Snapshot: Executive Summary
AMD is the only credible #2 in AI accelerators. It is the sole vendor outside Nvidia shipping a rack-scale, merchant data-center GPU at volume, and it pairs that with the share-gaining EPYC server-CPU franchise that already funds the company. The most recent print, Q1 FY26 (quarter ended Mar 28 2026), put revenue at $10.25B (+38% YoY)1, led by a Data Center segment of $5.78B (+57% YoY, ~$1.6B operating income)2 off the Instinct ramp. The October 2025 OpenAI agreement, a multi-year commitment to deploy 6 gigawatts of AMD Instinct GPUs, beginning with 1GW of MI450 in 2H 20263, and the February 2026 Meta agreement (another ~6GW of custom MI450-architecture Instinct + EPYC via the Helios rack platform) re-anchored the data-center narrative. The problem is price: AMD has run +264% over the trailing year to ~$522, and that re-rating now prices a successful MI450 ramp in.
Three franchises, three valuation regimes:
- Data Center (~$22B FY25 → ~$30B FY26E): Instinct GPU (MI350/MI355X in production, MI450/Helios rack in 2H 2026) plus EPYC "Turin"/"Venice" server CPUs at record ~46% server-CPU revenue share against Intel. The AI engine and the adjustable premium multiple in the SOTP.
- Client + Gaming (~$14B FY25E): Ryzen desktop/notebook CPUs taking client share, plus semi-custom console revenue. Profitable, cyclical, a market-multiple base.
- Embedded (~$3.5B FY25): the Xilinx FPGA/adaptive-SoC franchise. High-margin, recurring-ish, recovering off the 2024–25 industrial inventory trough.
At ~71x forward earnings (FY26E non-GAAP EPS $7.38) and ~40x FY27E, AMD trades at a steep premium, and after a +264% year the SOTP base needs the Data Center franchise to earn a rich ~40x forward EBITDA multiple just to justify spot. We move the rating from Buy to Hold, 12-month PT $560 via SOTP (bull $700 / bear $320). The franchise is exceptional and the debate is no longer whether it is real; it is whether a fully-priced ~40x DC multiple holds against Nvidia's ~86–90% accelerator-revenue dominance, ~20% eventual warrant dilution, and back-end-loaded H2 2026+ execution.
Tactical: AMD is trading at ~$522, roughly at our $560 12-month target, the AI re-rating has run the stock to fair value. The SOTP base (premium Data Center multiple on the Instinct + EPYC AI engine + a market multiple on Client/Gaming and Embedded, with a net-cash add-back) reconciles here. Rating Hold.
Investment Thesis
Bull Case
- MI450/Helios rack is competitive at the system level; OpenAI + Meta 6GW each ramp on schedule plus a third hyperscaler whale lands
- Data Center revenue reaches ~$56B FY27 (Instinct DC-GPU well above EPYC + continued server share gains)
- ROCm closes enough of the CUDA gap to win inference at scale
- Data Center segment holds a ~41.6x EV/EBITDA multiple
- FY27 non-GAAP EPS pushes above the ~$13.11 consensus
Base Case
- Instinct DC-GPU revenue scales further; OpenAI MI450 1GW deploys in 2H 2026; MI350/MI355X ramp through the year
- EPYC server share holds the record ~46% revenue share against Intel
- Client recovers modestly; Gaming declines on the console tail; Embedded recovers off the inventory trough
- Data Center ~40.5x, Client+Gaming 10x, Embedded 16x EV/EBITDA
Bear Case
- Nvidia's annual cadence (Rubin) plus CUDA lock-in cap AMD DC-GPU share in the low-teens; NVDA keeps ~86–90% of accelerator revenue
- OpenAI/Meta deals slip or downsize; MI450 system-level execution disappoints vs Vera Rubin racks; ~20% warrant dilution bites
- Hyperscalers favour in-house ASICs (TPU, Trainium, MTIA, Meta is both an AMD customer and building MTIA)
- Data Center multiple compresses toward ~29x on ~$40B FY27 revenue
Rating: Hold. The probability-weighted SOTP (45% base / 25% bull / 30% bear) yields a blended fair value of ~$523, essentially on top of the ~$522 snapshot. We frame the central case at the SOTP base ($560) to reflect our view that AMD is a structural, multi-year share-taker in data-center GPU, but after a +264% trailing-year run the market already pays for that, leaving the risk/reward balanced rather than skewed. The net-cash balance sheet still caps downside relative to a levered peer.
Business Overview
AMD reports four segments, Data Center (~$22B FY25), Client (~$11.5B), Gaming (~$3B), and Embedded (~$3.5B). For valuation, the more useful frame collapses these into three buckets, because the Data Center AI engine trades at a radically different multiple than the PC/console base and the Xilinx cash franchise:
FY25 revenue mix: four segments
Data Center is now the largest segment and the fastest grower; Client + Gaming + Embedded together form the profitable, cyclical base.
Revenue by product line
Within Data Center, Instinct DC-GPU and EPYC server CPU are split out; Client is Ryzen, Gaming is semi-custom consoles, Embedded is Xilinx.
Segment commentary
- Data Center: Two engines: Instinct data-center GPU (MI350/MI355X in production, MI450/Helios rack-scale system in 2H 2026, MI500 in 2027) and EPYC server CPUs ("Turin" gen-5 shipping, "Venice" gen-6 on TSMC N2 next). The ZT Systems acquisition added rack-scale system design; the OpenAI 6GW (Oct 2025) and Meta ~6GW (Feb 2026) commitments anchor multi-year GPU demand. Q1 FY26 segment revenue was $5.78B (+57% YoY).
- Client: Ryzen desktop and mobile CPUs (~$2.9B in Q1 FY26). AMD continues to take client CPU share from Intel in both desktop and premium notebook, aided by Zen 5 and AI-PC "Ryzen AI" parts.
- Gaming: Semi-custom SoCs for the PlayStation 5 and Xbox plus Radeon discrete GPUs (~$0.72B in Q1 FY26). Structurally declining as the current console cycle ages; a profitable but shrinking tail.
- Embedded: The Xilinx FPGA and adaptive-SoC franchise (~$0.87B in Q1 FY26, +6% YoY). High gross margin, diversified across industrial, comms, aero/defense, and automotive; recovering off the 2024–25 inventory-correction trough.
Data Center: Instinct + EPYC
The single most important section in this report. AMD's Data Center segment is two businesses fused: the Instinct accelerator line that makes AMD the credible #2 in AI compute, and the EPYC server-CPU franchise that already out-earns the rest of the company. Q1 FY26 Data Center revenue was $5.78B, up 57% YoY on ~$1.6B of operating income2, and back-to-back gigawatt-scale commitments, the OpenAI 6GW agreement (signed October 2025, first 1GW of MI450 in 2H 2026)3 and the Meta ~6GW agreement (February 2026, custom MI450-architecture Instinct + EPYC on the Helios rack), give the FY26–FY27 ramp two named, multi-year demand anchors. Both carry an analogous warrant for up to ~160M AMD shares (~10% each), so the demand is real but the eventual dilution is too.
Instinct DC-GPU revenue trajectory
FY24 ~$5B → FY25 ~$9B actual; FY26E ramping further as Q1 FY26 Data Center prints $5.78B (+57% YoY) on the MI350/MI355X ramp, the OpenAI MI450 1GW deployment, and a widening hyperscaler customer base. FY27E reflects MI450/Helios scaling.
Data Center GPU: illustrative customer share
Estimated split of Instinct demand across named AMD data-center customers. OpenAI, Microsoft, Meta, and Oracle are public Instinct adopters; weights are author estimates.
What the Instinct franchise actually is
Instinct is a merchant data-center GPU, AMD sells the same part to many customers, unlike a single-customer custom ASIC. The MI350 series and MI355X (now in production) compete on memory capacity and bandwidth (large HBM stacks favour inference and large-context workloads) and close the compute gap; and MI450 with the Helios rack-scale system (built on the ZT Systems engineering AMD acquired), shipping H2 2026 and ramping into 2027 ahead of MI500 in 2027, is the first AMD product designed to compete with Nvidia at the rack level rather than the chip level. The hard part is not silicon, it is software. ROCm 7, AMD's CUDA alternative, now ships with native Windows support and day-0 PyTorch and has closed materially on inference, but still trails on the breadth of training tooling and the ecosystem lock-in CUDA enjoys. That gap, plus Nvidia's ~86–90% hold on accelerator revenue, is the central bear argument and the key thing to monitor.
EPYC: the quieter compounder
EPYC server CPUs are the part of Data Center that already prints cash. AMD has taken server-CPU revenue share against Intel for years on a per-core performance and total-cost-of-ownership advantage, reaching a record ~46% server-CPU revenue share (~33% units), with server CPU revenue up >50% YoY in a fourth straight record quarter; "Turin" (Zen 5) is shipping and "Venice" (Zen 6, TSMC N2) is next. AMD raised its server TAM to >$120B by 2030. EPYC is a higher-margin, less-contested business than merchant GPU, and it is the franchise that funds the Instinct ramp. We estimate EPYC server is the larger share of Data Center segment profit today even as Instinct overtakes it on revenue.
The capex backdrop
Top-5 hyperscaler capex ($B)5
Microsoft + Google + Meta + Amazon + Oracle combined AI-infrastructure spend is scaling toward ~$500B by 2027E. AMD's exposure is differential, it benefits from the merchant-GPU share of that spend, which it must win from Nvidia. The OpenAI 6GW commitment and growing Microsoft/Meta/Oracle Instinct deployments are the proof points; Nvidia's annual cadence and in-house ASICs are the substitution risks.
Client / Gaming / Embedded: The Profitable Base
Outside Data Center, AMD runs three businesses that together form a profitable, self-funding base. They will not re-rate the stock, but they de-risk it: the Client + Gaming bucket throws off cash through the cycle, and Embedded is a high-margin franchise recovering off a trough. In the SOTP these map to the fixed-multiple "Client + Gaming" bucket and the adjustable "Embedded" bucket.
Client: Ryzen share gains
- Desktop and mobile CPUs (~$2.9B in Q1 FY26, ~$11.5B FY26E): AMD continues to take PC CPU share from Intel. Zen 5 desktop parts hold the enthusiast/gaming segment; "Ryzen AI" notebook silicon rides the AI-PC refresh.
- Client is cyclical with the PC market but structurally a share-gain story; the marginal unit is increasingly an AMD unit.
Gaming: the console tail
- Semi-custom + Radeon (~$0.72B in Q1 FY26): PlayStation 5 and Xbox semi-custom SoCs plus discrete Radeon GPUs. Structurally declining as the console generation ages; a profitable but shrinking line we model down.
Embedded: the Xilinx franchise
- FPGA + adaptive SoC (~$0.87B in Q1 FY26, +6% YoY): the Xilinx acquisition (closed 2022). High gross margin, diversified across industrial, communications, aerospace/defense, automotive, and test. Recovering off the 2024–25 inventory-correction trough; long product lifecycles make this the most annuity-like franchise in the company.
The risk lens
The base case here is not heroic. Client recovers modestly with the PC cycle, Gaming declines on the console tail, and Embedded recovers off the trough. The downside is a stalled Embedded recovery (industrial demand stays soft) or a sharper-than-expected console decline, neither of which threatens the thesis, because the thesis lives in Data Center. This bucket's job is to be the dependable, market-multiple ballast under the Instinct optionality.
Financial Health & Trends
Revenue by quarter ($B)
Non-GAAP gross margin (%)
FY21-FY26E: the Data Center arc
AMD's revenue line over the last five years is the story of a CPU-share-gainer becoming an AI-compute company. The Xilinx close (2022) added the Embedded franchise; the Instinct ramp (FY24 onward) is the step-change that re-rated the stock. Gross margin has expanded as Data Center and Embedded mix has grown, partially offset by the lower-margin merchant-GPU ramp and the console tail.
FY21-FY26E: revenue & non-GAAP GM
FY22 step: Xilinx consolidated. FY24–FY26E: the Instinct data-center-GPU ramp drives revenue. GM rises with Data Center + Embedded mix; the merchant-GPU ramp is a partial offset as Instinct gross margin is below the corporate average early in the curve.
Q1 FY26 print highlights
| Metric | Q1 FY26 | YoY | Driver | vs cons |
|---|---|---|---|---|
| Total revenue | $10.25B | +38% | Data Center-led | beat |
| Data Center revenue | $5.78B | +57% | Instinct + EPYC; ~$1.6B op inc | beat |
| Client + Gaming revenue | $3.6B | +~mid-teens | Client $2.9B + Gaming $0.72B | in-line |
| Embedded revenue | $0.87B | +6% | Xilinx off trough | in-line |
| Non-GAAP EPS | $1.37 | +~40% | DC operating leverage | beat |
| Non-GAAP gross margin | 55% | +~100bps | mix + Instinct yields | in-line |
| Q2 FY26 guide | ~$11.2B | +46% | ~56% non-GAAP GM | — |
Note: Q1 FY26 ended Mar 28 2026 (reported May 2026); free cash flow was ~$2.6B, operating margin 25%. FY26/FY27 estimates (revenue ~$49.4B/~$67B; non-GAAP EPS ~$7.38/~$13.11) are consensus. Instinct gross margin sits below the corporate average early in the ramp; blended GM improves as EPYC and Embedded mix and Instinct yield curves mature.
Capital Allocation & Returns
AMD's capital story is the inverse of a levered compounder: it carries a net-cash balance sheet (cash & investments exceed debt), pays no dividend, and reinvests aggressively in R&D, the Instinct roadmap, ROCm software, and EPYC. After cash, the priority is opportunistic buybacks under a multi-billion-dollar board authorization, plus strategic M&A (ZT Systems for rack-scale system design; earlier, Xilinx and Pensando).
Why the net-cash sheet matters
In the SOTP, AMD's net cash is a per-share add-back, equity value equals total enterprise value plus the net cash, not minus net debt. That structurally lifts the implied PT relative to a levered peer and limits downside: even in the bear case the balance sheet is not a constraint on funding the roadmap. The trade-off is the absence of a dividend, so the entire return case rests on Data Center execution and the multiple it earns.
The M&A question
AMD's M&A has been platform-building, not financial-engineering: Xilinx (Embedded), Pensando (DPU/networking), and ZT Systems (rack-scale system design for Helios). The most likely next moves are software/ecosystem tuck-ins that narrow the ROCm-versus-CUDA gap, the single highest-leverage investment the company can make.
Valuation Overview
AMD trades at ~71x fwd P/E (FY26E non-GAAP EPS $7.38) and ~40x FY27E (~$13.11) vs the SOX in the high-20s. Read at face value that looks very rich. But the headline multiple obscures the three-bucket structure, applying a single P/E to a business that is one part Data Center AI engine (deserves a premium multiple), one part cyclical PC/console base (deserves a market multiple), and one part high-margin Embedded franchise (deserves a software-like multiple) is the wrong frame. The right frame is SOTP, and the key tell is that, after a +264% year, the base case now needs the Data Center franchise to earn a rich ~40x forward EBITDA multiple just to reconcile to spot. We develop the full mechanics in the next section.
AMD forward P/E history
AMD's multiple has expanded with the Data Center re-rating. The 2024–2025 re-rate is fundamental, the earnings base changed as Instinct scaled, not just sentiment.
Peer NTM EV/EBITDA
AMD sits between Nvidia (AI pure-play, premium) and the diversified semis (Intel, Marvell, TSM). The closer the data-center-GPU share story tracks, the more AMD's multiple converges toward the AI-leader end of the range.
Why the SOTP
A blended P/E argument always loses to a fundamentals-aware SOTP when the underlying business mix is heterogeneous. We separate AMD into three buckets, Data Center, Client + Gaming, and Embedded, apply scenario-appropriate multiples to each, sum the enterprise values, add net cash, and divide by shares. That work is in the next section, with an interactive scenario tool.
Where the Street sits: the average published target (~$500) now sits below the ~$522 spot, even as the consensus rating stays Strong Buy. The freshest raises cluster $560–670 (BofA $560, Citi $575, TD Cowen / Bernstein $600, Mizuho $615, UBS $670). Our $560 base lands at the bottom of that fresh cluster and roughly at spot, which is precisely why we read AMD as fairly valued and rate it Hold rather than chasing the re-rating.
SOTP: Sum-of-the-Parts
The SOTP is the heart of the report. Each bucket is valued on its own FY27E revenue, EBITDA margin, and EV/EBITDA multiple. Toggle between Bull, Base, Bear, and Reverse-DCF scenario anchors; drag the Data Center and Embedded multiples directly to see the implied PT update. Note: AMD carries net cash, so the model adds the cash to enterprise value rather than subtracting net debt, the "net debt" line is negative.
| Bucket | Rev ($B) | EBITDA margin | Multiple | EV ($B) |
|---|---|---|---|---|
| Data Center (FY27E) | 48 | 43% | 40x | — |
| Client + Gaming (FY27E) | 14 | 16% | 10x | — |
| Embedded (FY27E) | 5.5 | 45% | 16x | — |
| Total enterprise value | — | |||
| Plus: net cash ($B) | +9 | |||
| Equity value ($B) | — | |||
| Implied per-share PT (÷ 1.62B sh) | — | |||
| vs current market | — |
Revenues and EBITDA margins update from the scenario anchors; multiples are user-editable via the sliders. Net debt is negative (AMD net cash), so equity value = total EV + cash. The "Reverse" scenario uses base revenues/margins, drag the multiples until the implied PT equals the current market price to see what the market is pricing.
Sensitivity grid: Data Center multiple × Embedded multiple (PT in $)
PT calculator (alternative simple-multiple view)
Risk / Reward calculator
Ask the Thesis AI-assisted checking…
Describe a scenario in natural language; the assistant returns a structured impact analysis against this dashboard's thesis, scenarios, and SOTP math. Powered by Claude via a Cloudflare Worker proxy (Anthropic key held server-side; same pattern as the live-quote feed).
Note: The assistant reasons from the dashboard's data snapshot and thesis sections, it does not browse the web, hit Bloomberg, or access real-time fundamentals beyond what's in data.js. Treat its responses as scenario-modeling support, not as primary research. Author judgments on rating, PT, and probabilities remain with the analyst.
Upcoming Catalysts
| Catalyst | Window | Why it matters |
|---|---|---|
| Q2 FY26 print | Aug 2026 | Tests the ~$11.2B (+46%) guide and ~56% non-GAAP GM; updated commentary on the OpenAI + Meta deployment cadence. |
| MI355X volume ramp | 2H 2026 | Evidence the compute gap to Nvidia is closing; hyperscaler adoption breadth beyond the initial anchor customers. |
| OpenAI MI450 1GW deployment | 2H 2026 | First gigawatt of the 6GW commitment goes live, the proof point for the multi-year Instinct demand anchor (and the warrant-vesting trigger). |
| MI450 / Helios rack launch | 2H 2026 | AMD's first rack-scale system competing with Nvidia at the system level, not the chip level; ramps into 2027 ahead of MI500. |
| EPYC "Venice" (Zen 6, N2) | 2026–27 | Next server-CPU generation; sustains the record ~46% server-CPU revenue-share annuity that funds the Instinct ramp. |
| New hyperscaler Instinct win | Any quarter | A third named whale beyond OpenAI/Meta/Microsoft/Oracle would be a substantial re-rate event. |
Risk Factors
- Valuation / the multiple itself. After a +264% trailing-year run to ~$522, AMD trades at ~71x FY26E EPS and the SOTP base needs a rich ~40x DC EBITDA multiple just to reconcile to spot. The average Street target (~$500) already sits below the price. A successful MI450 ramp is largely priced in, so the asymmetry that justified a Buy is gone, the core reason the rating moves to Hold.
- Nvidia dominance + CUDA lock-in. The pre-eminent fundamental risk. Nvidia still holds ~86–90% of AI-accelerator revenue; its annual product cadence (Rubin) and the entrenched ~20-year CUDA ecosystem make data-center-GPU share extraordinarily hard to take. If AMD's DC-GPU share stalls in the low-teens, the Data Center re-rating that anchors the PT compresses.
- ROCm-versus-CUDA software gap. AMD's silicon is competitive and ROCm 7 (native Windows, day-0 PyTorch) has closed materially on inference; the stack still trails on training tooling. If ROCm fails to close enough of the training-ecosystem gap, AMD wins only inference share and the revenue ramp disappoints relative to the bull case.
- OpenAI/Meta deals slip, downsize, or dilute. The two ~6GW commitments are the named demand anchors but are back-end-loaded to H2 2026+, and each carries a warrant for up to ~160M shares (~10%, ~20% combined). A delay to the 2H 2026 MI450 1GW deployment, a downsizing, or the dilution vesting at the share-price milestones (up to $600) cuts the per-share trajectory.
- Hyperscaler in-house ASICs. Google TPU, Amazon Trainium, and Meta MTIA are merchant-GPU substitutes for the largest buyers, and Meta is both an AMD customer and building MTIA. Broadcom + Marvell account for ~95% of custom-ASIC co-design. If the hyperscalers shift incremental AI compute to in-house silicon, the merchant-GPU market AMD is fighting for shrinks at the margin.
- Cyclical and console exposure. Client is sensitive to the PC cycle; Gaming is in structural decline on the console tail; Embedded depends on an industrial recovery that has been slow. A weak PC cycle or stalled Embedded recovery pressures the base bucket.
- Supply / Instinct gross-margin drag. Merchant data-center GPU carries below-corporate gross margin early in the ramp, and HBM4/CoWoS supply plus Helios readiness gate the volume. A faster Instinct mix shift without yield maturation compresses blended GM and the EBITDA margins the SOTP assumes.
Scenario Stress Tests
| Scenario | Mechanism | Anchor PT | Delta vs base $560 |
|---|---|---|---|
| Base | Mid-range DC-GPU share, EPYC gains, Embedded recovers | $560 | — |
| OpenAI/Meta deals slip a year | DC rev FY27 $42B (vs $48B base) | ~$495 | (12%) |
| ROCm fails to close on training | DC rev FY27 $42B, multiple compresses to 33x | ~$412 | (26%) |
| Embedded recovery stalls | Embedded FY27 rev $5B, 14x multiple | ~$555 | (1%) |
| Nvidia caps DC share in low-teens | DC rev FY27 $40B, multiple to 28.9x | ~$320 | (43%) |
| PC + console weakness | Client+Gaming FY27 $12B at 8x; DC unchanged | ~$556 | (1%) |
| Bull: MI450 system win + 3rd whale | DC rev FY27 $56B at 41.6x; Embedded $6B at 17x | ~$700 | +25% |
All stress-test PTs are derived from the same SOTP framework used in the interactive tool, each shock changes only the inputs noted. The base case is the central anchor we maintain.
Bull vs Bear Debate
| Issue | Bull view | Bear view |
|---|---|---|
| Can AMD take real DC-GPU share? | AMD is the only merchant GPU-2; memory-rich Instinct parts win inference, MI450/Helios competes at the rack, and OpenAI's + Meta's 6GW each are named, multi-year anchors. | Nvidia still holds ~86–90% of AI-accelerator revenue; its annual cadence + CUDA lock-in have repeatedly capped challengers. Low-teens share is the realistic ceiling; the bull case prices in share AMD has never demonstrated. |
| ROCm vs CUDA | ROCm has closed materially on inference; framework support is now good enough that switching cost is falling for the largest, most sophisticated buyers. | CUDA is a 15-year ecosystem moat. Training tooling, libraries, and developer mindshare still favour Nvidia by a wide margin; "good enough on inference" is not the whole market. |
| OpenAI + Meta 6GW durability | Two multi-year, gigawatt-scale commitments from the most demanding AI buyers validate the roadmap and de-risk FY26–FY27 demand. | Mega-deals slip and resize, and both carry ~10% warrants (~20% combined dilution). The first 1GW is 2H 2026; execution risk on MI450 and Helios at system scale is real, back-end-loaded, and unproven. Meta is also building its own MTIA. |
| Premium multiple justified? | ~71x fwd P/E (and a ~40x DC EBITDA multiple) is defensible for a structural data-center share-taker with a net-cash sheet and an EPYC annuity funding the GPU ramp. | After +264% the multiple already prices a successful MI450 ramp. If share stalls, the DC multiple compresses toward ~29x and the stock de-rates hard, the average Street target (~$500) already sits below spot. |
| Margins through the ramp | EPYC and Embedded carry high margins; as Instinct yields mature and mix normalizes, blended GM and EBITDA margin expand. | Merchant GPU GM is structurally below corporate; a fast Instinct mix shift drags blended margins and the SOTP's EBITDA assumptions look optimistic. |
Technical Analysis
AMD trailing-12-month monthly closes
RSI (multi-timeframe)
Daily 64 = neutral-to-bullish; Monthly 70 = trend intact off the 52-week trough but stretched after the +264% run.
MACD vs Signal
Crossed above signal and held through the OpenAI/Meta-deal rallies; trend constructive but extended into mid-2026.
Relative strength (trailing year)
AMD ran +264% over the trailing year, off the $133.50 low to the $562.99 all-time high on the OpenAI/Meta deals and the MI450/Helios ramp.
EMA stack (current)
Trader's view
- Price > 50-DMA > 200-DMA, bullish stack intact through the OpenAI/Meta-deal rallies, though price now sits well above the rising averages.
- Key support: the ~$450 area (prior consolidation shelf). A close below ~$450 invalidates the medium-term setup.
- Key resistance: the $562.99 all-time high. The stock trades ~$522, roughly at the $560 PT, so the risk/reward into resistance is balanced.
- Momentum: MACD constructive but extended; RSI elevated after the +264% run, limited room before it cools.
Glossary & Methodology Notes
- Instinct (MI300X / MI325X / MI350 / MI355X / MI450 / MI500)
- AMD's data-center GPU accelerator family. MI300X/MI325X are memory-rich parts favouring inference; the MI350 series and MI355X (in production) close the compute gap; MI450 with the Helios rack (H2 2026) is AMD's first rack-scale system, with MI500 following in 2027. The merchant alternative to Nvidia's data-center GPUs.
- Helios (rack)
- AMD's rack-scale AI system built around MI450-class GPUs, integrating compute, networking, and cooling at the rack level, designed to compete with Nvidia at the system level rather than the chip level. Engineered using the ZT Systems team AMD acquired; ships H2 2026 and ramps into 2027.
- ROCm
- AMD's open software stack for GPU compute, the CUDA alternative. The breadth and maturity of ROCm (frameworks, libraries, developer tooling) relative to CUDA is the central determinant of how much data-center-GPU share AMD can take.
- EPYC (Turin / Venice)
- AMD's server CPU line. "Turin" is the Zen 5 generation shipping today; "Venice" is the next Zen 6 generation on TSMC's N2 process. EPYC has taken server-CPU share from Intel for years and is the higher-margin franchise funding the Instinct ramp.
- SOTP (Sum-of-the-Parts)
- A valuation method that values each business segment separately, then sums the enterprise values, adjusts for net debt or net cash, and divides by shares to get a per-share fair value. Used when a single blended multiple obscures the underlying mix. AMD carries net cash, so the adjustment is an add.
- EV/EBITDA
- Enterprise value divided by EBITDA (earnings before interest, taxes, depreciation, amortization). A capital-structure-neutral valuation measure; appropriate for cross-comparing businesses with different balance sheets.
- Xilinx (Embedded)
- The FPGA and adaptive-SoC business AMD acquired in 2022, reported as the Embedded segment. High gross margin, diversified across industrial, comms, aero/defense, and automotive; the most annuity-like franchise in the company.
- Hyperscaler
- The largest cloud and AI infrastructure operators: Amazon (AWS), Microsoft (Azure), Google (GCP), Meta, and Oracle (OCI). Their AI-infrastructure capex and merchant-vs-in-house silicon choices drive AMD's data-center-GPU opportunity.
Methodology
- Snapshot anchor: June 26, 2026 close (~$521.58). Live price patches via the Cloudflare-Worker quote proxy on page load. Latest reported quarter is Q1 FY26 (ended Mar 28 2026).
- FY26–FY27 figures are consensus (revenue ~$49.4B/~$67B; non-GAAP EPS ~$7.38/~$13.11); segment splits are author triangulations of the reported results.
- SOTP scenarios use FY27E revenues; the EBITDA-margin and multiple anchors are the author's view, not consensus. The base reconciles to spot, which is why the rating is Hold.
- Conclusions are the author's view. Illustrative, not investment advice.
Sources & Citations
Inline citations
Superscripted numbers in the body link here. Click any N in the report to jump back to the source.
- Advanced Micro Devices, Inc., Q1 FY2026 results (quarter ended Mar 28 2026, reported May 2026): total revenue $10.25B (+38% YoY), non-GAAP EPS $1.37, non-GAAP gross margin 55%, operating margin 25%, ~$2.6B free cash flow; Q2 FY26 guide ~$11.2B (+46%) at ~56% non-GAAP GM. Segment splits are author triangulations of the reported results. ↩
- AMD Data Center segment disclosures: Q1 FY26 segment revenue $5.78B (+57% YoY) on ~$1.6B operating income, with Instinct DC-GPU plus EPYC at a record ~46% server-CPU revenue share. FY26/FY27 revenue (~$49.4B/~$67B) and non-GAAP EPS (~$7.38/~$13.11) are consensus. ↩ ↩
- AMD–OpenAI strategic agreement (announced October 2025): a multi-year commitment to deploy 6 gigawatts of AMD Instinct GPUs, first 1GW of MI450 targeted for 2H 2026, with a warrant for up to 160M AMD shares (~10%) at a $0.01 strike vesting on deployment + share-price milestones (up to $600); AMD frames >$100B of revenue over ~4 years. The February 2026 Meta agreement adds another ~6GW of custom MI450-architecture Instinct + EPYC via the Helios rack with an analogous up-to-160M-share warrant. ↩ ↩
- Combined hyperscaler capex aggregates from Microsoft, Alphabet, Meta, Amazon, and Oracle public filings and earnings releases. 2026E and 2027E figures are author estimates triangulating sell-side consensus, management qualitative guidance, and announced multi-year datacenter projects. AMD's exposure is the merchant-GPU share of this spend. ↩
Background reading
- Advanced Micro Devices, Inc. 10-K (FY24, FY25), annual financials, segment disclosures, M&A history (Xilinx, Pensando, ZT Systems).
- AMD 10-Q filings (FY25 through Q1 FY26), quarterly revenue, Data Center / Instinct disclosure, capital allocation.
- AMD Q1 FY26 earnings release + transcript (reported May 2026), $10.25B revenue, $5.78B Data Center, record ~46% server-CPU share, MI350/MI355X/MI450/Helios roadmap, ROCm 7 progress.
- AMD "Advancing AI" event materials (2024, 2025), Instinct and Helios roadmap, ROCm software updates.
- AMD–OpenAI announcement (October 2025) and AMD–Meta announcement (February 2026), the two ~6GW Instinct deployment commitments and warrant structures.
- Public press on Nvidia roadmap (Rubin), CUDA ecosystem, custom-ASIC co-design (Broadcom, Marvell), and hyperscaler in-house ASICs (TPU, Trainium, MTIA), competitive landscape.
Disclaimer. This report is the author's institutional equity-research view, prepared for portfolio and educational purposes. It is not a recommendation to buy, sell, or hold any security. Forward-looking statements are subject to risk and uncertainty; past performance is not indicative of future results. Consult a licensed financial advisor before making investment decisions. All third-party trademarks are the property of their respective owners.