asked only that Q4 Data Center growth exceed Q3's. Q4 Data Center is now guided up ~50% against Q3's 27%, about $12.8B on our arithmetic against the ~$10.2B we implied in August, and Q3 Data Center of $8,546M beat the ~$8.1B we implied. That is the evidence, so we are honoring the trigger. On an unchanged 30–37x multiple, the stock sits 1.3% below the bottom of the new range, with 13% upside to the midpoint. Conviction stays at 6/10 because gross margin, cash conversion and AMD's guarantees and backstops all moved the wrong way in the same quarter.
AMD financial model
Income Statement · Dollars in millions, except per share
| Income Statement | ||||||
|---|---|---|---|---|---|---|
| Actual | Estimate | |||||
| FY2023 | FY2024 | FY2025 | FY26E | FY27E | FY28E | |
| Revenue | $22,680 | $25,785 | $34,639 | $49,004 | $75,346 | $105,160 |
| Less: Cost of sales | ($11,278) | ($12,114) | ($16,456) | ($21,697) | ($32,266) | ($44,693) |
| Less: Amortization of acquisition-related intangibles (COGS) | ($942) | ($946) | ($1,031) | ($1,035) | ($982) | ($900) |
| Gross Profit | $10,460 | $12,725 | $17,152 | $26,271 | $42,098 | $59,567 |
| Less: Research and development | ($5,872) | ($6,456) | ($8,091) | ($11,503) | ($17,706) | ($24,187) |
| Less: Marketing, general and administrative | ($2,318) | ($2,735) | ($4,144) | ($5,709) | ($8,665) | ($11,568) |
| Less: Amortization of acquisition-related intangibles (OpEx) | ($1,869) | ($1,448) | ($1,223) | ($1,151) | ($1,098) | ($1,000) |
| Less: Restructuring charges | $0 | ($186) | $0 | $0 | $0 | $0 |
| Total Operating Expenses | ($10,059) | ($10,825) | ($13,458) | ($18,364) | ($27,469) | ($36,755) |
| Operating Income (Loss) | $401 | $1,900 | $3,694 | $7,908 | $14,629 | $22,813 |
| Less: Interest expense | ($106) | ($92) | ($131) | ($148) | ($148) | ($150) |
| Other income (expense), net | $197 | $181 | $577 | $660 | $660 | $660 |
| Pre-Tax Income (Continuing) | $492 | $1,989 | $4,140 | $8,420 | $15,141 | $23,323 |
| Less: Income tax provision (benefit) | $346 | ($381) | $103 | ($1,197) | ($2,271) | ($3,498) |
| Equity income in investee | $16 | $33 | $26 | $6 | $0 | $0 |
| Net Income from Continuing Operations | $854 | $1,641 | $4,269 | $7,228 | $12,870 | $19,824 |
| Income (Loss) from Discontinued Operations, net of tax | $0 | $0 | $66 | $11 | $0 | $0 |
| Net Income | $854 | $1,641 | $4,335 | $7,239 | $12,870 | $19,824 |
| Basic EPS | $0.53 | $1.01 | $2.67 | $4.42 | $7.79 | $11.98 |
| Diluted EPS | $0.53 | $1.00 | $2.65 | $4.37 | $7.70 | $11.84 |
| Ratios & Assumptions | ||||||
| Gross Margin | 46.1% | 49.4% | 49.5% | 53.6% | 55.9% | 56.6% |
| Operating Margin | 1.8% | 7.4% | 10.7% | 16.1% | 19.4% | 21.7% |
| Net Margin | 3.8% | 6.4% | 12.5% | 14.8% | 17.1% | 18.9% |
| Effective Tax Rate | (70.3%) | 19.2% | (2.5%) | 14.2% | 15.0% | 15.0% |
The full workbook adds 16 historical and 8 projected quarters, plus KPI Drivers · Balance Sheet · Cash Flow Statement — every subtotal a live formula, every projection traced to a driver.
Results vs. Consensus
Q3 2026 Scorecard
| Metric | Q3 2026 Actual | Consensus | Result | Magnitude |
|---|---|---|---|---|
| Revenue | $13,377M | $13.10B | Beat | +2.1% / +$277M |
| Revenue vs. company guide ($13.0B ± $300M) | $13,377M | n/a | Above range | +2.9% vs. midpoint; +0.6% vs. high end |
| Non-GAAP Gross Margin | 55.9% | ~56.1% | Slightly short | About −20 bps; met the ~56% guide |
| Non-GAAP Operating Expenses vs. guide (~$3,650M) | $3,560M | n/a | Lower | −$90M / −2.5% |
| Non-GAAP Operating Income | $3,923M | n/a | Record | 29.3% margin, up from 26.8% |
| Non-GAAP Diluted EPS | $2.09 | $1.95 | Beat | +7.2% / +$0.14 |
| GAAP Diluted EPS | $1.46 | n/a | n/a | After a $1,097M warrant charge and a $555M tax benefit |
| Data Center Revenue | $8,546M (+97% YoY) | ~$8.2B | Beat | About +4% |
| Client Revenue | $2,949M (+7% YoY) | ~$3.05B | Miss | About −3% |
| Free Cash Flow | $554M | n/a | Weak | 4.1% margin vs. 13.5% in Q2 |
| Q4 Revenue Guide (mid) | ~$17.0B ± 3% | ~$16.2B | Beat | +~$0.8B / +4.9% |
| Q4 Non-GAAP GM Guide | ~55% | ~55.3% | Slightly short | About −30 bps; down ~90 bps QoQ |
| Q4 EPS implied by guided components | ~$2.92 ($2.78–$3.07) | ~$2.68 | Beat | About +9% |
| 2027 Revenue Growth Guide | >80% | ~75% | Beat | >$93.9B vs. ~$90B Street |
Year-over-Year Comparison
| Metric | Q3 2026 | Q3 2025 | YoY |
|---|---|---|---|
| Revenue | $13,377M | $9,246M | +45% |
| Non-GAAP Gross Profit | $7,483M | $4,992M | +50% |
| Non-GAAP Gross Margin | 55.9% | 54.0% | +195 bps |
| Non-GAAP OpEx | $3,560M | $2,754M | +29% |
| Non-GAAP Operating Income | $3,923M | $2,238M | +75% |
| Non-GAAP Operating Margin | 29.3% | 24.2% | +5.1 ppts |
| Non-GAAP Net Income | $3,471M | $1,965M | +77% |
| Non-GAAP Diluted EPS | $2.09 | $1.20 | +74% |
| GAAP Operating Income | $2,832M | $1,270M | +123% |
| GAAP Net Income | $2,428M | $1,243M | +95% |
| GAAP Diluted EPS | $1.46 | $0.75 | +95% |
| Operating Cash Flow (continuing ops) | $1,202M | $1,788M | −33% |
| Capital Expenditures | $648M | $258M | +151% |
| Free Cash Flow | $554M | $1,530M | −64% |
Sequential Comparison
| Metric | Q3 2026 | Q2 2026 | QoQ |
|---|---|---|---|
| Revenue | $13,377M | $11,536M | +16% |
| Non-GAAP Gross Margin | 55.9% | 56.2% | −30 bps |
| Non-GAAP OpEx | $3,560M | $3,394M | +5% |
| Non-GAAP Operating Income | $3,923M | $3,094M | +27% |
| Non-GAAP Operating Margin | 29.3% | 26.8% | +2.5 ppts |
| Non-GAAP Diluted EPS | $2.09 | $1.66 | +26% |
| Data Center Revenue | $8,546M | $6,718M | +27% |
| Operating Cash Flow (continuing ops) | $1,202M | $2,366M | −49% |
| Capital Expenditures | $648M | $808M | −20% |
| Free Cash Flow | $554M | $1,558M | −64% |
| Inventories | $11,463M | $8,468M | +35% |
| Accounts Receivable, net | $9,209M | $7,281M | +26% |
| Cash, Equivalents and Short-Term Investments | $15,776M | $13,111M | +20% |
| Total Debt | $7,079M | $3,226M | +119% |
Quality of Beat
Revenue
The revenue beat is organic and concentrated where we want it. Data Center added $4,205M year over year, more than the company's total increase of $4,131M, because Client and Gaming fell $326M while Embedded grew $252M. Inside Data Center, management said cloud and enterprise server sales each more than doubled and data center AI revenue set a record with the first MI450 series shipments. The release and the call gave no dollar split between server and accelerators. Revenue reported here is also net of the customer warrant: the grant-date value of warrant shares probable of vesting is booked as a reduction of revenue as qualifying sales are recorded, and the 10-Q puts the unrecognized portion at $4.5B still to come. The headline is therefore conservative relative to gross shipments, not flattered by them.
The weak spot is client. Client revenue of $2,949M was about 3% below the Street, down 3.7% sequentially, as PC sell-through slowed late in the quarter on memory and component costs. Gaming was flat sequentially at $773M. Neither matters much to the thesis at their current size, but the Q4 guide takes Client and Gaming down more than 20% sequentially.
Margins
Non-GAAP gross margin of 55.9% met the guide and was a touch under the Street, down 30 bps sequentially as MI450 volume began. Operating margin did the work instead: OpEx of $3,560M came in $90M under the ~$3.65B guide and grew 5% sequentially against 16% revenue growth, which lifted the non-GAAP operating margin to 29.3%. The incremental operating margin was 45% sequentially and 41% year over year. That is the operating leverage the CFO promised in August, and it is now the main source of margin expansion, because gross margin is guided down in Q4.
Assessment: The margin structure is changing in the way we expected. Gross margin has stopped rising and will drift lower while data center AI outgrows server, and earnings growth now comes from revenue and OpEx discipline. That is acceptable at 45% revenue growth. It would be much less acceptable if revenue growth slowed.
EPS and Below the Line
Non-GAAP EPS of $2.09 is clean: it uses the 13% normalized tax rate, 1,661M diluted shares, and excludes the warrant remeasurement. GAAP EPS of $1.46 contains two large items that pull in opposite directions. Other income (expense) was a $902M loss, driven by a $1,097M charge to remeasure the warrant liability as AMD's stock price rose and as AMD changed its estimate of the probability that the customers meet the vesting conditions. Against that, the GAAP income tax line was a $555M benefit on $1,868M of pre-tax income, which the 10-Q attributes to FDDEI and R&D credits plus a $772M discrete benefit from stock-based compensation. GAAP net income of $2,428M was therefore higher than GAAP pre-tax income, which is not a pattern to extrapolate. Interest expense rose to $62M from $37M after the $4.75B August notes offering.
Assessment: We model on non-GAAP and keep the warrants in the share count, not in the earnings line. The GAAP line will be volatile for years: management said the quarterly remeasurement is expected to increase as AMD and its customers progress under the agreements, and the 10-Q shows the warrants carried at $178.58 per share against a weighted grant-date value of $14.25.
Cash Flow
Operating cash flow from continuing operations fell to $1,202M (9.0% of revenue) from $2,366M in Q2, and free cash flow fell to $554M (4.1%). Capex was not the cause: it fell to $648M from $808M. Working capital was. Inventories consumed $2,995M, receivables $1,928M and prepaid and other assets $1,070M, partly offset by a $2,313M increase in payables and $323M of accrued liabilities, a net use of $3,357M. The 10-Q attributes the nine-month prepaid increase primarily to prepayments of supply agreements. Inventory reached $11,463M, of which $7,416M is work in process and $2,212M is raw materials, against $909M of raw materials at the December year end.
Assessment: Each line has a growth explanation: inventory for a Q4 that is guided up 27%, receivables for a quarter that grew 16%, and prepayments to lock in wafers, substrates and memory in a shortage. Taken together, though, AMD is now financing its suppliers ahead of the ramp and carrying about 176 days of inventory on cost of sales, against about 152 days in Q2. The CFO said AMD will "continue to invest in inventory and capacity to support our strong growth in 2027". We read that as a guide to more of the same through at least the first half.
Segment Performance
Segment Revenue and Profit — Q3 2026
| Segment | Q3'26 Revenue | YoY | QoQ | Share of Revenue | Operating Income | Operating Margin | Margin a Year Ago |
|---|---|---|---|---|---|---|---|
| Data Center | $8,546M | +97% | +27% | 63.9% | $2,882M | 33.7% | 24.7% |
| Client and Gaming | $3,722M | −8% | −3% | 27.8% | $573M | 15.4% | 21.4% |
| Client | $2,949M | +7% | −4% | 22.0% | (within C&G) | n/a | n/a |
| Gaming | $773M | −40% | −1% | 5.8% | (within C&G) | n/a | n/a |
| Embedded | $1,109M | +29% | +14% | 8.3% | $453M | 40.8% | 33.0% |
| All Other | n/a | n/a | n/a | n/a | ($1,076M) | n/a | ($954M) a year ago |
| Total | $13,377M | +45% | +16% | 100% | $2,832M (GAAP) | 21.2% (GAAP) | 13.7% (GAAP) |
Data Center Growth Markers — What Management Quantified
AMD does not report server CPU and data center AI revenue separately. The table collects every growth marker management gave on the call, so readers can see what is a disclosure and what is our inference.
| Line | Q3 2026 marker | Q4 2026 marker | 2027 marker |
|---|---|---|---|
| Server CPU | Cloud and enterprise sales each "more than doubled" YoY; record enterprise sell-through; Venice revenue shipments began | Growing "really nicely, strongly"; first Venice OEM systems available | Revenue to "double year-over-year"; growth "mostly in units and core count and less so in ASPs" |
| Data center AI | Record revenue; MI455X shipped in volume; first Helios racks online at five customers | "Very significant step up"; "much higher than server business" in dollars | "Significantly more than double"; "grow even faster than we projected 90 days ago" |
| Data Center segment | $8,546M, +97% YoY, +27% QoQ, 64% of revenue | ~+50% QoQ, about $12.8B on our arithmetic | More than 80% of total revenue; full-year 2026 to "more than double" from 2025 |
Data Center: The Segment Is Now the Company
Data Center revenue of $8,546M grew 97% year over year and 27% sequentially, and segment operating income of $2,882M at a 33.7% margin was up 168% year over year. The segment's operating margin expanded 240 bps sequentially even as the first, lower-margin MI450 series shipments arrived, which says the server mix and the operating leverage outweighed the accelerator dilution at the segment level this quarter. Data Center now produces 74% of the three segments' combined operating income.
"Data center revenue growth has accelerated sequentially every quarter this year, with the segment contributing 64% of total revenue in the third quarter, up from 47% just 1 year ago. We expect that acceleration to continue in the fourth quarter, driven by increased demand for inference and agentic workloads and the ramp of our newest products." — Lisa Su, Chair and CEO
Within the segment, the server business delivered the larger upside surprise. In August management guided second-half server growth above 80%; on this call it said cloud and enterprise each more than doubled, that Turin is the most widely deployed EPYC generation in the public cloud with more than 540 instance types, and that AMD won head-node deployments with several of the largest model developers and cloud providers. OpenAI chose Turin as the host CPU for its first custom accelerator platforms.
Assessment: The August debate was whether the 2027 Data Center number was a commitment or a hope. It now looks like a commitment that has been raised. The server line in particular is beating plan with a product (Turin) that is a generation old, and Venice is only starting. The risk in this segment has moved from demand to execution and margin, which is where we would rather it sit.
Client: Share Gains Inside a Shrinking Market
Client revenue grew 7% to $2,949M on Ryzen mobile and a richer mix of high-end processors, but fell 3.7% sequentially and came in about 3% below the Street. Management said PC sell-through slowed late in the quarter as memory and component costs raised system prices, and that AMD is aligning sell-in with sell-through. The commercial franchise keeps widening: eight of the ten largest banks and seven of the ten largest healthcare and automotive companies now use Ryzen PRO.
"Looking at the market, PC sell-through slowed late in the quarter as higher memory and component costs weighed on demand. We are managing the business accordingly, aligning sell-in with sell-through to keep customer and channel inventory healthy. In 2027, we are planning for PC market unit shipments to decline by a double-digit percentage." — Lisa Su, Chair and CEO
Assessment: This is Bear-2 landing as forecast, and it lands harder than we had in August. A double-digit unit decline in 2027 is a real headwind for a business that is still about 22% of revenue. AMD's exposure is skewed to enterprise, premium consumer and desktop channel, and management noted that its supply is fungible between PC and data center, so lost client wafers can move to server. We model Client and Gaming down about 11% in 2027.
Gaming: Late-Cycle and Small
Gaming revenue fell 40% year over year to $773M on lower semi-custom sales at this stage of the console cycle, and was essentially flat sequentially. Radeon GPU channel sell-through rose sequentially, and Radeon AI Pro revenue "grew significantly" from Q2, which management attributes to developers running and fine-tuning models locally. Client and Gaming segment operating margin was 15.4%, against 21.4% a year ago.
Assessment: At 5.8% of revenue, gaming no longer moves the equity. The segment margin is the part to watch: it has held near 15% for two quarters while revenue fell, and the Q4 guide takes segment revenue down more than 20% sequentially, which will test that floor.
Embedded: Fourth Consecutive Acceleration
Embedded revenue rose 29% year over year and 14% sequentially to $1,109M, the fourth consecutive quarter of accelerating growth, with operating margin at 40.8%. Demand was broad across aerospace and defense, automotive, communications and industrial, and data center build-outs are lifting demand for embedded CPUs and adaptive SoCs in control-plane and optical networking. Design wins remain on track for more than $18B in 2026, a third consecutive record year.
"As these wins ramp over the next few years, we expect our embedded business to deliver strong double-digit annual revenue growth and continue outgrowing the market." — Lisa Su, Chair and CEO
Assessment: Embedded is the highest-margin segment and the one offset to accelerator dilution that is growing. Management expects embedded growth to accelerate in 2027. At a 41% segment margin, every dollar here is worth more to earnings than a dollar of Helios, which is why the CFO named it as a gross-margin help.
Key KPIs
| KPI | Q3 2026 | Q2 2026 | Q3 2025 | Trend |
|---|---|---|---|---|
| Data Center share of revenue | 63.9% | 58.2% | 47.0% | Rising fast |
| Non-GAAP gross margin | 55.9% | 56.2% | 54.0% | Peaked; Q4 guided ~55% |
| Non-GAAP operating margin | 29.3% | 26.8% | 24.2% | Expanding |
| Non-GAAP OpEx / revenue | 26.6% | 29.4% | 29.8% | Leverage |
| Free cash flow margin | 4.1% | 13.5% | 16.5% | Falling |
| Inventory days (on cost of sales) | ~176 | ~152 | n/a | Rising |
| Receivable days | ~63 | ~57 | n/a | Rising |
| Net cash (cash and short-term investments less debt) | $8,697M | $9,885M | $4,023M | Down $1.2B QoQ |
Key Topics & Management Commentary
Overall Management Tone: The most specific call AMD has given in the past year. Management put numbers on 2027 that it had declined to give in August and tied them to secured supply, and it answered the financing and custom-silicon questions directly rather than by deflection. The soft seams were the same as last quarter, in gross margin and cash, plus a new one in warrant dilution: management said it cannot frame 2027 gross margin quarter by quarter and asked investors to wait for share-count guidance.
1. 2027 Revenue Guided Above 80%, Twice the Analyst Day Model
The single most important statement on the call was a company-level 2027 growth number. In August, management gave Data Center markers only. It now expects total revenue to grow more than 80% in 2027, server revenue to double, data center AI to grow "significantly more than double", and Data Center to exceed 80% of revenue. On the FY26 base implied by the Q4 midpoint ($52,166M), that is more than $93.9B of 2027 revenue, against a Street near $90B (+~75%) entering the print and the $78–85B we published in August.
"Based on our current visibility, we now expect revenue to grow more than 80% in 2027, which is more than twice the long-term financial model we outlined at our Financial Analyst Day last November." — Lisa Su, Chair and CEO
Management tied the number to supply it has already secured, saying AMD has "worked across our supply chain to secure the wafers, substrates, advanced packaging, memory and other key components needed to support much higher growth next year", and said demand still exceeds supply.
Assessment: This is the revision that changes our estimates. It is a floor ("more than"), it is backed by named supply rather than by a TAM, and it arrives after a quarter in which the server line beat its own outlook. The caveat from August still applies: a number this large, given this early, will be marked to market several times before it is tested. But AMD's 2026 record so far is that each forward marker was beaten, not cut.
2. Q4: Data Center Up ~50% Sequentially, and Data Center AI Overtakes Server
The Q4 guide of ~$17.0B ± 3% implies 66% year-over-year and 27% sequential growth. The CFO broke the sequential step into a ~50% increase in Data Center, a decline of more than 20% in Client and Gaming, and double-digit Embedded growth. Solving those descriptions puts Q4 Data Center near $12.8B, Client and Gaming below $3.0B and Embedded above $1.2B. Management also said data center AI revenue passes server revenue in Q4, and in Q&A described the gap as "much higher" in dollars.
"Sequentially, we expect revenue to grow by approximately 27%, driven by a sharp increase in our data center segment of approximately 50% growth, a decline in our client and the gaming segment of greater than 20% and a double-digit growth in our Embedded segment." — Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: The August guided cadence was that Q4 Data Center growth would exceed Q3's, and from it we implied about $8.1B of Q3 Data Center and $10.2B of Q4. Q3 printed $8,546M, and on our arithmetic the ~50% Q4 guide is about 25% above our $10.2B. Management described Helios production as "ramping as planned". We read that as a statement against its internal plan, not against the August guide. This is the evidence our published upgrade trigger asked for: the Helios ramp is running materially ahead of the guided cadence. It also means Q4 carries more of the year than any quarter before it, which raises the cost of any slip in rack integration.
3. Helios Ships for Revenue; Five Customers Live in the First Quarter
Q3 was the first quarter of MI455X and Helios revenue. Management said lead customers Anthropic, Meta, Microsoft, OpenAI and Oracle brought their first racks online quickly and are running their latest models and workloads on them. HPE, Dell, Lenovo, Super Micro and other OEM and ODM partners begin shipping Helios systems in Q4. The CFO said data center AI was "actually better than our expectation when we got into the Q3 quarter".
"We started shipping MI455X in volume in the quarter, and lead customers, including Anthropic, Meta, Microsoft, OpenAI and Oracle, brought their first Helios racks online quickly. Customers are now running their latest models and workloads on Helios, and we are seeing very strong performance with the initial deployments." — Lisa Su, Chair and CEO
Looking further out, management said multiple frontier labs are working with AMD to optimize MI500 as they plan multi-gigawatt deployments, and that MI600 co-design is under way. EuroHPC selected MI430X for LUMI-AI, which comes online in the second half of 2027. ROCm 10 shipped with production-scale resilience features and the ROCm.AI agentic tuning tool.
Assessment: A rack-scale product reaching five customers in its first revenue quarter, with the OEM channel following one quarter later, is a better start than the August note assumed. The open questions are yields and margin on an early, complex product, which management did not update this quarter, and how much of the ramp depends on third-party rack components, which the 10-Q names as a risk factor.
4. Server CPU: Doubling in 2027 Off a Doubling Base
Server was the upside surprise. Management said it "significantly exceeded" its outlook for more than 80% growth, with cloud and enterprise sales each more than doubling, record enterprise sell-through and dozens of first-time EPYC customers. Venice began revenue shipments in Q3, OEM and ODM partners have committed more than 240 Venice platforms (20% more than Turin), and cloud providers are deploying Venice in Q4 ahead of instance launches in early 2027. The 2027 server marker moved from "more than 70%" in August to roughly 100%.
"Venice builds on that leadership with significantly better performance per watt, performance per core and performance per dollar than any other x86 or ARM server CPU, positioning us to gain share and significantly grow our server business as the TAM grows to approximately $220 billion by 2030." — Lisa Su, Chair and CEO
Assessment: Bull-1 is the strongest pillar in the thesis and got stronger. The composition matters: management said the 2027 growth is mostly units and core count, with only a modest ASP uplift, so the doubling does not depend on pricing power. The 2030 TAM stayed at about $220B this quarter after the three moves in nine months we flagged in August, which helps the credibility caveat we attached to Bull-4.
5. Agents Become a CPU Story: Helion and the All-CPU Rack
Management argued that personal and enterprise agents are becoming "the next killer app for CPUs in the cloud", citing more than 10 million people activating Meta's Muse agents in the first six weeks. AMD launched Helion, a Venice reference design for rack-scale agentic systems that can be configured as an all-CPU rack (up to 144 Venice processors and nearly 74,000 threads) or a mixed CPU and Instinct rack. Every major server OEM and ODM plans to offer it, with the first systems shipping in Q4.
"These agents work continuously and in parallel across applications and enterprise systems, adding to accelerator demand while creating a substantial increase in CPU demand. Adoption is just starting, and we expect demand to grow sharply as agents are deployed more broadly and take on more complex work." — Lisa Su, Chair and CEO
Assessment: Helion turns the agentic-CPU thesis (Bull-4) into a product with a part number, which is what it lacked. A CPU-dense rack is also better for AMD's margin mix than a GPU rack. We still do not capitalize the 2030 TAM, but the 2027 server doubling is the near-term proof that the agentic CPU demand is real.
6. Gross Margin Steps Down to ~55%
Q4 non-GAAP gross margin is guided to ~55%, down about 90 bps from Q3's 55.9% and a little under the ~55.3% Street, as data center AI passes server. The CFO called the accelerator business "a little bit dilutive to corporate average" and pointed to gross profit dollars rising sharply. Asked twice for a 2027 framework, she said the server and data center AI mix will set the margin, that quarterly variation is likely, and that Q4 is "a good benchmark".
"So we all know their margin is a little bit dilutive to corporate average, but the gross margin dollars are increasing very significantly." — Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: Bear-1 moves from EMERGING to MATERIALIZING. The size is modest (90 bps) and fully explained by mix, and management offered embedded and server doubling as offsets in 2027. We model 55% for 2027, which is the CFO's benchmark, and note that our estimate has no room for a second leg down if HBM and component inflation are not shared with customers as management says they will be.
7. The Customer Warrants Arrive in the P&L
Q3 was the first quarter in which AMD recognized the accounting impact of the customer warrants as MI450 series GPUs shipped. The 10-Q identifies the holders as OpenAI and Meta, each entitled to buy up to 160 million shares at $0.01, for a combined 320 million shares, or about 19% of Q3's diluted count. Tranches vest as the customers reach purchase milestones and AMD's stock meets price targets, and become exercisable only after technical and commercial conditions are met. As of September 26, no warrant shares had vested, the liability stood at $1.2B and $4.5B of grant-date value remained to be booked as a reduction of revenue.
"Going forward, these quarterly noncash fair value adjustments are expected to increase as we and our customers progress under this agreement. The amount will vary each quarter based on AMD's stock price and other valuation assumptions. These noncash adjustments would not affect our non-GAAP results or free cash flow." — Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: The P&L charge does not matter to us. The share count does. Q4 is guided without warrant dilution, and management will guide 2027 share count only once vesting conditions are met. We include about 60 million weighted warrant shares in our FY27 share count. That is our assumption for partial vesting as roughly 3 GW of planned deployments land during 2027, not a company figure. Full vesting of both warrants would reduce our FY27 EPS by about 13%; it would also require far more deployment than our FY27 revenue assumes.
8. Cash Conversion: Free Cash Flow Margin Falls to 4%
Free cash flow of $554M was a 4.1% margin, down from 13.5% in Q2 and 16.5% a year ago. The decline came from working capital, not capex: inventory rose $2,995M, receivables $1,928M and prepaid and other assets $1,070M. Inventory now stands at $11.5B, and other non-current assets rose to $10,020M from $5,452M at the year end. The CFO framed the build as fulfilling data center demand and guided more of it.
"Free cash flow was $554 million, reflecting increased working capital investment and ongoing capital expenditures to secure equipment and capacity for the rapid growth of our data center business, including back-end capacity dedicated to support our data center demand. We expect to continue to invest in inventory and capacity to support our strong growth in 2027." — Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: Bear-5 moves from EMERGING to MATERIALIZING. In August we expected free cash flow conversion "in the teens for the next two to three quarters"; the first of those quarters printed 4%. The working-capital build is a down payment on a Q4 guided up 27%, so Q4 cash flow should recover some of it as receivables collect. But a business that holds 176 days of inventory and prepays its suppliers is a different cash machine from the one the market valued two years ago, and our multiple already reflects that.
9. Building the Customer's Data Center: Guarantees, Leases and Compute Backstops
The 10-Q filed alongside the results describes arrangements that the call referred to only in general terms. AMD guarantees certain commercial partners' data center lease and debt obligations, with a maximum potential exposure of $8.5B over terms of up to 15 years. It has signed $13.3B of data center leases that it expects to assign or sublease to commercial partners, and the leases are expected to commence beginning in Q2 2027. It has committed up to $6.2B to buy GPU compute capacity from cloud providers to the extent third-party customers do not. It also has up to $5.1B of investment commitments, and invested $1.0B in a public company's equity offering in October. No partner was in default and nothing had been paid under the guarantees.
"And then we are selectively investing with our lead partners where we see it's helpful to have, for example, the data center infrastructure identified way in advance, and that's multiple years in advance." — Lisa Su, Chair and CEO
Assessment: This is the most important new disclosure of the quarter and the reason conviction does not rise with the rating. The three support categories add up to $28.0B of gross exposure, about $17 per share if every dollar were lost with no recovery, which we do not expect. The issue is less the size than the signal: AMD is now using its balance sheet to make sure its customers' data centers exist. Financing of AI data centers was the worry that weighed on the group in October, and AMD now carries some of that risk directly. We open a new Bear-6 at EMERGING to track it.
10. Supply: Demand Above Supply, and the Bill to Secure It
Management said repeatedly that demand exceeds supply in both server and data center AI, and that AMD is working with partners to add more. The 10-Q puts unconditional operating commitments, mainly wafers, substrates, memory, components and assembly and test, at $48.8B, of which $33.4B falls in the rest of 2026 and in 2027. The CFO cited "back-end capacity dedicated to support our data center demand" among the capex uses. Management also said its supply is fungible between PC and data center, which lets it move capacity as the PC market weakens.
"We have worked across our supply chain to secure the wafers, substrates, advanced packaging, memory and other key components needed to support much higher growth next year." — Lisa Su, Chair and CEO
Assessment: The supply language is the reason we take the 2027 number seriously. A company that has committed $33.4B for the next five quarters is not guessing at demand. The other side is that those commitments are fixed while demand is not, which is the classic semiconductor risk if the AI build-out pauses.
11. Taalas and World Labs: $11.1B for Inference Silicon and Model Research
AMD agreed to buy Taalas, a developer of low-latency inference silicon, and World Labs, a spatial-intelligence model lab led by Dr. Fei-Fei Li, who will join AMD as Chief Scientist. The call gave no prices. The 10-Q values Taalas at about $2.9B in cash and stock, including up to about $1.4B of contingent consideration, and World Labs at about $8.2B, all in stock. Both are expected to close by year end, and the Q4 diluted share estimate of 1,668M includes the expected issuance.
"Fei-Fei will join AMD as Chief Scientist, leading our model research and helping shape our future hardware, software and systems road maps. We expect both acquisitions to close by the end of the year." — Lisa Su, Chair and CEO
Assessment: The World Labs deal is a large price for a research team, paid in stock at a high share price, and it moves AMD further toward building models of its own. We treat both deals as R&D spending that will show up in OpEx and dilution rather than revenue. The CFO's 2027 OpEx commitment (growth below half of revenue growth) should be read as including them.
Guidance & Outlook
| Metric | Q3'26 Actual | Q4'26 Guide | Implied Change | vs. Street |
|---|---|---|---|---|
| Revenue | $13,377M | ~$17.0B ± 3% ($16.49–17.51B) | +27% QoQ, +66% YoY | +4.9% vs. ~$16.2B |
| Data Center | $8,546M | ~+50% QoQ | About $12.8B (our arithmetic) | n/a |
| Client and Gaming | $3,722M | Down >20% QoQ | Below $3.0B | n/a |
| Embedded | $1,109M | Up double digits QoQ | Above $1.2B | n/a |
| Non-GAAP Gross Margin | 55.9% | ~55% | About −90 bps | −30 bps vs. ~55.3% |
| Non-GAAP OpEx | $3,560M | ~$3,750M | +5% QoQ | n/a |
| Non-GAAP Other Income, net of Interest | n/a | ~+$4M | n/a | n/a |
| Non-GAAP Tax Rate | 13% | 13% | Unchanged | n/a |
| Diluted Share Count | 1,661M | ~1,668M | Includes acquisition shares; no warrants | n/a |
| Implied Non-GAAP EPS | $2.09 | ~$2.92 ($2.78–$3.07) | +40% QoQ | About +9% vs. ~$2.68 |
2027 Framework
| Line | August Marker | November Marker | Change |
|---|---|---|---|
| Total revenue | Not given | Growth >80% (>$93.9B on our FY26 base) | New, above Street |
| Server CPU revenue | >70% | About double | Raised |
| Data center AI revenue | "Well over 100%" for the segment | "Significantly more than double"; faster than projected 90 days ago | Raised |
| Data Center share of revenue | Not given | More than 80% | New |
| Non-GAAP OpEx | Grows slower than revenue | Grows less than half the revenue growth rate | Tightened |
| Gross margin | Not given | Mix-driven; Q4 (~55%) is "a good benchmark" | Partial |
| Share count | Not given | Guidance deferred until warrant conditions are met | Deferred |
| PC market units | Softer second half | Down double digits in 2027 | Lowered |
The guide is a raise on every axis except gross margin. At the midpoint, revenue grows 27% sequentially, OpEx grows 5%, and the implied non-GAAP EPS of about $2.92 is 40% above Q3. FY26 revenue lands near $52.2B (+51%), with Data Center near $33.9B, more than double FY25's $16.6B, as management said.
Implied ramp into 2027: To grow 80% on $52.2B, AMD needs about $94B of 2027 revenue, or an average of about $23.5B a quarter. Q4's ~$17B exit rate annualizes to $68B. The 2027 framework therefore needs roughly 38% growth from the Q4 exit rate across 2027, at a time when Client is guided to shrink. Nearly all of it has to come from Data Center, consistent with the "more than 80%" mix marker.
Street at: ~$16.2B for Q4 and ~$90B for 2027 before the print. Post-print targets moved in both directions, but the published 2027 revenue view from at least one desk is already near $95B. We expect the Street to converge on the $94–96B range.
Guidance style: AMD has beaten the midpoint of its own revenue guide by about 3% in each of the last two quarters ($11,536M against $11.2B; $13,377M against $13.0B). Applied to Q4, that pattern points to about $17.5B. We use the $17.0B midpoint in our estimates and treat the pattern as upside, because Q4 depends on OEM Helios systems that have not shipped yet.
Analyst Q&A Highlights
Whether the 2027 Plan Is Backed by Supply, and Whether Demand Exceeds It
The opening question asked whether the 2027 framework is supported by wafers, packaging, substrates and memory already in hand, and whether demand would outrun supply even so. Management answered both halves: the supply has been planned with customers and suppliers over several years, and demand is still higher than supply. A later question asked whether the server doubling meets all demand; the answer was that it does not.
Q: "So maybe can you talk about, first and foremost, like is this backed up by supply, whether that be wafers, packaging, substrate, memory? And then if you look at the demand profile behind that, would demand outstrip that supply if you were able to get some more?"
— Thomas O'Malley, Barclays
A: "It is true that demand is higher than supply. So we continue to work with our supply chain partners to increase supply. And I do believe that there is a good opportunity for us to do that, just given the strong relationships we have throughout the ecosystem."
— Lisa Su, Chair and CEO
Assessment: A clean answer. Supply-gated growth with excess demand behind it is the best kind of guide to underwrite, because upside comes from supply additions rather than from demand forecasts. It also means the 2027 number is a supply number, which caps how far it can rise in-year.
Whether Anchor Customers Can Finance and House the 2027 Deployments
The second exchange went straight to the October worry: do the anchor customers have the financing, land, power and shells for roughly 3 GW of planned AMD deployments in 2027, and are competitors' residual-value guarantees changing the competitive terms? Management said it knows which data centers and locations the GPUs are going to, and that AMD is "selectively investing" with lead partners to identify infrastructure years in advance. The answer did not size that investment. The 10-Q, filed that evening, does.
Q: "So kind of a 2-parter, what's the level of confidence, right, that AMD has that your anchor customers have the financing and the physical infrastructure to support, I think it's roughly 3 gigawatts or so of planned AMD deployments into '27? And then, as we look beyond, many of your competitors are kind of investing in the ecosystem, providing residual value guarantees, and such."
— Vivek Arya, Bank of America Securities
A: "So we know where the GPUs are going in terms of which data centers and which locations and all of that stuff. ... And then we are selectively investing with our lead partners where we see it's helpful to have, for example, the data center infrastructure identified way in advance, and that's multiple years in advance."
— Lisa Su, Chair and CEO
Assessment: The site-level visibility is reassuring for 2027. "Selectively investing" is a soft description of $8.5B of guarantees, $13.3B of leases to sublease and $6.2B of compute backstops. We would have preferred management to put the figures on the call next to the description.
Whether Higher HBM and Component Costs Can Be Passed Through
A question on memory economics asked how AMD has kept its revenue-per-gigawatt framing steady while HBM prices rise, and whether higher input costs can be passed to the largest customers. Management said early programs ran on long-term agreements with suppliers and customers, and that AMD is now working with customers so that cost increases are shared. It framed the competitive test as tokens per dollar rather than as AMD's own margin.
Q: "Is that because early builds are fairly program-specific and pricing was agreed upon a long time ago? And again, as we think about that business longer term, do you have the ability to pass on these higher input costs even to your big 3 customers?"
— Joshua Buchalter, TD Cowen
A: "We are also working with our customers to include -- to ensure that those costs are shared as we go forward. ... this is all about TCO and ensuring that from a tokens per dollar standpoint, that we continue to offer very strong advantages for choosing AMD Instinct in our overall Helios system."
— Lisa Su, Chair and CEO
Assessment: "Shared" is not "passed through". The answer implies AMD absorbs part of the memory inflation, which is consistent with the Q4 gross margin step down. It is the most direct link between Bear-1 and the HBM market that management has given.
How Fast OpEx Can Grow Against 80% Revenue Growth
With the 2027 revenue framework at more than twice the long-term model, a caller asked how OpEx should grow and whether a 55% gross margin floor holds through the Helios ramp. The OpEx half got the most precise answer on the call: growth below half of the revenue growth rate. The gross margin half got a mix explanation and no floor.
Q: "How should we think about OpEx growth in '27? And I guess while I'm in the neighborhood, do you think you can hold your target gross margin range sort of that 55% floor as we move through the year and Helios continues to ramp?"
— Joshua Buchalter, TD Cowen
A: "That dynamic, we believe, will carry into 2027, Lisa talked about is that we expect the whole company growing more than 80%. We should expect OpEx to grow less than the top line revenue growth rate -- less half -- less than half of the top line revenue growth rate. So that will drive a very significant operating margin expansion in 2027."
— Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: The OpEx commitment is quantified and testable, and it is the reason our FY27 operating margin reaches about 35%. The non-answer on a 55% floor makes this the second consecutive call without a 2027 gross margin framework, and it is why our margin estimate sits at the CFO's Q4 benchmark rather than above it.
Whether There Is a 2027 Gross Margin Entrance and Exit Rate
A second attempt asked for a 2027 gross margin entrance and exit rate. The CFO said the company-level margin depends on the mix between server and data center AI, both supply-constrained, and that she does not have visibility to give a quarterly path. She offered Q4, when data center AI is much larger than server, as the benchmark.
Q: "Do you think you could give us some color or some highlights on like a good entrance rate and a good exit rate for gross margins in 2027, given that the ramps of the different products and everything that are happening?"
— Stacy Rasgon, Bernstein Research
A: "So the mix between those 2 businesses actually determine what the company's gross margin will be. Right now, I actually don't have that visibility. ... Q4 actually, is a good benchmark to think about it is because Q4 data center AI business is much higher than CPU business."
— Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: An honest answer, and a useful one if taken literally. If Q4's mix, with data center AI well ahead of server, produces about 55%, then a 2027 in which server doubles and embedded accelerates should not average much below it. The risk is a mix tilted further toward accelerators than Q4.
How the Warrants Will Layer Into the Model
A caller asked for guardrails on how the customer warrants enter the model through 2027. The CFO split the answer into three parts: revenue is always reported net of the warrant's grant-date value; the shares enter the diluted count only when all vesting and exercise conditions are met; and the quarterly remeasurement is a non-cash item excluded from non-GAAP. On timing of the share count, she asked investors to wait.
Q: "How are they going to layer in through the year? Any sort of like guardrails you can give us on sort of how to roll that into our model through the year and into 2027?"
— Timothy Arcuri, UBS
A: "The way to think about this, the share count only comes into diluted shares once all the vesting conditions and the exercising conditions are met by our customers. That means the shipments need to be there and some commercial conditions that need to be met. ... So on the share count, just wait for us to provide you a clear guidance when we get there."
— Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: The mechanics are clear and the timing is not. Up to 320 million shares can enter the count, conditioned on shipments, and the company will not say when. We carry a partial-vesting assumption in FY27 and show the full-vesting case in our valuation.
How Large the Q4 Crossover of Accelerators Over Server Will Be
A caller asked whether data center AI's Q4 crossover over server is marginal or large, to size the entry rate into 2027. The CFO said the MI450 step-up is "very significant" and that data center AI will be "much higher" than server in Q4, and confirmed on follow-up that she meant dollars.
Q: "I was wondering if you could help us with the magnitude of that crossover? Is it just barely there? Or like is it like significantly higher?"
— Stacy Rasgon, Bernstein Research
A: "I think the way server business, of course, is very strong, but the data center AI will be much higher than server business in Q4."
— Jean Hu, Executive Vice President, CFO and Treasurer
Assessment: The clearest sizing of the accelerator ramp AMD has offered. With Q4 Data Center near $12.8B and server growing strongly, "much higher" implies a data center AI quarter well above $6.4B, which is a multi-billion sequential step from Q3. It confirms the ramp is ahead of the guided cadence from August.
Whether the Server Doubling Is Units or Price
A caller asked how much of the roughly 100% 2027 server growth comes from units versus higher core counts and Venice pricing. Management said most of it is units and core count, with a modest ASP uplift that follows core count.
Q: "I'm curious how we should be thinking about units versus mix, including higher core count as well as higher ASPs associated with Venice. Is there a way to break that down to help us understand that 100% growth?"
— Christopher Muse, Cantor Fitzgerald
A: "So on the server CPU side, when we talk about 100% growth, you should imagine that we are definitely shipping significantly more units in that. Of course, some of those units have higher core counts as we go into Venice. But I would say the growth in revenue is mostly in units and core count and less so in ASPs, although there are some -- there's an ASP uplift as we increase core count."
— Lisa Su, Chair and CEO
Assessment: Unit-led growth is the healthier composition and the harder one for a competitor to undercut on price. It also means the server doubling consumes substrate, packaging and DDR5 supply in proportion, which is why management tied it to secured components.
What They're NOT Saying
- The size of AMD's support for its customers' data centers, on the call. The 10-Q discloses $8.5B of maximum guarantee exposure, $13.3B of data center leases to be assigned or subleased to partners, $6.2B of cloud compute backstops and up to $5.1B of investment commitments. On the call they were described only in general terms, most directly as "selectively investing with our lead partners". For a market that sold the group in October on AI-financing worries, the figures belonged in the prepared remarks.
- No dollar figure for data center AI or server revenue. Management gave "record", "more than doubled", "much higher than server" and "significantly more than double" for 2027, but no dollars for Q3, Q4, 2026 or 2027. The Street is backing into the split from segment guidance, as it did in August.
- No 2027 gross margin range, for the second consecutive quarter. Asked twice, the CFO offered Q4's ~55% as a benchmark and declined a floor or an entrance and exit rate. Management will size 2027 revenue to "more than 80%" and will not bracket margin.
- No warrant vesting schedule or 2027 share-count range. Up to 320 million shares can enter the diluted count. Management said only that guidance will come when the conditions are met.
- No capex guide, again. The CFO cited "ongoing capital expenditures to secure equipment and capacity", including dedicated back-end capacity. No run rate was given. Capex fell to $648M from $808M, so the August concern about an unguided step-up was not borne out this quarter, but the number is still unguided.
- No buybacks for a second consecutive quarter. The release shows no repurchases under the program in Q3, against $1,495M of shares withheld for employee taxes and $4.7B of new debt. The 10-Q shows $9.2B of authorization remaining. Management offered no capital-return commentary.
- No update on Helios yields or early-ramp margin. In August management said yields would improve over the first few quarters of the ramp. This call, the first quarter of revenue, gave no yield update.
- No China commentary. China was not mentioned on the call. The 10-Q says the U.S. government granted licenses in the first half of 2026 to ship MI325 products to certain China-based customers, subject to a U.S. inspection that carries a 25% tariff. Neither the volume nor any revenue was discussed.
- The acquisition prices were left to the filings. The call described Taalas and World Labs strategically. The 10-Q gives the values: about $2.9B for Taalas, including up to $1.4B contingent, and about $8.2B in stock for World Labs.
- The $555M GAAP tax benefit was explained only in the 10-Q. The release shows it on the income statement, and the call did not mention it. The 10-Q attributes it to FDDEI and R&D credits plus a $772M discrete benefit from stock-based compensation, which fits the $1,495M of shares withheld for employee taxes at a high stock price. It does not affect non-GAAP EPS, but it lifted GAAP net income above GAAP pre-tax income in the same quarter the warrant charge depressed it.
Market Reaction
- Pre-print setup: Closed at $483.90 on November 3, up 126.0% year to date against the S&P 500's 10.6%, and up 86.4% over twelve months. The stock was down 23.7% over the trailing thirty days and 25.5% below its $649.42 52-week closing high of October 6. In the two sessions before the print it fell 17.6%, from $566.42 on October 29 to $466.84 on November 2, including a 10.9% drop on October 30 after Intel's weak outlook, then rose 3.65% on the day of the release.
- After-hours move: Up roughly 6% to 7.5% after the release and through the call, with the high near $519–520.
- Reaction session: Opened at $485.25, a 0.3% gap up, after trading 4% to 5% higher before the open. Traded $476.50 to $514.61 and closed at $488.78, up 1.0% or $4.88.
- Volume: 50.3M shares, 2.4x the 30-day average of 20.7M.
- Peers and indices: The S&P 500 rose 1.6% and the Nasdaq-100 tracker 1.0%, but semiconductors fell: the SMH ETF was down 2.7%, NVIDIA 3.2%, Intel 3.8%, Broadcom 2.0%, TSMC 2.2%, Arm 5.3% and Micron 7.0%. AMD outperformed SMH by 3.7 points.
- Since our August note: $483.36 on August 7 to $488.78, up 1.1%, against a 0.9% decline in the S&P 500. The lowest close in between was $456.16 on September 3.
An After-Hours Gain That Faded With the Sector
The after-hours move was a straightforward read of the guide: a Q4 about 5% above the Street and a 2027 framework above the ~75% the Street had assumed. Most of that gain was gone by the close, and the obvious reading is a fifth beat-and-fade. We think that reading is incomplete. The fade came on a day when semiconductors fell sharply and the broad market rose, and AMD closed up while every large peer we track closed down. Relative to the sector, the stock kept about four points of the print.
The Setup Was a Washout, Not a Run-Up
This is the opposite of August. Then, the stock had rallied 20.7% in four sessions into the print and needed a blowout. This time it had fallen 25.5% from its October 6 high. The decline ran through a CoreWeave bond sell-off in mid-October and a broad tech sell-off on October 22, and ended with a 17.6% drop in the two sessions after Intel's guide-down (October 30 and November 2). The bar was low, as post-print sell-side commentary acknowledged. A beat-and-raise against a low bar that holds only a 1% gain says investors are not yet willing to pay for 2027 while the financing question is open. That is a positioning problem, not a fundamental one.
What the Market Is Discounting
At $488.78 the stock trades at roughly 30x our new FY27 estimate, against 36x on the same year at our August downgrade. The market has absorbed a 24% increase in our 2027 earnings estimate without moving the price. Some of that is fair: gross margin, cash conversion and the new customer-support exposures all argue for a discount. But we think the market is treating a 2027 that management now backs with secured supply as if it were as uncertain as it was in August, and that gap is the basis for the upgrade.
Street Perspective
Debate: Is 80%-Plus 2027 Growth a Real Number?
Bull view: Every forward marker AMD gave in 2026 has been met or beaten, the server line just beat its own outlook by a wide margin, and the 2027 number is backed by secured wafers, substrates, packaging and memory rather than by a TAM. Demand is above supply, so the risk is to the upside if supply can be added. The Street was at about 75% and will move up.
Bear view: The number depends on a handful of frontier labs and hyperscalers deploying about 3 GW of AMD capacity on schedule, at a time when AI-financing worries have already hit the group and AMD itself is guaranteeing partners' leases. PC units are planned down double digits, and the Q4 exit rate needs roughly 38% further growth across 2027 to get there.
Our take: The bull view has the better evidence for 2027. Site-level visibility, five customers already live on Helios, and $33.4B of supply commitments through 2027 are not the profile of a guess. We underwrite about $95B, a little above the floor, and treat the customer concentration as a risk to 2028 rather than to 2027.
Debate: Is AMD Financing Its Own Demand?
Bull view: The support is small relative to the business. The $8.5B of guarantees, $13.3B of leases to sublease and $6.2B of compute backstops add up to about $28B, against a 2027 revenue run rate near $95B and $15.8B of cash. No partner is in default, the guarantees' fair value is immaterial, and the leases are expected to pass to partners. Competitors are doing versions of the same thing to secure deployments, as one question on the call noted.
Bear view: Combine the guarantees, the backstops, $5.1B of investment commitments and warrants for up to 320 million shares given to two customers, and AMD is paying for a meaningful part of its anchor demand with equity and balance sheet. Free cash flow was 4% of revenue in the quarter. If one partner's financing fails, AMD inherits a data center lease and loses the revenue in the same move.
Our take: The bear view is right about the direction and wrong about the scale, for now. At about $17 per share in a total-loss case, the exposure does not change the rating. It does change the quality of the revenue, and it is why we open a new bear pillar and hold conviction at 6 rather than raising it. The test is whether the commitments grow faster than revenue over the next two quarters.
Debate: Does Gross Margin Hold Near 55% as Accelerators Dominate?
Bull view: The Q4 guide already reflects a quarter in which data center AI is much larger than server, and still prints about 55%. In 2027, server doubles and embedded accelerates, both accretive. Gross profit dollars are growing about 50% a year, and OpEx growth below half of revenue growth turns that into operating margin expansion of several points.
Bear view: Margin is now falling, the CFO declined both a floor and an entrance-and-exit path for 2027, and management said memory costs will be "shared" with customers rather than passed through. Q3 and the Q4 guide were both a little under the Street. A 2027 skewed further toward accelerators than Q4 would take margin lower.
Our take: We sit with the CFO's benchmark at 55% for 2027. The operating-margin story does not need gross margin to rise, only to stop falling, and the OpEx commitment carries the EPS. A gross margin below 54% in any 2027 quarter would challenge that view.
Debate: Entry Point or Value Trap After a 25% Drawdown?
Bull view: The stock is back where it was in August while 2027 estimates rose by a quarter. At about 30x FY27, the multiple is below the 36x it carried at our August downgrade, for a business growing more than 50% this year and guided above 80% next year. The drawdown came from the sector and from Intel, not from AMD's own results.
Bear view: The stock still trades near 61x this year's earnings, the after-hours gain faded, and three of the things that could hurt it (margin, cash conversion and dilution) all moved the wrong way this quarter. A sector that has just learned to worry about financing will not pay up for a vendor that guarantees its customers' leases.
Our take: On the year that matters for a twelve-month view, the bull case is the better-evidenced one. Our fair value range rises to $495–610 on an unchanged multiple, which puts the stock just below the bottom of it. We would rather own the franchise at 30x a supply-backed 2027 than at the 36x we declined in August.
Model Update & Valuation Framework
| Item | Prior Estimate (Aug 9) | New Estimate | Reason |
|---|---|---|---|
| Q4 2026 Revenue | ~$15B | ~$17.0B | Company guide midpoint; Data Center ~+50% QoQ |
| FY26 Revenue | $49–51B | ~$52.2B | $35,166M nine-month actual plus the $17.0B Q4 midpoint |
| FY26 Data Center Revenue | ~$31B | ~$33.9B | $21,039M nine-month actual plus ~$12.8B Q4 |
| FY26 Non-GAAP Gross Margin | 55.5–56% | 55.5–56% (unchanged) | Q3 55.9% and a ~55% Q4 guide offset each other |
| FY26 Non-GAAP EPS | ~$7.30 | ~$8.04 | $1.37 + $1.66 + $2.09 actual plus ~$2.92 Q4 from the guided components |
| FY26 Free Cash Flow Margin | 15–18% | 11–12% | Nine months at 13.3%; Q4 working-capital build continues |
| FY26 Capex | $3.0–3.5B | $2.5–2.8B | Nine months $1,845M; Q3 fell to $648M |
| FY27 Revenue | $78–85B | $93–98B (mid ~$95B) | Guide of >80% growth on ~$52.2B; the $93B low end is the 80% floor on a Q4 at the bottom of its range ($51.7B base); the ~$95B base case sits a little above the floor |
| FY27 Data Center Revenue | >$62B | ~$77–78B | More than 80% of revenue; server about double; data center AI significantly more than double |
| FY27 Client and Gaming Revenue | ~$14B | ~$12.5B | PC units planned down double digits; AMD outgrows the market |
| FY27 Non-GAAP Gross Margin | ~55% | ~55% (54–56%) | CFO's Q4 benchmark |
| FY27 Non-GAAP OpEx | ~$18.5B | ~$19.1B | FY26 ~$13.8B grown ~38%, under the <41% ceiling (half of revenue growth); includes acquisitions |
| FY27 Diluted Shares | ~1.72B | ~1.74B | 1,668M Q4 base plus employee grants plus ~60M weighted warrant shares (our assumption) |
| FY27 Non-GAAP EPS | $13.00–13.75 (~$13.30) | $15.60–17.50 (~$16.50) | $95B revenue, 55% GM, $19.1B OpEx, 13% tax, 1.74B shares |
| FY27 Free Cash Flow Margin | not modeled | 14–18% | Working-capital build slows as Helios matures; capex 3–4% of revenue |
| FY28 Revenue | $100–110B | $120–130B | MI500 multi-gigawatt plans; Anthropic and Meta deployments scale; 30–35% growth on FY27 |
| Long-term EPS power | Significantly above $20, timing undefined | Unchanged | The 2027 framework now reaches about $16.50; warrant dilution is the main swing |
Valuation impact. Our August fair value of $400–500 was set at 30–37x FY27 EPS of about $13.30. We keep the multiple range unchanged: faster growth argues for more, and lower cash conversion, the new customer-support exposures and warrant dilution argue for less. On FY27 EPS of about $16.50, the range becomes $495–610, with a midpoint near $553.
At the November 4 close of $488.78, AMD trades at about 29.6x our FY27 estimate and 60.8x our FY26 estimate, with a market capitalization near $812B on 1,661M diluted shares. On the same FY27 year, the August price of $483.36 was 36.3x our then-estimate. The stock is 1.3% below the bottom of the new range, with 13.1% upside to the midpoint and 24.8% to the top.
Thesis Scorecard Post-Earnings
August Commitments, Graded
| Commitment to watch (from our Q2 note) | Outcome | Grade |
|---|---|---|
| (1) Q3 revenue $12.7–13.3B, ~56% GM, ~$3.65B OpEx, 13% tax, ~1.66B shares; ~$1.93 EPS | $13,377M; 55.9%; $3,560M; 13%; 1,661M; $2.09 | Beat |
| (2) Helios initial shipments in Q3, ramp through Q4 | MI455X shipped in volume; five customers live; "big step up" in Q4 | Met |
| (3) Server revenue >80% YoY in 2H26 | Cloud and enterprise each more than doubled; outlook "significantly exceeded" | Beat |
| (4) Q4 Data Center sequential growth exceeds Q3's | ~+50% guided against +27% in Q3 | Beat |
| (5) Capex near $800M+ per quarter, and a guided run rate | $648M; no run rate given | Mixed |
| (6) OpEx keeps growing slower than revenue | +5% QoQ vs. +16% revenue; 2027 commitment tightened to less than half of revenue growth | Met |
| (7) Gross margin holds 56% or breaks lower | 55.9% in Q3; ~55% guided for Q4 | Breaking lower |
| (8) Shipped Helios memory configurations drift down | Not addressed; costs to be "shared" with customers | Open |
| (9) Any 2027 gross margin framework | Q4 as "a good benchmark"; no range or floor | Partial |
| (10) Buybacks resumed | None under the program; $4.7B of new debt; $1,495M withheld for employee taxes | Not met |
Standing Thesis
| Thesis Point | Status | Notes |
|---|---|---|
| Bull #1: Server CPU share compounding | Confirmed + | Cloud and enterprise each more than doubled; 2027 server about double, mostly on units; Venice shipping with 240+ platforms; head-node wins including OpenAI's custom accelerator. Status tag holds ON TRACK. |
| Bull #2: Instinct has real hyperscale anchors | Confirmed + | Anthropic, Meta, Microsoft, OpenAI and Oracle live on Helios in the first revenue quarter; data center AI passes server in Q4; MI500 multi-gigawatt planning with several labs. Status tag holds ON TRACK. |
| Bull #3: Operating leverage holds as revenue scales | Split | Operating margin 29.3% and a tightened 2027 OpEx commitment strengthen the income-statement half. Free cash flow fell to a 4.1% margin, weakening the cash half further. Status tag holds AT RISK. |
| Bull #4: Agentic AI structurally expands the server CPU TAM | Confirmed | Helion all-CPU racks and a 2027 server doubling give the thesis a product and a number; the 2030 TAM held at about $220B after three moves in nine months. Status tag holds ON TRACK. |
| Bear #1: Instinct dilutes gross margin as it ramps | Materializing | Q4 guided to ~55%, about 90 bps below Q3 and under the Street; the CFO called data center AI "a little bit dilutive"; no 2027 floor. Status tag moves EMERGING to MATERIALIZING. |
| Bear #2: Memory/component cost drag on client and gaming | Materializing | PC sell-through slowed late in Q3; Client missed the Street; Q4 Client and Gaming guided down more than 20%; 2027 PC units planned down double digits. Status tag holds MATERIALIZING. |
| Bear #3: China / export-control risk on accelerators | Dormant | Not mentioned on the call; the 10-Q discloses MI325 licenses for certain China customers, subject to a 25% tariff on U.S. inspection. Status tag holds CONTAINED. |
| Bear #4: Price has outrun the thesis | Reversed | Price unchanged since August while our FY27 EPS estimate rose about 24%; FY27 multiple 30x against 36x; the stock now sits just below our fair-value range. Status tag moves MATERIALIZING to CONTAINED. |
| Bear #5: Cash conversion is structurally lower selling racks than chips | Materializing | Free cash flow margin 4.1%; inventory up $3.0B to $11.5B (about 176 days); supply prepayments; the CFO guided continued inventory and capacity investment into 2027. Status tag moves EMERGING to MATERIALIZING. |
| Bear #6 (NEW): AMD is underwriting its customers' build-out | New — Emerging | $8.5B of lease and debt guarantees, $13.3B of leases to sublease, $6.2B of compute backstops, up to $5.1B of investment commitments, and warrants for up to 320M shares to two customers. No defaults; not sized on the call. Status tag opens at EMERGING. |
Overall: The operating thesis strengthened again and the investment thesis turned. All four bull pillars held, and two improved materially on the 2027 framework. Three bear points escalated or opened (gross margin, cash conversion and customer support), but the one that drove the August downgrade, price, reversed: the stock did not move while our earnings estimate rose by about a quarter. That is the mirror image of August's configuration, and our framework, built on relative twelve-month return, reads it the other way.
Action: Upgrading to Outperform from Hold; raising fair value to $495–610 from $400–500; conviction stays at 6/10. We grade our August triggers as published. The upgrade trigger read: "We would return to Outperform on a move toward the low $400s absent thesis damage, or on evidence that either the Helios ramp is running materially ahead of the guided cadence or that gross margin can expand through the dilution." The low-$400s condition did not occur (the lowest close since August was $456.16), and the gross-margin condition did not occur. The Helios condition did. The CEO said Helios production is "ramping as planned", but the August guided cadence asked only that Q4 Data Center growth exceed Q3's. Q4 Data Center is guided up ~50% against Q3's 27%, about $12.8B on our arithmetic against the ~$10.2B we implied in August, and Q3 Data Center of $8,546M beat the ~$8.1B we implied. In the CFO's words, data center AI will be "much higher" than server in Q4, and Q3 data center AI was "actually better than our expectation". Server outperformance explains part of the Data Center upside, but not the Q4 crossover. The downgrade trigger read: "We would move to Underperform only if cash conversion stays in the low teens into 2027 while the multiple holds." It has not fired: free cash flow conversion fell below the low teens, but 2027 has not started and the FY27 multiple compressed rather than held. We carry it forward unchanged. We add two new triggers, both outside our estimate range. We would return to Hold if the combined free cash flow margin for Q1 and Q2 2027 is below 10% (our FY27 range is 14–18%), or if management cuts the 2027 revenue framework below 70% growth (our range is 78–88%). If the Q4 print beats on revenue and misses on cash, we will not let the first offset the second: the first-half 2027 cash test decides.