The Business Got Bigger, the Cash Got Smaller, and the Price Got There First: Downgrading AMD to Hold
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 | $44,962 | $52,359 | $60,997 |
| Less: Cost of sales | ($11,278) | ($12,114) | ($16,456) | ($19,908) | ($22,442) | ($25,924) |
| Less: Amortization of acquisition-related intangibles (COGS) | ($942) | ($946) | ($1,031) | ($1,035) | ($982) | ($900) |
| Gross Profit | $10,460 | $12,725 | $17,152 | $24,020 | $28,935 | $34,173 |
| Less: Research and development | ($5,872) | ($6,456) | ($8,091) | ($10,554) | ($12,304) | ($14,029) |
| Less: Marketing, general and administrative | ($2,318) | ($2,735) | ($4,144) | ($5,245) | ($6,021) | ($6,710) |
| 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) | ($16,949) | ($19,424) | ($21,739) |
| Operating Income (Loss) | $401 | $1,900 | $3,694 | $7,070 | $9,511 | $12,434 |
| 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 | $7,582 | $10,023 | $12,944 |
| Less: Income tax provision (benefit) | $346 | ($381) | $103 | ($1,076) | ($1,504) | ($1,942) |
| Equity income in investee | $16 | $33 | $26 | $6 | $0 | $0 |
| Net Income from Continuing Operations | $854 | $1,641 | $4,269 | $6,512 | $8,520 | $11,003 |
| Income (Loss) from Discontinued Operations, net of tax | $0 | $0 | $66 | $11 | $0 | $0 |
| Net Income | $854 | $1,641 | $4,335 | $6,523 | $8,520 | $11,003 |
| Ratios & Assumptions | ||||||
| Gross Margin | 46.1% | 49.4% | 49.5% | 53.4% | 55.3% | 56.0% |
| Operating Margin | 1.8% | 7.4% | 10.7% | 15.7% | 18.2% | 20.4% |
| Net Margin | 3.8% | 6.4% | 12.5% | 14.5% | 16.3% | 18.0% |
| Effective Tax Rate | (70.3%) | 19.2% | (2.5%) | 14.2% | 15.0% | 15.0% |
The full workbook adds 17 historical and 7 projected quarters, plus KPI Drivers · Balance Sheet · Cash Flow Statement — every subtotal a live formula, every projection traced to a driver.
Key Takeaways
- A double beat that cleared the bar by inches. Revenue $11,536M (+50% YoY, +13% QoQ) beat the $11.3B Street by 2.2%; non-GAAP EPS $1.66 beat $1.62 by 2.5%. Both margins of victory are well below Q1's (+3.7% revenue, +7.9% EPS). Data Center did the heavy lifting at $6,718M, +107% YoY and 58% of the company. Management delivered on every Q1 commitment we were tracking: revenue landed above its own $11.2B guide, gross margin hit the guided 56%, and EPYC grew the promised 70%-plus.
- 2027 got materially bigger, again. Management now guides Data Center segment revenue to more than double year over year in 2027, with server alone growing more than 70% off a much higher base, and told the Street on the call that "well over 100%" is the better read. Anthropic joined OpenAI, Meta and Microsoft as a named anchor, committing to up to 2 GW of MI450 in Helios racks. The 2030 server CPU TAM was raised a second time in two quarters, from $120B to roughly $220B. The long-term EPS anchor moved from more than $20 to a commitment to "significantly exceed" it.
- The cash statement is where this quarter turned. Free cash flow fell to $1,558M (14% of revenue) from $2,566M (25%) in Q1, on revenue that grew 13% sequentially. Capex more than doubled sequentially to $808M from $389M, against $282M a year ago. Receivables rose $1,246M and inventory reached $8,468M, a build management frames as supporting data center demand. Buybacks were zero in the quarter. None of this is alarming on its own; together it marks the point where AMD stopped being a self-funding compounder and became a capital-intensive one.
- Gross margin is guided flat into the dilutive ramp. Q3 non-GAAP gross margin is guided to approximately 56%, unchanged from Q2, even as revenue steps up 13% sequentially. Helios and MI450 carry gross margins the CFO described as slightly below the corporate average, and they begin shipping this quarter. The offsetting accretion from server, Embedded and commercial client is now fully spoken for. The margin algorithm buys AMD nothing in the Q3 guide.
- Rating: Downgrading to Hold from Outperform. Raising fair value to $400–500 from $325–400. Nothing here breaks the thesis; the thesis is working. The problem is that the stock has worked faster. AMD closed at $483.36 on August 7 after a +142% year-to-date run into the print, which puts it at roughly 66x our FY26 EPS estimate of about $7.30 and 36x our FY27 estimate of about $13.30. That already sits in the upper half of our raised fair-value range. A fourth consecutive beat-and-fade tells you the bar is now set by the buy side, not the sell side. We do not sell an improving franchise into strength, but we no longer expect it to beat the index over the next twelve months from this price.
Results vs. Consensus
| Metric | Q2 2026 Actual | Consensus | Result | Magnitude |
|---|---|---|---|---|
| Revenue | $11,536M | $11,284M | Beat | +2.2% / +$252M |
| Non-GAAP Gross Margin | 56% | ~56% (company guide) | In line | Met the guide exactly |
| Non-GAAP Operating Income | $3,094M | n/a | Beat | 27% margin, up from 25% |
| Non-GAAP Diluted EPS | $1.66 | $1.62 | Beat | +2.5% / +$0.04 |
| Non-GAAP EPS vs. buy-side whisper | $1.66 | $1.69 | Short | −1.8% / −$0.03 |
| GAAP Diluted EPS | $1.38 | n/a | n/a | Includes a $483M investment gain |
| Data Center Revenue | $6,718M (+107% YoY) | ~$6.0B | Beat | +~12% |
| Free Cash Flow | $1,558M | n/a | Weak | 14% margin vs. 25% in Q1 |
| Q3 Revenue Guide (mid) | $13.0B ± $300M | $12.5B | Beat | +$500M / +4.0% |
| Q3 Guide vs. buy-side whisper | $13.0B | ~$14B | Short | −~7% |
| Q3 Non-GAAP GM Guide | ~56% | Above 56% hoped | Flat | No sequential expansion |
Year-over-Year Comparison
| Metric | Q2 2026 | Q2 2025 | Reported YoY | Comparable YoY |
|---|---|---|---|---|
| Revenue | $11,536M | $7,685M | +50% | +50% |
| Non-GAAP Gross Profit | $6,488M | $3,326M | +95% | +57% |
| Non-GAAP Gross Margin | 56% | 43% | +13 ppts | +over 200 bps |
| Non-GAAP OpEx | $3,394M | $2,429M | +40% | +40% |
| Non-GAAP Operating Income | $3,094M | $897M | +245% | +82% |
| Non-GAAP Diluted EPS | $1.66 | $0.48 | +246% | +~82% |
| GAAP Operating Income | $1,990M | $(134)M | n/a | n/a |
| GAAP Net Income | $2,297M | $872M | +163% | n/a |
| Free Cash Flow | $1,558M | $1,180M | +32% | +32% |
| Capital Expenditures | $808M | $282M | +187% | +187% |
Comparable-basis columns exclude the approximately $800M of inventory and related charges recorded in Q2 2025 in connection with U.S. export controls on MI308. Revenue, OpEx and cash flow were unaffected by that charge, so those rows are identical on both bases.
Sequential Comparison
| Metric | Q2 2026 | Q1 2026 | QoQ |
|---|---|---|---|
| Revenue | $11,536M | $10,253M | +13% |
| Non-GAAP Gross Margin | 56% | 55% | +1 ppt |
| Non-GAAP OpEx | $3,394M | $3,145M | +8% |
| Non-GAAP Operating Income | $3,094M | $2,540M | +22% |
| Non-GAAP Operating Margin | 27% | 25% | +2 ppts |
| Non-GAAP Diluted EPS | $1.66 | $1.37 | +21% |
| Data Center Revenue | $6,718M | $5,775M | +16% |
| Operating Cash Flow | $2,366M | $2,955M | −20% |
| Capital Expenditures | $808M | $389M | +108% |
| Free Cash Flow | $1,558M | $2,566M | −39% |
| Free Cash Flow Margin | 14% | 25% | −11 ppts |
| Inventories | $8,468M | n/a (not disclosed in release) | $7,920M at Dec 27, 2025 |
Revenue
The 50% headline is clean and unusually broad for a company this size. Every reportable segment grew year over year, and the growth is driven by the businesses with the most durable competitive position rather than by a single lumpy product. Data Center contributed $3,478M of the $3,851M of incremental year-over-year revenue, which is to say Data Center is the growth story now at 58% of the company, up from 42% a year ago. Within it, management disclosed a fifth consecutive quarter of record server CPU revenue with cloud and enterprise sales each growing more than 70%, and Instinct sales more than doubling. That is a rare combination: the mature franchise is accelerating at the same time the new one is scaling.
The complication is the beat's size relative to the setup. A 2.2% revenue beat is a perfectly good result in isolation and a thin one against a stock that had rallied 20.7% in the four sessions immediately preceding the print. Management guided Q2 to $11.2B at the Q1 call and delivered $11,536M, which clears its own high end. The buy side had already spent that.
Margins
Non-GAAP gross margin of 56% met the guide exactly and expanded 80 basis points sequentially. On the comparable basis that strips the year-ago MI308 charge, the year-over-year expansion is over 200 basis points rather than the reported 13 percentage points. Readers should hold onto that distinction: the reported gross-profit growth of 95% and EPS growth of 246% are arithmetic artifacts of a depressed base, and the honest underlying numbers are 57% gross-profit growth and 82% EPS growth. Management gave the 82% on the call; the 57% is our derivation. Those are still very good numbers. They are not the numbers in the press-release table.
Non-GAAP operating margin reached 27%, up two points sequentially, because gross margin expanded while OpEx grew 8% against revenue growth of 13%. That is real operating leverage after several quarters where OpEx outran the top line. OpEx is still up 40% year over year, and the CFO restated the framework in Q&A: expenses are managed to grow slower than revenue. This quarter that promise was kept for the first time in a while.
EPS and Below the Line
Non-GAAP EPS of $1.66 is an operational number and we treat it as clean. GAAP EPS of $1.38 is not comparable to prior periods without care: it includes a $483M net gain on long-term investments, which is why GAAP net income of $2,297M grew 163% while GAAP operating income grew off a negative base. Other income was $598M in the quarter against $165M in Q1. Strip the investment gain and GAAP earnings power looks materially less dramatic. The non-GAAP reconciliation correctly removes it, so the $1.66 is unaffected, but anyone modeling from GAAP should not extrapolate the other-income line.
The share count was 1,659M diluted, essentially flat sequentially, and there were no repurchases in the quarter. There is no share-count assistance in this EPS figure.
Cash Flow
This is the section the market read. Operating cash flow fell 20% sequentially to $2,366M on revenue that rose 13%. Capex more than doubled to $808M. Free cash flow of $1,558M translates to a 14% margin, against 25% in Q1 and 15% in the year-ago quarter. The working-capital drivers are visible in the statement: receivables consumed $1,246M, inventories $423M and prepaid and other assets $683M, partly funded by a $2,274M increase in payables. Inventory ended at $8,468M against $7,920M at the December year end.
Assessment: Every individual line has a benign explanation. Receivables build when revenue steps up 13% late in a quarter. Inventory builds ahead of a Helios ramp. Capex rises when you are adding test and packaging capacity into a supply-constrained ramp. Taken together, though, the shape has changed, and the change is directionally permanent: AMD is now pre-funding a rack-scale systems business with balance sheet rather than harvesting a fabless chip business. The payables increase is doing a lot of work to hold operating cash flow up, and payables are a source you can only pull once. We expect free-cash-flow conversion in the teens for the next two to three quarters, not the mid-twenties. That belongs in the model and it belongs in the multiple.
Segment Performance
| Segment | Q2'26 Revenue | YoY | QoQ | Operating Income | Operating Margin | Margin Y/Y |
|---|---|---|---|---|---|---|
| Data Center | $6,718M | +107% | +16% | $2,103M | 31% | from $(155)M loss |
| Client and Gaming | $3,841M | +6% | +7% | $582M | 15% | 21% a year ago |
| Client | $3,062M | +23% | +6% | (within C&G) | n/a | n/a |
| Gaming | $779M | −31% | +8% | (within C&G) | n/a | n/a |
| Embedded | $977M | +19% | +12% | $386M | 40% | 33% a year ago |
| All Other | n/a | n/a | n/a | $(1,081)M | n/a | $(1,021)M a year ago |
| Total | $11,536M | +50% | +13% | $1,990M (GAAP) | 17% (GAAP) | from $(134)M loss |
Data Center
Revenue of $6,718M more than doubled year over year and grew 16% sequentially, and operating income of $2,103M at a 31% margin is the number that deserves attention. A year ago this segment lost $155M. It now generates more operating income than the rest of AMD combined, before the corporate cost pool. The segment crossed 58% of company revenue from 42% a year ago, which means AMD's consolidated financial profile is now essentially a data center company with a large, profitable PC attachment.
Two engines, and management was unusually clear about both. Server CPU delivered a fifth consecutive record quarter with cloud and enterprise sales each up more than 70%, and the growth was more unit-led than price-led, which is the healthier composition. Instinct sales more than doubled on the MI350 series with adoption broadening across labs, clouds, startups, national labs and sovereign deployments.
"Growth was driven by greater than 70% year-over-year increase in EPYC sales with record enterprise sell-through and robust demand across our cloud customer base. Both unit shipments and ASP increased significantly year-over-year reflecting the continued mix shift powered to our latest Zen 5 generation processors." — Jean Hu, CFO
The forward setup is where the segment gets interesting and where the debate now sits. Sixth-generation EPYC Venice, built on Zen 6 and a 2-nanometer process, is in production with every major OEM tracking to launch and lead cloud providers deploying later this year. Management characterized customer demand for Venice as stronger than for any prior EPYC generation. On the accelerator side, Helios entered production with initial shipments beginning in Q3 and a ramp through Q4 into 2027.
"In server CPUs with very strong customer demand and improved supply, we now expect server revenue to grow more than 80% year-over-year in the second half of 2026 and more than 70% for the full year 2027, off a much higher base. Taken together, we now expect data center segment revenue to more than double year-over-year in 2027." — Lisa Su, Chair and CEO
Assessment: This is the strongest segment disclosure AMD has produced. A 31% operating margin on a business doubling annually is what the bull case looked like on paper three years ago, and it has arrived. Our reservation is not about the segment; it is that the segment's 2027 trajectory is now the entire equity story, and the stock has already been marked to a successful version of it. Note also that Data Center's incremental operating margin is flattering the corporate figure while the corporate cost pool grew only 6% year over year; that leverage is finite.
Client and Gaming
The segment grew 6% to $3,841M with the two halves moving in opposite directions. Client rose 23% to $3,062M on record mobile processor revenue and continued share gains, with commercial the standout: Ryzen Pro sales grew more than 50% year over year on new wins across healthcare, technology, automotive and financial services, and the major OEMs launched a broad commercial lineup on Ryzen AI Pro 400. Gaming fell 31% to $779M on lower semi-custom revenue at this point in the console cycle, with gaming graphics also down as industry-wide component costs pushed card prices up.
The margin line is the one to watch. Segment operating income of $582M at 15% compares to $767M at 21% a year ago. Revenue grew and profit fell.
"Client and Gaming segment operating income was $582 million or 15% of revenue compared to $767 million or 21% a year ago, reflecting continued investment in go-to-market activities and expanding product road map." — Jean Hu, CFO
Management guided the segment down modestly in Q3, with slight client growth offset by a strong double-digit gaming decline, and flagged the second-half PC market explicitly.
"Looking to the second half of the year, we are planning for a softer PC market as higher memory and component costs weigh on demand." — Lisa Su, Chair and CEO
Assessment: Client is executing well and the commercial mix shift is genuinely valuable, but the six-point operating-margin decline on positive revenue growth is a real deterioration that the segment's headline growth obscures. Management attributes it to go-to-market spend and roadmap breadth rather than pricing, which we accept for now. Gaming is doing what a late-cycle console business does and is no longer large enough to matter to the thesis at 6.8% of revenue. The memory-cost warning we flagged last quarter as an active bear point is now visibly landing in the numbers.
Embedded
Revenue of $977M grew 19%, the segment's strongest year-over-year growth in more than three years, and 12% sequentially, with operating margin expanding to 40% from 33%. Demand was broad across networking, aerospace and defense, test and measurement, and communications. The structural change is embedded x86, which grew significantly as hyperscalers and networking customers adopted AMD CPUs for networking and control-plane functions inside the data center. Design wins are tracking to more than $18B for the year.
"Embedded x86 is becoming a significant growth driver for the segment." — Lisa Su, Chair and CEO
Assessment: Embedded has quietly become the best-margin business in the company and finally has a growth vector attached to it. At 40% operating margin it is the only segment structurally accretive to corporate gross margin, and it is the offset management is counting on to absorb Helios dilution in 2027. Small in revenue, disproportionately important to the margin algorithm.
Key Topics & Management Commentary
Overall Management Tone: Assured and forward-leaning, with the emphasis shifted almost entirely from the current quarter to 2027. Where the Q1 call spent its energy establishing that demand was real, this call took that as settled and spent its energy raising the size of the prize, which is a different and more confident posture. The one place the tone thinned was margins and cash: answers there were framework-level and deferred to a later update rather than quantified, and that asymmetry between a precise revenue outlook and an unquantified profitability outlook was the call's only soft seam.
1. The Server CPU TAM Was Raised Again, from $120B to $220B
Three quarters ago the 2030 server CPU TAM was roughly $60B. At the Q1 call it became $120B-plus at a greater-than-35% growth rate. It is now approximately $220B at a greater-than-50% rate. Alongside it, the data center AI accelerator market was put at approximately $1.4 trillion by 2030, and the combined high-performance and AI compute market at approaching $2 trillion.
"we now expect the data center AI accelerator market to grow more than 45% annually to approximately $1.4 trillion by 2030. And we expect the server CPU market to grow more than 50% annually to approximately $220 billion by 2030." — Lisa Su, Chair and CEO
The mechanism management describes is agentic AI, decomposed into general-purpose compute, AI head nodes, and agentic sandbox workloads, with the last described as the smallest bucket today and the fastest-growing through 2030.
Assessment: We wrote last quarter that the risk to the $120B number was to the upside, and it was. That does not mean we should treat a near-quadrupling of a 2030 TAM in nine months as a measurement. It is a forecast, revised upward three times as the near-term demand signal improved, which is exactly the pattern a cyclical peak also produces. We continue to underwrite the demand as real and the trajectory as directionally right, and we decline to capitalize a 2030 TAM figure into today's multiple.
2. Data Center Guided to More Than Double in 2027
The single most consequential forward statement on the call was not a TAM number but a growth number: Data Center segment revenue more than doubling year over year in 2027, with server alone up more than 70% off a much higher base. Pressed twice in Q&A on how much work the word "more" was doing, management upgraded the phrasing to "well over 100%," and separately told a caller whose model implied roughly $30B of 2027 Instinct revenue that the figure was probably too low.
Run the arithmetic from the Q3 guide. Backing into the segment split implied by the guided components leaves Data Center at roughly $8.1B in Q3, and management said Q4 growth will exceed Q3's. That puts FY26 Data Center around $31B. More than doubling from there is a 2027 Data Center segment above $62B, which on our estimates lands total company revenue somewhere in the high $70s to low $80s billions.
Assessment: That number is far above where published consensus sits, and management volunteered it fourteen months ahead of the fact. Two readings are available. The generous one is that customer commitments are now firm enough to underwrite it, which the Anthropic, Meta, OpenAI and Microsoft roster supports. The cautious one is that a company guiding 2027 this precisely, this early, in a market where its own TAM estimate has moved three times, is making a forecast that will be marked to market four more times before it is tested. We think the first reading is more likely and the second is not dismissible.
3. Anthropic Becomes the Fourth Named Anchor
Anthropic committed to deploy up to 2 GW of MI450 series GPUs in Helios racks, with the first gigawatt beginning in the first half of 2027, alongside a multiyear engineering collaboration using Claude to optimize Instinct workloads and accelerate ROCm development. Microsoft separately expanded to deploy Helios at scale on Azure for frontier-model inference, and added EPYC VM series and broader Pensando deployment.
"Anthropic will deploy up to 2 gigawatts of MI450 series GPUs in Helios with deployment of the first gigawatt beginning in the first half of 2027." — Lisa Su, Chair and CEO
With Meta at up to 6 GW and OpenAI's multi-gigawatt commitment, management characterized revenue per gigawatt as double-digit billions, unchanged from prior framing.
Assessment: The bear case that AMD wins benchmarks and loses deployments is now comprehensively dead. Three named frontier-scale anchor customers, plus Microsoft deploying Helios at scale on Azure, is a different competitive position than AMD occupied twelve months ago. The caveat is timing: Anthropic's first gigawatt begins in the first half of 2027, which places it beyond the current guidance horizon entirely. This is a 2027 and 2028 revenue event being priced in 2026.
4. Gross Margin Is Guided Flat Into the Ramp
Q3 non-GAAP gross margin is guided to approximately 56%, identical to Q2's actual, on revenue stepping up 13% sequentially. The reason is mix. Helios and MI450 begin shipping this quarter at gross margins management described as slightly below the corporate average, and the accretion from server, Embedded and commercial client is now precisely offsetting rather than exceeding that drag.
"I think as we talked before, our gross margin is primarily driven by the mix -- business mix. We're actually very pleased with our gross margin progress in the first half of 2026 and our guidance for Q3 2026." — Jean Hu, CFO
Asked directly about the 2027 shape, management declined to size it, describing the outcome as determined by the relative pace of the server and data center AI ramps and promising more color closer to the year.
Assessment: We flagged this a quarter ago as the bear point most likely to move from dormant to active, and it has. The flat guide is not bad; 56% on a heavily diluting mix is a good outcome. It is that the margin algorithm has stopped contributing incremental EPS, which matters when the revenue outlook is already fully reflected. From here, growth in earnings must come from revenue and OpEx discipline alone.
5. Free Cash Flow and the Capex Inflection
Capex of $808M nearly tripled year over year and more than doubled sequentially. Management did not address it in prepared remarks and was not asked about it in Q&A, which is itself notable given it was the single largest sequential change in the cash statement. The CFO's only balance-sheet color was the inventory build.
"Inventory increased sequentially to approximately $8.5 billion to support strong data center demand." — Jean Hu, CFO
Assessment: A rack-scale systems business carries working capital and fixed-asset intensity that a chip business does not. That is the cost of moving up the stack, and it is a reasonable price for the revenue it buys. The issue is that it was neither guided nor discussed. AMD has trained investors to model a high-conversion, capital-light franchise, and this quarter quietly retired that assumption without saying so. We would rather management sized the new capex run rate than left the Street to discover it in the cash flow statement.
6. Supply Moved From Constraint to Enabler
Management characterized the server CPU supply chain as tight through the first half, with the tightness a consequence of demand that was not forecast rather than of any manufacturing problem, and described 2027 supply as better because the demand is now visible in advance. Improved supply was cited explicitly as a reason the second-half server outlook could be raised to more than 80% growth. On the 2-nanometer ramp for Venice, management noted that AMD's chiplet approach requires fewer leading-node wafers than a monolithic design.
Assessment: This is the most tangible upgrade in the quarter and the least discussed. AMD spent the first half leaving server revenue on the table for want of wafers, substrates and back-end capacity. Removing that constraint converts demand into revenue mechanically. The chiplet point is a genuine structural advantage at 2 nanometers, where leading-node capacity is the binding constraint for everyone.
7. ROCm Reaches Its Inflection
Management described the software stack as having reached the point where customers can deploy in production at scale, with more than 3 million models running out of the box, day-zero support on leading open models, and a tenfold increase in open-source contributions over the past year. ROCm.ai, introduced last month, lets developers use Claude, Codex and Cursor to port and optimize code for Instinct, and was cited as delivering more than twice the training performance and roughly 3x the inference performance of ROCm 7.
Assessment: Software has been the durable bear argument against Instinct for a decade and it is finally getting quantitative rather than aspirational. The AI-assisted porting angle is more interesting than the benchmark claims: if the cost of moving a workload from the incumbent stack falls toward zero, the switching-cost moat that has protected the competitor's share erodes from the direction nobody was defending. We are not yet ready to score this as a confirmed thesis pillar, but it is the item most likely to become one.
8. The Long-Term Model Was Raised Twice in One Paragraph
The November Financial Analyst Day framework was greater than 35% revenue growth and more than $20 of annual EPS within the strategic timeframe. Management raised both.
"We now expect revenue to grow substantially above our prior target of greater than 35%, and we expect to significantly exceed our $20 annual EPS target within our strategic timeframe." — Lisa Su, Chair and CEO
Assessment: Reaffirmed and raised in the same breath, nine months after it was set. The signal is real. The complication for a rating is arithmetic: at roughly 1.66B diluted shares, $20 of EPS is about $33B of net income, and at AMD's target margins that requires revenue well north of $120B. The market is now paying a substantial multiple for a number that management has placed inside an undefined "strategic timeframe." That is a fine thing to own and a difficult thing to underwrite at 36x our 2027 estimate.
9. MI500 and the Annual Cadence
Management committed to a new rack-scale platform every year, with the 2027 generation combining MI500 series GPUs, Verano CPUs and Pensando networking, and expanded scale-up domains in both copper and optical interconnects. MI500 was described as the largest generational leap in Instinct history, putting AMD on track to increase inference performance more than 2,000 times in four years, with multiple customers already engaged on next-generation planning.
Assessment: The annual cadence commitment is the more important disclosure. Matching the incumbent's release rhythm is the entry ticket to being designed into multi-year data center plans at all, and AMD has now committed to it publicly with a named 2027 platform. Execution risk on an annual rack-scale cadence is substantial and should not be underestimated; this is a systems, networking and software problem more than a silicon one.
10. Client Holds Up Better Than the Market It Sells Into
Management is planning for a softer second-half PC market on memory and component costs while expecting the client business to outgrow it on Ryzen strength and commercial adoption, and noted that the market has actually held up better than most expected so far. The next-generation Ryzen AI Halo platform on the Gorgon Halo processor carries 192 GB of unified memory and can run models to 300 billion parameters locally, bundled with a year of Hugging Face Pro.
Assessment: Consistent with last quarter and appropriately scoped. Client is a share-gain story inside a shrinking market, which is a fine place to be and not a swing factor for the equity. The local-AI positioning is a legitimate multi-year option that is not worth capitalizing yet.
11. Embedded Design Wins at a Record
More than $18B of new design wins are tracking for the year, led by networking, data center, communications, test, and aerospace and defense. Embedded x86 adoption inside hyperscaler networking and control-plane infrastructure is the newest and fastest-moving piece, and the Ryzen AI Embedded X100 series and Kria robotics platform extend the portfolio toward edge and physical AI.
Assessment: Design wins convert to revenue over three to five years, so the $18B figure is a statement about 2028 through 2030, not about now. It is the cleanest evidence that the Xilinx franchise has been successfully repositioned rather than merely absorbed. At a 40% operating margin, every incremental dollar here is worth roughly double a dollar of Instinct revenue to the earnings line.
Guidance & Outlook
| Metric | Q2'26 Actual | Q3'26 Guide | Implied Change | vs. Street |
|---|---|---|---|---|
| Revenue | $11,536M | $13.0B ± $300M | +13% QoQ, +41% YoY | +$500M / +4.0% vs. $12.5B |
| Non-GAAP Gross Margin | 56% | ~56% | Flat | Below hoped expansion |
| Non-GAAP OpEx | $3,394M | ~$3,650M | +7.5% QoQ | Grows slower than revenue |
| Non-GAAP Other Income | n/a | +$55M | n/a | n/a |
| Non-GAAP Tax Rate | 13% | 13% | Unchanged | n/a |
| Diluted Share Count | 1,659M | ~1,660M | Flat | n/a |
| Implied Non-GAAP EPS | $1.66 | ~$1.93 ($1.84–$2.02) | +16% QoQ | Our calculation from guided components |
The guide is a genuine raise against published consensus and a disappointment against the buy-side bar, and both facts are load-bearing. At the midpoint, revenue grows 13% sequentially for the second consecutive quarter with gross margin flat and OpEx growing more slowly than revenue, which produces roughly $1.93 of non-GAAP EPS on our arithmetic from the guided components. That is 16% sequential earnings growth. It is a good quarter being guided.
Implied segment composition. Management guided Data Center to strong double-digit sequential growth, Embedded to strong double-digit sequential growth, and Client and Gaming to a modest sequential decline with slight client growth offset by a strong double-digit gaming decline. Solving the $13.0B midpoint against those descriptions puts Data Center near $8.1B, Client and Gaming near $3.75B, and Embedded near $1.1B. Management also said Q4 Data Center growth will exceed Q3's, which implies a fourth quarter around $15B and a full year near $49–51B.
Where the Street sits. Published consensus entering the print was $12.5B for Q3. Post-print, the sell side raised numbers and price targets rather than cutting them, which tells you the disappointment lived on the buy side. The whisper was approximately $14B, roughly a billion dollars above the guide. A company that beats the published number by 4% and misses the unpublished one by 7% will sell off, and did.
Guidance style. AMD's recent pattern has been to guide slightly above whisper and then beat by two to five points. Q2 broke that: the company guided $11.2B and printed $11,536M, a 3% beat of its own guide, which is at the low end of its own range. The Q3 guide of $13.0B against that pattern implies an actual near $13.3–13.6B. Management is guiding conservatively into a supply situation it has described as improving, which we read as prudent rather than as a signal of weakness.
Analyst Q&A Highlights
Whether "Acceleration" in the Press Release Survives Contact With the Guide
The sharpest exchange on the call was a direct challenge to the press release's own language. The release said Data Center sales would accelerate in the second half; the Q3 guide implies year-over-year Data Center growth in the eighties against 107% in Q2, which is deceleration on any quarterly reading. Management clarified that the acceleration claim compares the second half to the first half in aggregate, not Q3 to Q2. That is a defensible construction and not the one a casual reader of the release would form.
Q: "First thing, I wanted to pick on something in some of the press release. It tells you that data center accelerates in the second half. So it grew 107% year-over-year in Q2, you're guiding it up like in the 80s, though, like in Q3. So how do I interpret which is not acceleration, how do I interpret that statement?"
— Stacy Rasgon, Bernstein Research
A: "I think when we talk about the acceleration, we are talking about the second half versus the first half. If you look at the first half, our data center business year-over-year growth versus the second half, we do think there's an acceleration. That's the reflection there."
— Jean Hu, CFO
Assessment: Management answered directly and the answer holds up, but the exchange is a useful reminder that the release's framing was written to flatter. The half-over-half comparison is the correct one given the Helios ramp lands in Q4, and the Q3 number is not the tell. We would prefer the release said so.
How Much Work the Word "More" Is Doing in the 2027 Outlook
A recurring line of questioning tried to convert management's qualitative 2027 framing into a number, on the reasonable grounds that "more than 70%" and "more than double" are floors rather than estimates. Management upgraded the language in real time from more than 100% to "well over 100%" for the Data Center segment, which is about as close to a guide as a company will give fourteen months out.
Q: "So you said like more -- I think servers more than 70%, the total more than doubling. I guess what I'm asking is how much work is the more in that statement doing?"
— Stacy Rasgon, Bernstein Research
A: "Stacy, you bring a smile to my face. So I would say, we're trying to give you a way to think about 2027. And yes, the server CPU we're saying is more than 70%, which we think at this point is a very strong statement just given where the business is. And we do expect the overall data center business to be well over 100%."
— Lisa Su, Chair and CEO
Assessment: The upgrade from "more than" to "well over" inside a single Q&A session is meaningful and was clearly deliberate. It also confirms that the published floor is not the internal expectation, which is the useful piece of information for anyone modeling 2027.
Whether a $30B Instinct Model for 2027 Is Too High or Too Low
One caller put a number on the table by working backwards from the Data Center framing to roughly $30B of 2027 Instinct revenue and asked management to react. The response was that the number was probably too low, which is the most specific quantitative signal management gave about the accelerator business all call. The companion question on Instinct gross margin trajectory was not answered.
Q: "If I take your data center guide, Lisa, it sort of implies Instinct revenues of $30 billion, give or take. And so curious, 2 parts on this. How do you see kind of the revenue cadence first half, second half and then if we isolate to instinct only and start thinking about 455X and Helios, how should we think about the underlying gross margins for that business starting in Q1 and then exiting in Q4?"
— Christopher Muse, Cantor
A: "I think what you're hearing from us is that your data center AI number is probably too low. And maybe back to Stacy's question without going into exact numbers, I think this notion of over 100% should consider be well over 100%."
— Lisa Su, Chair and CEO
Assessment: A remarkable disclosure on the revenue half and a complete non-answer on the margin half, in the same response. That asymmetry is the through-line of the call: management will size the opportunity to the dollar and will not size the profitability of it. The gross margin question was re-asked and deferred again to a later update.
Memory Cost Exposure From the HBM-Heavy Design
The most technically substantive challenge concerned whether AMD's architectural choice to carry materially more high-bandwidth memory than the competition, historically a performance advantage, becomes a cost liability in an inflating memory market. Management asserted allocation visibility through 2027 and, more interestingly, volunteered that memory configuration is a variable AMD can adjust downward where the total-cost benefit does not justify it.
Q: "When we look at the specs of Helios, it has, I think, almost 50% more HBM than the competition. And I think you have always had more HBM in your products. So how much of the benefit is because of having 50% more HBM and then doesn't that expose AMD more to memory cost inflation."
— Vivek Arya, Bank of America Securities
A: "We do recognize in this memory environment though, every customer is looking at how to optimize their memory footprint. And we have the opportunity, if desired to also modify that memory footprint if the total cost of ownership is not as significant in certain workloads."
— Lisa Su, Chair and CEO
Assessment: The most quietly important answer of the call. Management just told the market that the memory advantage is a dial rather than a fixed specification, which protects gross margin in a cost squeeze at the price of some of the performance differentiation. That is the right commercial trade and it is a partial retreat from a pillar of the product argument. Worth watching whether shipped configurations drift downward through the ramp.
Whether Server Supply Is Still the Binding Constraint
Questioning returned repeatedly to whether the raised second-half server outlook reflects better demand or merely better supply. Management was unusually candid that the first-half tightness was a forecasting failure rather than a manufacturing one, and that 2027 supply should be materially easier because the demand is now visible far enough in advance to plan against.
Q: "In the server CPU business, are you supply constrained now? It seems like the market is very tight. And do you anticipate that tightness persisting? And when you sort of think about next year, just the general ability of the supply chain to support the level of growth that you're talking about specifically in CPU?"
— Joseph Moore, Morgan Stanley
A: "I would say the server CPU supply chain is tight right now, and it has been tight for the first half of the year because much of this demand was unforecasted. As we get into 2027, the demand is better forecasted. And so we would expect that the 2027 server supply situation should be better than '26."
— Lisa Su, Chair and CEO
Assessment: This reframes the second-half raise. Part of the more-than-80% server growth is demand that already existed in the first half and could not be shipped. That makes the number more reliable, not less, because it is backed by an order book rather than a forecast. It also implies first-half revenue understated true demand, which is the cleanest positive in the quarter.
Whether the Helios "Ahead of Forecast" Comment Was Volume or Yield
Management's prepared-remarks claim that customer pull for Helios was tracking ahead of internal forecasts drew a request to separate demand from manufacturing readiness, and a companion question on whether early-ramp gross margins should be expected to improve. The answer confirmed volume as the driver and set expectations for the usual early-ramp yield curve on a complex product.
Q: "And Lisa, in your prepared remarks, you mentioned that Helios was ahead of your original forecast. Can we unpack that comment a little bit? Was that a comment on volumes? Was it yields? And then if it yields, how should we think about sort of the first quarter or 2 of Helios gross margins compared to as it gets later into its ramp, should we expect gross margins to improve as it ramps?"
— Joshua Buchalter, TD Cowen
A: "So Josh, when I was talking about Helios was ahead of our initial forecast, it was as it relates to the overall volumes. ... One would expect that the overall yields will improve as we go through the next few quarters, the starting quarter is this quarter here in Q3, and we will be ramping over the next couple of quarters and we always would expect that the yields will continue to improve as we go through the first few quarters, especially on a product like this, which is highly complex."
— Lisa Su, Chair and CEO
Assessment: Volume ahead of plan is the better of the two possible answers. The yield commentary is honest and correctly sets up Q3 and Q4 as the trough of Helios profitability, which is consistent with the flat gross margin guide. It also means the mix drag on gross margin is at its worst precisely when the revenue ramp is steepest.
Whether OpEx Discipline Survives the Investment Cycle
The final question of the call went to the operating-expense trajectory against the roughly 40% market growth management had described. This was the same line of questioning that produced the most skeptical exchange of the Q1 call, and the answer this time was a cleaner commitment.
Q: "And one for Jean. How should we think about OpEx growth relative to that overall 40% market growth that you talked about in the next few years?"
— Atif Malik, Citi
A: "From an OpEx perspective, we'll continue to invest given the large opportunities we have ahead of us. But you should expect us to manage this OpEx increase less than the top line revenue growth. That's what our business model is designed for so we can drive in more operating leverage to deliver earnings per share."
— Jean Hu, CFO
Assessment: An explicit commitment where last quarter there was a rationale, and the quarter backs it: OpEx grew 8% sequentially against 13% revenue growth, and the Q3 guide holds that relationship. With gross margin now flat, this constraint is the only source of margin expansion left in the model, which makes it the most important promise management made on the call.
What They're NOT Saying
- The capex step-up went unmentioned and unasked. Capital expenditures more than doubled sequentially to $808M and nearly tripled year over year. It was not in prepared remarks, no forward capex range was offered, and no analyst raised it on the call. It is the largest unexplained change in the financials and the single biggest input to whether the free-cash-flow decline is a quarter or a regime.
- No 2027 gross margin range, asked twice. Both the opening and a later question sought a 2027 margin framework and both received a puts-and-takes answer with a promise of color "when we get there." Management will size 2027 revenue to the percentage point and will not bracket 2027 margin at all.
- Zero buybacks in the quarter. AMD repurchased no stock in Q2 after $221M in Q1 and $1,227M in the first half of last year. With $13.1B of cash and short-term investments, this is a choice, and it aligns with the capex ramp. Management offered no capital-allocation commentary.
- No Helios or Instinct revenue quantification for 2026. "Initial shipments" in Q3 and a Q4 ramp remain unsized. The Street is left backing into it from segment guidance, which is why one caller's $30B 2027 estimate was the only number on the table and management would only say it was too low.
- Client and Gaming's six-point operating margin decline got one clause. Segment profit fell from $767M to $582M on higher revenue and the explanation was go-to-market investment and roadmap breadth. No analyst followed up. If that compression is competitive rather than discretionary, it matters more than the segment's growth rate.
- No China or export-control commentary of any kind. A year after an $800M MI308 charge, there was no update on China policy, residual inventory, or any licensing path. Most likely genuinely immaterial; still an unaddressed tail risk in a business now built on export-sensitive accelerators.
- The MEXT acquisition was one bullet in the release and absent from the call. AMD acquired an AI predictive-memory technology company and spent $129M on acquisitions in the quarter. Given how central memory economics were to the Q&A, the silence is odd.
- The $483M investment gain was never discussed. It is a fifth of GAAP net income and the reason GAAP EPS looks as strong as it does. Correctly excluded from non-GAAP, and worth a sentence it did not get.
- No FY26 revenue guide. Structural rather than signaling, as always, but the gap between the implied full-year walk near $49–51B and published consensus is wide enough that the silence has consequences.
Market Reaction
- Pre-print setup: Closed at $518.58 on August 4, up 142.1% year to date and 193.3% over the trailing twelve months, against the S&P 500's 13.0% year-to-date gain. Critically, the stock had rallied 20.7% in the four sessions immediately before the print, from $429.56 on July 29. Trailing thirty days was flat at +0.1%, masking a 22% drawdown and a full retracement inside the month. The 52-week closing range was $151.14 to $580.91.
- After-hours move: Down 8% to 9% on the release and through the call, with reported figures clustering between "more than 8%" and 8.8%.
- Next-day session: Opened at $484.50, a 6.6% gap down, traded $478.20 to $502.00, and closed at $482.05, down 7.0% or $36.53. Volume of 49.6M was 1.7x the 30-day average. The S&P 500 was down 0.2% that session.
- Since: $489.28 on August 6 and $483.36 on August 7. The move has held. The stock now sits 16.8% below its 52-week closing high.
Three things explain a 7% decline on a double beat and a raised guide, and they compound rather than compete. The first is the setup: a 20.7% four-session rally into a print prices a blowout, and this was a 2.2% revenue beat. The second is the whisper gap: the guide beat the published $12.5B by 4% and missed an approximately $14B buy-side expectation by roughly 7%, and in a stock this crowded the unpublished number is the one that clears. The third is composition: flat gross margin guidance and a free-cash-flow margin that halved shifted the conversation from how fast AMD grows to how much of that growth reaches cash.
This is the fourth consecutive AMD beat-and-fade, and we think the pattern has stopped being a curiosity and started being information. Three quarters ago we read the fade as positioning unwind on an otherwise clean print, and that was right at the time. Four in a row is something else: it says the marginal buyer now requires a result well beyond the guide to be paid, which is the definition of a stock whose expectations have caught up with its execution. The sell side responded by raising targets and estimates, which confirms the disappointment was seated entirely on the buy side.
Street Perspective
Debate: Is a $220B Server CPU TAM Credible After Three Revisions in Nine Months?
Bull view: Each prior revision proved conservative within two quarters, and the current one is corroborated by observable results rather than extrapolation: cloud and enterprise server revenue each grew more than 70%, growth was more unit-led than price-led, and management raised the second-half outlook specifically because supply was added to meet demand that already existed. Forecasts that keep being beaten are usually too low, not too high.
Bear view: A 2030 TAM that has moved from roughly $60B to $120B to $220B in nine months is tracking the near-term demand signal, not measuring a structural end state, and that is exactly what a forecast does near a cycle peak. The agentic-AI mechanism underpinning the largest bucket is the least proven part of the market and is being sized off customer plans rather than deployed workloads.
Our take: The bull view is better evidenced on the near term and the bear view is better calibrated on the far term, and both can be right. Server demand through 2027 is backed by an order book and improved supply, so we underwrite it. The 2030 figure is a directional statement about a market nobody can size yet, and we decline to pay for it today. The important distinction is that the 2027 number is a commitment and the 2030 number is an aspiration; the market is increasingly pricing them the same way.
Debate: Does the Cash Flow Deterioration Signal a Regime Change or a Quarter?
Bull view: Every driver is growth-related and reversible. Receivables build when revenue accelerates into quarter end, inventory builds ahead of a rack-scale ramp with committed customers behind it, and capex rises when you are adding capacity into a supply constraint you have publicly said is limiting revenue. Operating cash flow of $2,366M is still up 62% year over year, and $13.1B of cash comfortably funds the build.
Bear view: Free-cash-flow margin fell from 25% to 14% in a single quarter on accelerating revenue, and the operating line was held up by a $2,274M payables increase that cannot repeat at that scale. Rack-scale systems carry structurally more working capital and more fixed assets than merchant silicon, so the conversion rate that supported AMD's multiple is not coming back. Management neither guided capex nor was asked about it.
Our take: The bear view is closer to right, though for a less dramatic reason. This is not deterioration, it is a change in business model that has been visible in the strategy for two years and has now arrived in the financials. Moving from selling chips to selling racks means financing inventory, carrying customer-specific configurations, and owning more test and packaging capacity. That is a good trade for revenue and a worse one for cash conversion. We model mid-teens free-cash-flow margins through the Helios ramp rather than a return to the mid-twenties, and we think a business converting 15% of revenue to cash deserves a different multiple than one converting 25%.
Debate: Has the Stock Outrun a Thesis That Is Objectively Improving?
Bull view: The 2027 outlook improved more this quarter than the stock did. Data Center guided to more than double, a fourth frontier-scale anchor customer signed, supply constraints lifting, software reaching production readiness, and the long-term EPS anchor raised. On our own 2027 estimate the stock trades at roughly 36x, which is not demanding for a business compounding revenue above 50% with a widening competitive position.
Bear view: Up 142% year to date and 193% over twelve months into the print, at roughly 66x this year's earnings, on a quarter where the beat margins halved, gross margin expansion stopped, cash conversion halved and the guide missed the buy-side bar. Four consecutive fades say the price already embeds the good version of 2027, and a 2030 EPS anchor inside an undefined strategic timeframe is being discounted as though it were a 2027 number.
Our take: We are moving to the bear side of this one, on price rather than on thesis. Our fair-value work supports $400 to $500 and the stock closed at $483.36, inside the upper half of that band, which leaves single-digit upside to the top and roughly 7% downside to the midpoint. That is not a distribution that beats the index. Everything we admire about this company is now in the number, and the two things that could re-rate it higher, a faster Helios ramp and a gross margin surprise, are exactly the two things management declined to quantify. We would rather own this at $400 than at $483, and we expect to get the chance.
Model Update Needed
| Item | Prior Model | Suggested Change | Reason |
|---|---|---|---|
| FY26 Revenue | $46–48B | $49–51B (mid ~$50B) | H1 actual $21.8B plus $13.0B Q3 guide plus a Q4 near $15B on the stated Q4-above-Q3 Data Center step |
| FY26 Data Center Revenue | not separately modeled | ~$31B | $5.8B + $6.7B actual, plus ~$8.1B Q3 and ~$10.2B Q4 implied by the guided segment descriptions |
| FY26 Non-GAAP Gross Margin | 56% ± 100bps | 55.5–56% (hold, bias lower) | Q3 guided flat at 56%; Helios dilution is worst in Q3–Q4 while yields climb |
| FY26 Non-GAAP EPS | $7–9 | ~$7.30 | $1.37 + $1.66 actual, ~$1.93 Q3 from guided components, ~$2.34 Q4; lands at the low end of the prior range |
| FY26 Free Cash Flow Margin | ~25% | 15–18% | Q2 printed 14%; capex more than doubled sequentially with no guided run rate |
| FY26 Capex | ~$1.6B | $3.0–3.5B | H1 actual $1,197M with Q2 alone at $808M and the ramp still ahead |
| FY27 Revenue | $60–65B | $78–85B (mid ~$81B) | Data Center more than doubling off ~$31B, plus client near $14B and embedded near $5B |
| FY27 Non-GAAP EPS | ~$11–12 | ~$13.00–13.75 | Revenue at ~$81B, gross margin ~55%, OpEx ~$18.5B, 13% tax, ~1.72B shares |
| FY28 Revenue | $70–78B | $100–110B | Anthropic's first gigawatt lands in 2027 and scales into 2028; MI500 and Verano on the annual cadence |
| Long-term EPS power | $20+ | Significantly above $20, timing undefined | Management raised the anchor; the strategic timeframe remains unspecified and we decline to date it |
Valuation impact. Our prior fair-value range of $325 to $400 was set on FY27 EPS of $11 to $12 at 28x to 33x. Both inputs move up. We now estimate FY27 non-GAAP EPS in a $13.00 to $13.75 band centred near $13.30, and we raise the multiple range to 30x to 37x to reflect a faster growth rate offset by structurally lower cash conversion. That produces a fair value of $400 to $500, a midpoint near $450.
At the August 7 close of $483.36, AMD trades at roughly 66x our FY26 estimate and 36x our FY27 estimate, with a market capitalization near $789B. That is 3.4% below the top of our range and 7.4% above the midpoint. The upside case requires either a Helios ramp materially faster than guided or a gross margin surprise, and management specifically declined to quantify either.
Thesis Scorecard Post-Earnings
| Thesis Point | Status | Notes |
|---|---|---|
| Bull #1: Server CPU share gains continue with Turin and Venice | Confirmed + | Fifth consecutive record server CPU quarter; cloud and enterprise each up more than 70%; Venice in production with demand described as stronger than any prior EPYC generation; supply constraint lifting into 2027. Status tag holds ON TRACK. |
| Bull #2: Instinct establishes sustainable hyperscale anchor customers | Confirmed + | Anthropic added at up to 2 GW alongside Meta at 6 GW, OpenAI, and Microsoft deploying Helios at scale on Azure. Three named anchors plus Microsoft at scale on Azure. Status tag holds ON TRACK. |
| Bull #3: Operating leverage holds as topline accelerates | Confirmed, with a caveat | Operating margin expanded to 27% and OpEx grew slower than revenue for the first time in several quarters. But free cash flow margin halved to 14% and capex more than doubled. Income-statement leverage confirmed; cash leverage broke. Status tag moves ON TRACK to AT RISK. |
| Bull #4: Agentic AI structurally expands the server CPU TAM | Confirmed | TAM raised again to approximately $220B by 2030 at more than 50% annual growth, with agentic workloads named as the largest and fastest-growing bucket. Corroborated by actual server results. Status tag holds ON TRACK, with the caveat that the estimate has now moved three times. |
| Bear #1: AI gross margins compress materially as Instinct ramps | Emerging | Q3 gross margin guided flat at 56% into the Helios ramp; data center AI margin described as slightly below corporate average; 2027 margin framework declined twice. Status tag moves CONTAINED to EMERGING. |
| Bear #2: Memory pricing creates a second-half consumer demand drag | Materializing | Management planning for a softer second-half PC market; gaming guided to a strong double-digit sequential decline; gaming graphics already down on component-driven price increases. Sized and contained, but landing as forecast. Status tag moves EMERGING to MATERIALIZING. |
| Bear #3: China policy risk on AI accelerator exports | Dormant | No mention on the call, no residual MI308 commentary, no licensing update. Unaddressed rather than resolved. Status tag holds CONTAINED. |
| Bear #4: The stock has run ahead of fundamentals | Materializing | Fourth consecutive beat-and-fade; +142% year to date into the print; roughly 66x our FY26 estimate; price now inside the upper half of our raised fair-value band. This is the pillar that drives the rating change. Status tag moves EMERGING to MATERIALIZING. |
| Bear #5 (NEW): Cash conversion structurally lower as AMD sells racks rather than chips | New — Emerging | Free cash flow margin 14% versus 25%; capex more than doubled sequentially with no guided run rate; inventory at $8.5B; the $2,274M payables increase nearly equaled the entire $2,366M of operating cash flow. Neither guided nor questioned on the call. Status tag opens at EMERGING. |
Overall: The operating thesis is stronger than it was a quarter ago and the investment thesis is weaker. All four bull pillars remain intact and two of them improved materially. Three of five bear points escalated a status tag in the same quarter, and a fifth was opened. That combination, an improving business against a deteriorating risk/reward, is precisely the configuration a rating framework based on relative twelve-month return is built to catch.
Action: Downgrading to Hold from Outperform; raising fair value to $400 to $500 from $325 to $400. This is a valuation call, not a thesis call. We are not sellers of the franchise and holders with a multi-year horizon have no reason to act on a rating built for a twelve-month window. 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. We would move to Underperform only if cash conversion stays in the low teens into 2027 while the multiple holds.