Outcome reveal. Read only after Part A was frozen. This report binds the outcome-blind Part A bundle at
c9a4bfb2d28e9a61c209ce2becc82a0dac4e7b7dbeb90f57f01a035086e66ae1, frozen August 16, 2026 at 16:22:35 UTC.
Executive summary
-
The selected business outcome was favorable, but no investment outcome is calculated. AMD later reported 2017 Ryzen and EPYC launches, customer-authored Microsoft and Google adoption signals appeared, and the selected consolidated operating endpoints improved substantially by FY2021 and FY2025. No actual investor action is observed, and the record lacks the governed price, corporate-action, capitalization, counterfactual cash-flow, mandate, exposure, and risk-budget package needed for TSR, target price, price attractiveness, or position size. (claim.amd.outcome.actual-roadmap-execution; claim.amd.outcome.customer-authored-adoption; claim.amd.outcome.return-valuation-abstention)
-
FY2021 is the cleanest selected pre-Xilinx operating bridge. AMD reported FY2021 revenue of USD 16,434 million, operating income of USD 3,648 million, net income of USD 3,162 million, and operating cash flow of USD 3,521 million before Xilinx closed. Yet Enterprise, Embedded and Semi-Custom growth included both EPYC and semi-custom products, so it is not an EPYC-only series. (claim.amd.outcome.fy2021-pre-xilinx-scale; claim.amd.outcome.fy2021-mixed-segment-drivers; table.amd.outcome.pre-xilinx-bridge)
-
FY2025 is acquisition-expanded and multi-product. AMD reported continuing-operations revenue of USD 34,639 million and operating income of USD 3,694 million. Data Center revenue was USD 16,635 million and segment operating income USD 3,603 million, but the segment included CPUs, GPUs, AI accelerators, DPUs, AI NICs, FPGAs, APUs, and SoCs. Xilinx, Pensando, ZT, Instinct demand, discontinued operations, acquisition amortization, and export controls prevent an organic or Zen-only reading. (claim.amd.outcome.endpoint-continuing-results; claim.amd.outcome.endpoint-data-center-multi-product; claim.amd.outcome.acquisition-perimeter; claim.amd.outcome.zt-perimeter)
-
The Part A process remains defensible. At the May 2016 cutoff, independent technical validation, production conversion, stable product economics, complete manufacturing and funding schedules, and valuation readiness were missing. Later favorable evidence could clear gates prospectively; it cannot retroactively enter the cutoff record. (judgment.amd.outcome.process-quality)
What actually happened
No real investor trade is asserted. Part A asked a simulated long-only committee whether to add, maintain, defer, or escalate a possible reduction. It chose defer and re-underwrite, sized no position, and required human approval for any external action. The later public record instead shows company and customer events:
- Microsoft announced an Azure Lv2-Series configuration powered by AMD EPYC on December 5, 2017. This is customer-authored availability evidence, not audited AMD volume or contribution margin. (claim.amd.outcome.customer-authored-adoption; evidence.amd.outcome.microsoft-lv2-adoption)
- AMD's FY2017 filing reported first-generation Zen-based Ryzen desktop CPUs and EPYC datacenter processors launched during 2017. The same filing showed early consolidated profitability but a mixed Enterprise, Embedded and Semi-Custom segment in which EPYC and IP-related revenue were offset by other changes. (claim.amd.outcome.actual-roadmap-execution; claim.amd.outcome.fy2017-early-economics; evidence.amd.outcome.fy2017-zen-launch; evidence.amd.outcome.fy2017-epyc-mixed-segment)
- Google announced on August 8, 2019 that it was using EPYC for internal workloads and planned second-generation EPYC virtual machines for Google Cloud customers. Google's performance language remains a customer expectation rather than an independently reproduced benchmark. (claim.amd.outcome.customer-authored-adoption; evidence.amd.outcome.google-epyc-adoption; evidence.amd.outcome.google-performance-claim)
- AMD's FY2021 filing revealed much larger pre-Xilinx consolidated operating scale and said Enterprise, Embedded and Semi-Custom growth was driven by both semi-custom and EPYC sales. (claim.amd.outcome.fy2021-pre-xilinx-scale; claim.amd.outcome.fy2021-mixed-segment-drivers)
- AMD closed Xilinx in February 2022 and Pensando in May 2022, adding FPGA, adaptive-SoC, and DPU scope. The selected filing recorded USD 48,800 million of Xilinx consideration, USD 46,400 million net of cash acquired, 429 million AMD shares issued, and USD 1,655 million of Pensando purchase consideration. (claim.amd.outcome.acquisition-perimeter; table.amd.outcome.acquisition-perimeter)
- The original FY2025 filing supplied the common audited endpoint: continuing and discontinued operations, a multi-product Data Center segment, ZT acquisition and divestiture effects, acquisition-related amortization, export-control charges, balance-sheet and commitment coordinates, and a conditional OpenAI warrant maximum. (claim.amd.outcome.endpoint-continuing-results; claim.amd.outcome.endpoint-data-center-multi-product; claim.amd.outcome.export-control-charge; claim.amd.outcome.zt-perimeter; claim.amd.outcome.dilution-boundary; claim.amd.outcome.liquidity-and-commitments)
The selected record does not establish that AMD management or any real investor used Part A's six gates. It tests the quality of a decision framework, not compliance with that framework.
Selected operating endpoints
| Measure | FY2017 launch year | FY2021 pre-Xilinx | FY2025 acquisition-expanded continuing operations |
|---|---|---|---|
| Revenue (USDm) | 5,329 | 16,434 | 34,639 |
| Gross profit (USDm) | 1,823 | 7,929 | 17,152 |
| Operating income (USDm) | 204 | 3,648 | 3,694 |
| Net income measure (USDm) | 43 total | 3,162 total | 4,269 continuing |
| Operating cash flow measure (USDm) | 68 total | 3,521 total | 6,493 continuing |
| PPE purchases measure (USDm) | 113 total | 301 total | 974 continuing |
| Deterministic gross margin | 34.2090% | 48.2475% | 49.5164% |
| Deterministic operating margin | 3.8281% | 22.1979% | 10.6643% |
| Deterministic net margin | 0.8069% total | 19.2406% total | 12.3243% continuing |
| OCF less PPE purchases (USDm; not FCF) | (45) | 3,220 | 5,519 continuing |
Reported facts and deterministic checks are preserved in table.amd.outcome.fy2017-launch-year, table.amd.outcome.pre-xilinx-bridge, and table.amd.outcome.endpoint-continuing. The OCF-minus-PPE line is only a descriptive residual on each labeled perimeter. It is not free cash flow, maintenance capital, required reinvestment, or distributable cash. (model-run.amd.outcome-descriptive-checks)
The selected FY2021-to-FY2025 calculation shows reported revenue growth of 110.7764% and a 2.1078-times revenue scale ratio. Both endpoints are correct as reported, but the comparison crosses the Xilinx and Pensando acquisitions, changing product and segment definitions, and later ZT activity. It is explicitly cross-perimeter and not organic growth. (table.amd.outcome.selected-scale-bridge; claim.amd.outcome.acquisition-perimeter; claim.amd.outcome.attribution-bounded)
FY2017 proves launch and early presence, not EPYC economics
The FY2017 filing establishes AMD's own report that Zen-based Ryzen and EPYC launched. It also reports USD 5,329 million of consolidated revenue, USD 204 million of operating income, USD 43 million of net income, USD 68 million of operating cash flow, and USD 113 million of property, plant and equipment purchases. Those are company-wide facts, not Zen- or EPYC-level economics. (evidence.amd.outcome.fy2017-operations; evidence.amd.outcome.fy2017-cash-flow)
The Enterprise, Embedded and Semi-Custom segment was flat at about USD 2.3 billion of revenue in 2017 and operating income fell to USD 154 million from USD 283 million. AMD's narrative mixed EPYC sales and IP-related revenue with lower semi-custom sales, lower NRE revenue, datacenter expenses, and foundry-related costs. That record blocks the tempting inference that EPYC was already independently profitable or caused the consolidated turnaround. (evidence.amd.outcome.fy2017-epyc-mixed-segment; claim.amd.outcome.fy2017-early-economics)
FY2021 is favorable, but its segment driver remains mixed
The FY2021 bridge is valuable because Xilinx remained pending. AMD reported USD 16,434 million of revenue and USD 3,648 million of operating income, giving a deterministic operating margin of 22.1979%. Operating cash flow was USD 3,521 million and PPE purchases were USD 301 million; their USD 3,220 million difference is not FCF. (evidence.amd.outcome.fy2021-operations; evidence.amd.outcome.fy2021-cash-flow; table.amd.outcome.pre-xilinx-bridge)
Enterprise, Embedded and Semi-Custom reported USD 7,102 million of revenue and USD 1,979 million of operating income. AMD attributed the 113% revenue increase to both semi-custom products and EPYC processors. The evidence is consistent with EPYC contribution, but it does not isolate units, ASP, product gross margin, contribution, retention, or a counterfactual without EPYC. (evidence.amd.outcome.fy2021-segment-table; evidence.amd.outcome.fy2021-segment-mix; judgment.amd.outcome.causal-attribution)
Acquisitions changed the economic and share perimeters
Xilinx added adaptable hardware platforms, FPGAs, adaptive SoCs, and ACAP products; Pensando added DPUs and distributed-services capabilities. AMD issued 429 million shares to Xilinx stockholders. Compared only as a scale coordinate with 1,207 million FY2021 ending shares, that issuance equals about 35.5%. The inputs are reported in whole millions, so the model deliberately stops at one decimal. This is actual transaction issuance, not a complete dilution attribution or acquisition-return calculation. (evidence.amd.outcome.fy2022-product-perimeter; evidence.amd.outcome.fy2022-xilinx-consideration; table.amd.outcome.acquisition-perimeter)
FY2025 acquisition-related intangible amortization was USD 2,254 million. That cash-flow reconciliation line is a noncash period expense add-back, not current acquisition cash spending and not an acquisition IRR. The selected record lacks complete timed incremental cash flows, synergies, opportunity cost, terminal value, and a human-approved counterfactual. (evidence.amd.outcome.fy2025-acquisition-amortization; claim.amd.outcome.acquisition-perimeter; claim.amd.outcome.return-valuation-abstention)
The diluted weighted-average share count rose from 1,229 million in FY2021 to 1,636 million in FY2025, a one-decimal scale comparison of 33.1%. It is a share-count comparison across periods, not Xilinx-only dilution. Ending shares, weighted-average diluted shares, transaction shares, and conditional instruments remain different measures. (table.amd.outcome.selected-scale-bridge; claim.amd.outcome.dilution-boundary)
FY2025 continuing, discontinued, and Data Center perimeters
The original FY2025 filing reported continuing-operations net income of USD 4,269 million and discontinued-operations net income of USD 66 million, totaling USD 4,335 million. Continuing operating cash flow was USD 6,493 million, discontinued operating cash flow USD 1,216 million, and total operating cash flow USD 7,709 million. Continuing PPE purchases were USD 974 million and discontinued PPE purchases USD 38 million. These perimeters are retained separately in table.amd.outcome.continuing-discontinued-boundary.
ZT Systems explains why. AMD recorded USD 4,409 million of purchase consideration and issued 8,335,849 shares at the acquisition date. Retained ZT Design was included in continuing Data Center operations and described as immaterial, while ZT Manufacturing was discontinued and sold to Sanmina in October 2025. A total-cash-flow or total-net-income figure cannot silently replace continuing operations in the endpoint model. (claim.amd.outcome.zt-perimeter; evidence.amd.outcome.fy2025-zt-acquisition; evidence.amd.outcome.fy2025-zt-perimeter; evidence.amd.outcome.fy2025-zt-sale)
FY2025 Data Center revenue was USD 16,635 million and segment operating income USD 3,603 million, producing a deterministic 21.6592% segment operating margin. AMD attributed revenue growth to both EPYC processors and Instinct GPU accelerators, and the segment scope included multiple CPU, GPU, accelerator, networking, FPGA, and SoC categories. It is a useful segment endpoint but not a Zen-only profit series. (table.amd.outcome.data-center-multi-product; claim.amd.outcome.endpoint-data-center-multi-product)
Export-control charge: gross, reversal, and net remain visible
AMD described approximately USD 800 million of MI308 inventory and related charges after new U.S. export restrictions, approximately USD 360 million of reversal after licensed shipments, and an approximately USD 440 million net FY2025 effect. The deterministic subtraction reconciles USD 440 million. The amounts are reported facts; the regulatory and licensed-shipment causal explanation is an AMD company claim, not independent causal adjudication. (claim.amd.outcome.export-control-charge; evidence.amd.outcome.fy2025-warrant-export; evidence.amd.outcome.fy2025-export-net; table.amd.outcome.export-control-bridge)
The gross-to-net bridge matters. Treating only USD 440 million as the event would hide the inventory exposure and later license path; treating USD 800 million without the reversal would omit the reported recovery. Neither amount predicts future license decisions, demand, collectability, or cash recovery.
Liquidity, commitments, and conditional dilution are separate gates
At December 27, 2025, AMD reported USD 5,539 million of cash and cash equivalents and USD 5,013 million of short-term investments. Their selected sum is USD 10,552 million. Current carrying debt of USD 874 million plus long-term carrying debt of USD 2,348 million gives USD 3,222 million; subtracting that from the selected liquid-asset sum gives USD 7,330 million. This is a selected scale check, not net cash, surplus cash, distributable value, or complete capitalization. AMD separately reported approximately USD 12,200 million of unconditional commitments. (evidence.amd.outcome.fy2025-balance; evidence.amd.outcome.fy2025-liquidity-commitments; table.amd.outcome.liquidity-and-commitments)
The OpenAI warrant covered a maximum of 160 million AMD shares, while no vesting or exercise conditions had been met at the FY2025 endpoint. Compared as a one-decimal scale coordinate with 1,630 million ending shares, the maximum is about 9.8%. It is a conditional maximum, not issued, vested, exercisable, exercised, outstanding, or realized dilution. It remains separate from the actual 429 million Xilinx issuance. (evidence.amd.outcome.fy2025-warrant-export; table.amd.outcome.dilution-boundary; claim.amd.outcome.dilution-boundary)
Causal analysis: contribution without monopoly on explanation
The primary bounded hypothesis is hypothesis.amd.zen-epyc-contribution. AMD's reported launches, Microsoft and Google customer signals, and AMD's own description of EPYC participation in FY2021 segment growth support a plausible contribution to the pre-Xilinx operating turnaround. Confidence is moderate because the evidence is observational and product-level economics are absent. It does not support a causal percentage or sole attribution. (judgment.amd.outcome.causal-attribution)
The rival hypothesis is hypothesis.amd.portfolio-demand-and-external-context. Semi-custom products, GPUs and Instinct accelerators, acquired FPGA and DPU scope, AI and cloud demand, manufacturing and software execution, ZT, and export restrictions could explain an unknown and potentially large share of later results. These mechanisms are not mutually exclusive: Zen and EPYC can contribute while other products, acquisitions, demand, and external shocks also matter. (conflict.amd.outcome.zen-contribution-versus-expanded-scope; claim.amd.outcome.attribution-bounded)
A credible falsifier would be a controlled, stable-scope product bridge that either finds no incremental Zen or EPYC contribution or reliably isolates a different causal share after controlling for product mix, acquisitions, supply, demand, competition, and regulation. The selected filings and customer pages do not provide that design.
What was observable at the May 2016 cutoff
Part A's six gates were grounded in evidence available before the outcome reveal:
- Technical validation: Polaris sampling, Zen roadmap risk, and contemporaneous independent context made future claims testable but did not provide reproduced production-workload performance, efficiency, reliability, security, or total-system-cost proof. (evidence.amd.cutoff.polaris-sampling; evidence.amd.cutoff.zen-roadmap-risk; evidence.amd.cutoff.pcworld-zen-context)
- Roadmap execution: sampling and expected timing were visible, while qualification, volume, yield, customer availability, and retained adoption were not. (evidence.amd.cutoff.polaris-sampling; evidence.amd.cutoff.zen-roadmap-risk)
- Segment economics: FY2015 and Q1 2016 segments were mixed and pressured, and THATIC licensing recognition showed why licensing could not substitute for product proof. (evidence.amd.cutoff.fy2015-segments; evidence.amd.cutoff.q1-segments; evidence.amd.cutoff.thatic-recognition)
- Manufacturing and capital: GlobalFoundries obligations and the issuer's ATMP expectation showed material constraints but no complete qualified-capacity, yield, commitment, or required-reinvestment schedule. (evidence.amd.cutoff.gf-obligations; evidence.amd.cutoff.atmp-issuer-expectation)
- Funding resilience: the latest balance sheet, negative operating cash flow, and ratings and cash-risk disclosures made downside funding a real gate despite transaction proceeds. (evidence.amd.cutoff.q1-balance-sheet; evidence.amd.cutoff.q1-cash-flow; evidence.amd.cutoff.fy2015-ratings-cash-risk)
- Valuation readiness: historical results were available, but the governed packet lacked exact price, complete diluted capitalization, approved forecast and valuation, mandate, exposure, and risk budget. (evidence.amd.cutoff.fy2015-results; evidence.amd.cutoff.fy2015-balance-sheet; evidence.amd.cutoff.q1-results; evidence.amd.cutoff.q1-balance-sheet)
These were gates, not predictions of inevitable failure. The feasible counterfactual counterfactual.amd.follow-part-a-defer-and-reunderwrite keeps new capital deferred and routes any existing holding to human review until all six gates and a separate valuation package clear. Its return cannot be estimated without actual action and point-in-time security data.
Transferable candidate rules
- rule.stage-semiconductor-roadmaps-behind-independent-and-customer-proof requires technical, customer-production, stable-scope economic, manufacturing-funding, and valuation gates. Launches and customer announcements are signals, not complete evidence.
- rule.bridge-acquisition-and-reporting-perimeters-before-causal-attribution requires deterministic pre-close, acquired, retained, discontinued, consolidated, and share bridges before organic-growth or sole-cause language.
- rule.separate-regulatory-charges-and-conditional-dilution-gates preserves gross charge, reversal, and net effects and keeps every conditional-equity state distinct from actual issuance and realized dilution.
All three rules are candidate, not corpus-validated. Each has explicit indicators, reversal conditions, falsifiers, human-approval boundaries, and a null expected range because one episode cannot calibrate universal effect size, threshold, or horizon.
Reproducibility and abstentions
All derived values come from deterministic Python using an explicit Decimal context with precision 40 and half-even rounding. The hostile-context regression changes ambient precision, rounding, exponent bounds, and traps and reproduces the same outputs without leaking state. The code SHA-256 is f458fb2271eb0441ef805a5d2febcd73e1675aac9a789e2e764b827456867799; the canonical invocation SHA-256 is 9f790a9c15de68f87182461dbb9e77b6f0847bbc9ce8cfc13fbf088ce34e55b8; and the canonical execution-result envelope SHA-256 is a45234a9258b9eb66b8e5daecbd41f38e2ea1d1de24e7bef5ae979e26aad9a1. The run remains non-authoritative because no frozen model-registry receipt or governed validation-suite result exists. (model-run.amd.outcome-descriptive-checks)
The report does not calculate or imply TSR, intrinsic value, target price, price attractiveness, position size, acquisition IRR, Zen-only economics, maintenance capital, distributable cash, organic FY2021-to-FY2025 revenue growth, or realized OpenAI-warrant dilution. These are substantive abstentions.
Primary-source ledger
- @src.amd.outcome.fy2017-10k — AMD FY2017 Form 10-K, accepted February 27, 2018 at 21:18:58 UTC; issuer-reported launches, audited consolidated financials, cash flow, and mixed-segment boundary. SEC filing
- @src.amd.outcome.microsoft-lv2 — Microsoft Azure article dated December 5, 2017; customer-authored Lv2-Series EPYC announcement. The live page was retrieved in 2026 without a cutoff archive and does not establish volume or economics. Microsoft Azure announcement
- @src.amd.outcome.google-epyc — Google Cloud article dated August 8, 2019; customer-authored internal-use and planned-cloud-availability evidence. The live page was retrieved in 2026 without a cutoff archive. Google Cloud announcement
- @src.amd.outcome.fy2021-10k — AMD FY2021 Form 10-K, accepted February 3, 2022 at 22:22:50 UTC; pre-Xilinx audited operating, cash, balance-sheet, share, and original segment evidence. SEC filing
- @src.amd.outcome.fy2022-10k — AMD FY2022 Form 10-K, accepted February 27, 2023 at 21:20:39 UTC; Xilinx and Pensando close, consideration, issued shares, and acquisition-expanded product scope. SEC filing
- @src.amd.outcome.fy2025-10k — original AMD FY2025 Form 10-K, accepted February 3, 2026 at 23:14:52 UTC; continuing and discontinued operations, multi-product Data Center, ZT, export charge, liquidity, commitments, and warrant endpoint. The next-day amendment and later evidence are excluded. SEC filing
Cutoff observability reuses the frozen Part A source record. Part B evidence was researched only after the Part A freeze, and later evidence does not alter the May 2, 2016 knowledge state.