Advanced Micro Devices Doubled Data Center Sales. FCF Must Grow 41%

AMD's latest operating performance provides genuine evidence that its AI and server products have moved beyond development-stage adoption. Q2 FY2026 revenue increased 50.1% year over year to $11.54 billion. Data Center revenue rose 107.0% to $6.72 billion and represented 58.2% of consolidated sales. Gross margin reached 53.8%, GAAP operating income was $1.99 billion, operating margin rose to 17.3%, and diluted EPS increased 155.6% to $1.38. Data Center segment operating income was $2.10 billion, equivalent to a 31.3% segment margin before centralized acquisition amortization and stock compensation.

The comparison nevertheless contains accounting noise. Q2 FY2025 included approximately $800 million of inventory and related charges caused by MI308 export restrictions, while full-year FY2025 net charges were approximately $440 million after subsequent developments. The prior-year quarter also included an $853 million tax benefit from releasing uncertain tax positions. H1 FY2026 included $549 million of long-term investment gains, and Q2 other income of $598 million caused net income to exceed operating income. H1 also included $1.10 billion of acquisition-related amortization and $990 million of stock compensation.

Cash flow was the weak point. Q2 FCF declined 9.8% year over year and 39.3% sequentially to $1.56 billion. That does not negate the operating improvement, but it shows that revenue and accounting earnings have not yet translated proportionally into cash. The 14.0-point year-over-year gross-margin improvement also should not be extrapolated mechanically because the comparison includes the prior export-control charge.

The longer financial record shows both AMD's progress and its cyclicality. Revenue increased from $16.43 billion in FY2021 to $34.64 billion in FY2025. Over the same period, GAAP operating margin moved from 22.2% in FY2021 to 5.4% in FY2022, 1.8% in FY2023, 7.4% in FY2024 and 10.7% in FY2025. FCF was $3.22 billion in FY2021, $3.12 billion in FY2022, $1.12 billion in FY2023, $2.40 billion in FY2024 and $6.74 billion in FY2025. The business is improving rapidly, but neither margins nor cash generation have followed a straight line.

Data Center includes EPYC CPUs, Instinct GPUs, Pensando data-processing units and AI network-interface cards, FPGAs and adaptive SoCs. Customers include hyperscalers, server manufacturers, original-design manufacturers, enterprises and systems integrators. Client consists primarily of Ryzen and Ryzen AI processors and chipsets sold to PC OEMs, distributors and commercial PC vendors. Gaming includes Radeon GPUs and semi-custom chips for consoles and handheld devices. Embedded supplies CPUs, FPGAs, adaptive SoCs and systems-on-module to industrial, automotive, aerospace, communications and healthcare buyers.

The four businesses carry different economics. Data Center produced a 31.3% segment operating margin in Q2 FY2026. Client and Gaming, reported together for operating profit, generated $582 million at a 15.2% margin. Embedded generated $386 million at a 39.5% margin. Consolidated GAAP operating margin was lower because AMD recorded $1.08 billion of centralized costs in All Other, primarily acquisition-related amortization and stock compensation.

Revenue growth currently depends on units and mix more than broad-based pricing. FY2025 client processor units rose 31% and average selling price increased 15%. In Q2 FY2026, client units increased 34%, but average selling price declined 6%. Data Center's rise to 58.2% of quarterly revenue improved consolidated mix. Other incremental drivers include cloud and server design-win conversion, embedded product qualifications and market-share changes against Nvidia in accelerators and Intel in CPUs.

The recurring-revenue characteristics are limited. Approximately 9% of FY2025 revenue was recognized over time from custom products and development services. Embedded and semi-custom design wins can create multiyear shipment streams, but those remain product sales rather than subscriptions. Customer forecasts usually have no minimum purchase commitments, and purchase orders may often be revised or canceled. Consequently, headline deployment plans and backlog-like figures require scrutiny of pricing, timing, cancellation rights and financing conditions.

AMD's opportunity spans more than accelerator silicon. EPYC CPUs support general-purpose cloud, enterprise, high-performance-computing and AI-host workloads. Instinct accelerators address training and inference. Pensando products provide infrastructure offload and AI networking. Xilinx-derived FPGAs and adaptive SoCs serve configurable workloads. ZT Systems' retained design capabilities extend AMD toward server and rack-level architecture. Together, these assets allow AMD to compete as a compute platform rather than as a single-chip vendor.

EPYC's advantages come from architecture, chiplets, access to advanced process technology and product cadence rather than an unassailable instruction-set monopoly. Server customers evaluate core density, performance per watt, memory capacity, virtualization, reliability, security, x86 compatibility and total cost of ownership. Qualification is extensive, creating meaningful but not prohibitive switching costs. Intel remains the principal x86 alternative, while Arm-based CPUs and internally designed hyperscaler processors are indirect competitors.

For AI accelerators, usable system performance matters more than a chip benchmark. Customers evaluate memory capacity, uptime, power consumption, networking efficiency, model compatibility and developer productivity. The full technical stack includes compute dies, high-bandwidth memory, advanced packaging, interconnects, drivers, compilers, kernels, libraries, framework integration and cluster-management software. AMD identified MI350 demand as a principal Q2 Data Center growth driver and also launched MI350P for installation into existing infrastructure.

The main weakness is the surrounding software and systems ecosystem. Nvidia's libraries, installed base, developer tools and proprietary interconnects lower deployment risk. ROCm offers a more open alternative, but customers may bear code-porting, model-validation, developer-retraining, performance-tuning and cluster-operating costs. ROCm does not have to replace CUDA universally. It must become reliable and economical for enough high-volume workloads to support durable accelerator share and acceptable margins.

AMD reported approximately 7,200 U.S. patents and 2,200 pending U.S. applications at December 27, 2025, but patents are supporting assets rather than the central moat. Defensibility comes from architecture, software, process-node access, packaging expertise, customer qualification, x86 compatibility, FPGA toolchains, supplier commitments and engineering talent. The hardware is credible and difficult to reproduce. The AI software moat remains materially weaker than Nvidia's.

Nvidia overlaps with AMD in AI accelerators, networking, gaming GPUs and CPUs. Its FY2026 revenue was $215.9 billion, up 65%, and its primary advantages are software, scale and networking integration. Its weaknesses relative to AMD include high pricing, customer concentration and a more closed ecosystem. Nvidia also disclosed $279 billion of supply commitments as of July 26, 2026, illustrating both the scale of AI demand and the capital required to secure it.

Intel competes in server and client CPUs, accelerators and FPGAs. Q2 2026 revenue was $16.1 billion, up $3.3 billion year over year. Intel retains an installed base, manufacturing assets and x86 scale, but its execution record and process transitions create risk. Broadcom and Marvell compete through custom accelerators, networking and interconnects. Broadcom's strength is custom silicon and switching scale; Marvell brings cloud custom-silicon relationships but a smaller compute ecosystem.

The Arm ecosystem competes in server and client processors through power efficiency and a broad licensing model, offset by platform qualification and software fragmentation. Hyperscalers are also direct competitors through TPUs, Trainium, Maia, MTIA and other internal chips. Their workload-specific economics and internal integration can displace merchant processors where workloads are stable and high-volume. Huawei and other Chinese vendors benefit from domestic access and policy support but face export restrictions and software-maturity constraints. In programmable and embedded products, AMD faces Altera, Lattice, Microchip, NXP, Qualcomm and Texas Instruments.

Customers choose AMD for a credible second source, competitive performance per dollar or watt, high-memory configurations, ROCm's open approach, an integrated CPU-GPU-FPGA-networking portfolio, customization, co-engineering and access to constrained capacity. The commercial risk is that customers may principally use AMD to negotiate lower prices from Nvidia or Intel. A design win can benefit the buyer without producing attractive lifetime economics for AMD shareholders.

The likely long-term market is mixed rather than winner-take-all. Merchant GPUs retain advantages when workloads change quickly, compatibility matters, general-purpose flexibility is valuable and time to market outweighs unit cost. Custom ASICs are stronger for predictable workloads at enormous scale. AMD can succeed in that mixed market, but broad AI infrastructure growth does not guarantee that it captures the associated revenue or profit pool.

AMD disclosed a product purchase agreement with OpenAI covering deployment of six gigawatts of AMD GPUs, with the first gigawatt expected to use MI450 products. This is more substantial than an engineering sample or memorandum of understanding. It is not equivalent to six gigawatts of non-cancellable backlog because pricing, timing, minimum volumes, termination rights and financing conditions were not disclosed.

OpenAI and Meta can each earn warrants for up to 160 million AMD shares at an exercise price of $0.01. Vesting depends on Instinct purchase milestones, technical and commercial conditions, stock-price targets and, for OpenAI, additional stock-performance thresholds. No warrant shares had vested or become exercisable as of June 27, 2026. The relationships are therefore real commercial validation, but their value to existing shareholders cannot be judged from deployment scale alone.

No customer generated at least 10% of consolidated FY2025 or FY2024 revenue, although one customer represented approximately 11% of accounts receivable at December 27, 2025. That historical diversification may not describe AMD's future AI economics. OpenAI, Meta and other hyperscalers can create substantial economic concentration even where purchases flow through OEMs, ODMs or other authorized channels.

The disclosed evidence of adoption includes $6.72 billion of quarterly Data Center revenue, 107.0% year-over-year growth, the OpenAI purchase agreement, purchase-linked warrants and hyperscaler buying through direct and indirect channels. Missing evidence includes customer-specific accelerator revenue, the MI350-versus-EPYC split, purchase-price commitments, non-cancellable backlog, gross margins by product or customer, deployment utilization and the economics of individual warrant milestones.

The missing unit economics are equally important. AMD does not disclose Instinct units, accelerator revenue or gross profit per unit, EPYC shipments, accelerator utilization, revenue per wafer, high-bandwidth-memory cost, packaging yield, customer acquisition cost, customer lifetime value or customer-level contribution margin. A six-gigawatt agreement could produce mediocre economics if won through low pricing, financing support, warrants or expensive supply commitments.

H1 FY2026 net income was $3.68 billion. Adding $1.52 billion of depreciation and amortization and $990 million of stock compensation, removing $549 million of investment gains, and including deferred taxes, other items and working capital produced $5.32 billion of operating cash flow. Capital expenditure was $1.20 billion, leaving reported H1 FCF of $4.12 billion.

The working-capital movements deserve attention. Receivables increased by $966 million, inventory by $548 million, and prepayments and other assets by $991 million. Those uses of cash were largely offset by a $2.17 billion increase in accounts payable and a $61 million increase in accrued and other liabilities. AMD attributed the payable movement primarily to payment timing.

Reported TTM FCF was $8.4 billion. The memo's normalization removes half of the H1 payable increase, or approximately $1.09 billion, while allowing the remainder because suppliers naturally finance part of AMD's higher revenue and inventory. That produces H1 normalized FCF of approximately $3.04 billion. Adding approximately $4.28 billion of reported H2 FY2025 FCF gives normalized TTM FCF of about $7.3 billion.

The 50% adjustment is an explicit analyst assumption rather than company guidance. At $7.3 billion, normalized FCF yield is approximately 0.9%, price-to-normalized FCF is about 114 times, and normalized H1 FCF conversion is approximately 83% of net income. Stock compensation is not deducted again because dilution is captured through the share count; treating stock compensation as costless while also ignoring dilution would overstate value.

Growth creates shareholder value only if FCF rises faster than diluted shares, customer incentives remain below lifetime profit, supply commitments do not create obsolete inventory, gross margin remains above 50%, and R&D sustains differentiation. AMD's operating leverage is encouraging, but the customer warrants make per-share cash flow more important than consolidated sales.

At June 27, 2026, AMD held $5.09 billion of cash and equivalents and $8.03 billion of short-term investments, for $13.11 billion of liquid assets. Gross debt principal was $3.25 billion, leaving net liquid cash of $9.86 billion. Current assets were $31.52 billion against $12.08 billion of current liabilities, producing a 2.6 times current ratio. H1 EBIT of $3.47 billion covered $74 million of interest expense approximately 46.8 times.

Debt maturities are manageable: $875 million of notes mature in 2026, followed by $625 million in 2028, $750 million in 2030, $500 million in 2032 and $500 million in 2052. AMD also has an undrawn $5.0 billion revolver and a $5.5 billion commercial-paper program. Funded leverage is not the central financial risk.

Working capital and contractual obligations are more consequential. Inventory rose from $5.73 billion at FY2024 year-end to $8.47 billion in Q2 FY2026. Finished goods more than doubled during FY2025, from $1.09 billion to $2.24 billion. Approximate Q2 metrics were 57 receivable days, 145 inventory days, 91 payable days and a 111-day cash-conversion cycle. Product ramps explain part of the buildup, but rapid product cycles create obsolescence risk.

AMD disclosed $30.3 billion of unconditional commitments at June 27, 2026, including $17.4 billion due during the rest of FY2026. It also disclosed $1.2 billion of commenced lease commitments, $4.5 billion of uncommenced leases and $4.1 billion of data-center lease guarantees. Subsequent commitments included investments of up to $5.0 billion and $9.5 billion of data-center leases. Categories may overlap and do not all represent probable losses, but the exposure is much larger than reported debt.

AMD does not own leading-edge fabs, yet it must secure wafers, high-bandwidth memory, substrates, packaging, testing, networking components and data-center capacity. H1 capex increased 142% to $1.20 billion, while construction in progress rose from $508 million to $1.15 billion. Unconditional commitments increased from $12.2 billion at FY2025 year-end to $30.3 billion six months later. If MI450-class demand materializes, those commitments are rational. If deployments slip, they can create underused capacity, inventory write-downs, lower selling prices and weaker margins.

AMD's diluted share count increased from approximately 1.61 billion in FY2023 to 1.66 billion in Q2 FY2026. Basic shares outstanding were 1.632 billion at June 27 and again on July 29, 2026. At FY2025 year-end, AMD also had approximately 2 million options, 33 million unvested time-based restricted shares and 2 million unvested performance shares. Another 107 million shares remained available under the amended equity plan, although plan availability is not current dilution.

The OpenAI and Meta warrants are much larger. Together they cover 320 million shares at $0.01 per share and would generate maximum cash proceeds of only approximately $3.2 million. Relative to 1.632 billion basic shares, full vesting would increase the share count by 19.6% and reduce existing holders' ownership by 16.4%. At the supplied $503.60 price, the gross value of the warrant shares would be approximately $161.2 billion.

That $161.2 billion is conditional and is not a current accounting expense. It nevertheless represents a potentially enormous transfer of future upside. A valuation that assumes large OpenAI and Meta revenue while excluding the related warrants is internally inconsistent. The realistic valuation share count is approximately 1.95–2.02 billion in a successful AI scenario, compared with current GAAP dilution of 1.659 billion.

AMD repurchased $221 million of shares during H1 FY2026 while recording $990 million of stock compensation. It also spent $341 million on tax withholding for vested awards, yet basic shares increased by approximately two million during the six months. FY2025 repurchases totaled $1.3 billion, and $9.2 billion of authorization remained at June 27, 2026. Buybacks have not been large enough to offset the potential customer dilution.

The warrants should be evaluated as customer-acquisition costs. More than 160 million warrants vesting without at least $5 billion of incremental sustainable annual FCF would fail the memo's stated economic test. Revenue growth alone is insufficient; the purchases must produce enough cumulative cash flow to compensate shareholders for the shares transferred.

Lisa Su has served as chief executive since October 2014 and chair since February 2022. During her tenure, AMD moved from approximately $4.3 billion of revenue and negative operating margins in FY2016 to $34.6 billion of revenue and a 10.7% GAAP operating margin in FY2025. Her record includes restoring a competitive CPU roadmap, establishing EPYC, executing chiplet-based products, acquiring adaptive-computing and networking assets, building an accelerator business and preserving balance-sheet strength.

Mark Papermaster, chief technology officer, joined in 2011 after roles at Cisco, Apple and IBM. His experience spans silicon, device hardware, microprocessors and servers, fitting AMD's transition toward systems integration. Forrest Norrod leads Data Center Solutions, where H1 FY2026 revenue reached $12.49 billion, up 81%, and segment operating income rose to $3.70 billion. CFO Jean Hu oversees finances as commitments, acquisitions, investment spending and customer warrants grow.

R&D remains AMD's most defensible use of capital. FY2025 R&D was $8.09 billion, or 23.4% of revenue. H1 FY2026 R&D increased 36% to $4.93 billion, or 22.6% of revenue, supporting accelerators, ROCm, CPUs, networking, packaging, FPGA tools and systems engineering. The $4.4 billion ZT Systems acquisition included $3.2 billion of cash, $860 million of AMD stock and $361 million of contingent consideration. AMD later sold ZT's manufacturing business and retained system-design capabilities.

Insider ownership is limited. All 16 directors and executive officers collectively owned approximately 8.11 million shares in the 2026 proxy, less than 0.5% of the current basic count. Disclosed holdings included approximately 1.83 million shares for Papermaster, 425,000 for Norrod, 185,000 for Darren Grasby and 148,000 for Hu. Lisa Su's exact beneficial ownership was not surfaced in the extracted proxy data and is not estimated.

Q2 FY2026 Rule 10b5-1 plans permitted maximum sales of 450,000 shares by Su, 116,784 by Norrod, 30,000 by Hu and 135,000 by Joseph Householder. These are prearranged plans rather than evidence of misconduct, but recent disclosed activity was directed toward sales rather than open-market purchases. Su's FY2025 compensation was $55.16 million, or 341 times median employee compensation. Incentives emphasize financial performance, strategic milestones, non-GAAP income or EPS and relative shareholder return; FCF per diluted share and ROIC deserve greater explicit weight.

AMD's governance structure is above average for a large technology company. All directors other than Lisa Su are independent, an independent lead director is in place, directors are elected annually by simple majority, and qualifying 3%-for-three-year holders have proxy access for up to 20% of board seats. Board committees are independent, no dual-class voting structure was identified, no FY2025 related-party transactions were reported, employee and director hedging is prohibited, and pledging requires preapproval.

The weaknesses are a combined CEO-chair role, low insider ownership, reliance on relative total shareholder return in compensation, broad equity-plan capacity and potential customer dilution outside conventional employee awards. Su's record and the independent lead director make the combined role tolerable, but warrant economics, acquisitions and long-term commitments require strong board scrutiny.

The forensic accounting record is clean. Ernst & Young has served as auditor since 1970 and issued unqualified FY2025 financial-statement and internal-control opinions. No accountant disagreements, material weaknesses, material Q2 legal proceedings, related-party transactions or financial restatements were identified in the reviewed materials. AMD filed a 10-K/A on February 4, 2026, but the supplied documents do not describe it as a financial-statement restatement.

The filings are direct about export controls, inventory charges, conditional warrants, purchase commitments, order cancellations, TSMC dependence, customer-designed chips and working-capital timing. Commercial economics are less transparent. AMD publicizes deployment scale, partnerships and addressable markets without disclosing customer prices, minimum purchases, expected margins or the value required for warrant vesting. The appropriate concern is not an integrity discount but an economic-governance discount around dilution and commitments.

FY2025 revenue by billing location was $11.36 billion, or 32.8%, in the United States; $7.75 billion, or 22.4%, in China including Hong Kong; $5.19 billion, or 15.0%, in Taiwan; $4.28 billion, or 12.4%, in Singapore; and $6.06 billion elsewhere. Billing location is not the same as end demand because products billed in Taiwan or Singapore may support customers in other markets.

U.S. export controls restrict selected Instinct and Versal products. The MI308 restriction caused approximately $800 million of Q2 FY2025 inventory and related charges. AMD received licenses in February 2026 to ship certain MI325 products to Chinese customers, but said it did not yet know whether China would allow the imports. The products must first be inspected in the United States and are subject to a 25% U.S. import tariff during that process.

A policy stress case assumes AMD loses half of FY2025 China-billed revenue, or approximately $3.9 billion, at a 55% contribution margin and incurs $1.0 billion of inventory and redesign charges. The resulting pre-tax effect would be approximately $3.1 billion before mitigation. AMD would remain solvent, but earnings and valuation would decline sharply.

Taiwan is the more existential operating dependency. TSMC manufactures AMD's advanced-node products, and Asian suppliers provide much of the packaging and testing. A severe disruption would not be replaceable in the short term. Net cash would provide liquidity but could not replace advanced-node wafer production.

Tax policy also affects reported earnings. H1 FY2026's effective tax rate was 11.8%, supported by foreign-derived-income deductions and R&D credits. Applying the full 21% U.S. statutory rate to $4.16 billion of pre-tax continuing income would have reduced H1 net income by approximately $383 million, or about 10%. AMD does not primarily depend on direct subsidies, but it benefits from favorable tax treatment, trade policy and a globalized supply chain.

AMD is exposed to semiconductor inventories, PC replacement cycles, gaming-console generations, discrete graphics demand, enterprise capital spending, embedded corrections, hyperscaler AI spending, and foundry and packaging capacity. These cycles do not move together, which provides some portfolio diversification, but they can still reinforce one another during an industry downturn.

Current GAAP earnings are not obviously at a cyclical peak. FY2021 operating margin of 22.2% remains above FY2025's 10.7% and Q2 FY2026's 17.3%. Embedded is recovering from an inventory correction, AI accelerators are ramping, and consolidated margins remain below what the bull case assumes. A non-peak business, however, does not make a 129.5 times trailing P/E inexpensive.

Q2 Data Center growth of 107.0% will not persist indefinitely. Inventory is elevated, purchase and lease commitments have expanded rapidly, AI customers are building internal processors, and gross-margin comparisons benefited from the prior export charge. The business can grow materially from current levels while still failing to meet the expectations embedded in the stock.

At $503.60, AMD's market capitalization is $835.5 billion and enterprise value is approximately $825.6 billion. The shares trade at 129.5 times trailing earnings, 99.5 times reported TTM FCF and approximately 114 times normalized FCF. EV is 20.0 times TTM revenue, 127.2 times EBIT and approximately 87 times reconstructed EBITDA. The reported FCF yield is 1.0%; normalized yield is about 0.9%.

A forward illustration using analyst assumptions—not company guidance—puts FY2027 revenue at $55 billion, operating margin at 22%, tax rate at 17%, diluted shares at 1.85 billion, EPS at approximately $5.40 and FCF at $10.5 billion. Even then, the current price represents approximately 93 times forward earnings and 80 times forward FCF. AMD is being valued more like an established dominant platform than an improving challenger.

A simple terminal-value test demonstrates the hurdle. Supporting the current $825.6 billion enterprise value requires sustainable annual FCF of $41.3 billion at a 20 times multiple, $33.0 billion at 25 times or $27.5 billion at 30 times. Those figures ignore the time needed to reach the terminal level and therefore understate the operating requirement.

The five-year reverse DCF begins with $8.4 billion of TTM FCF, uses a 10% discount rate, includes interim annual cash flows, applies a 25 times terminal FCF multiple and adds $9.9 billion of current net liquid cash. Under those assumptions, annual FCF must grow approximately 41% for five years, reaching roughly $47 billion in year five, to support $503.60 per share. Even using a more generous 30 times terminal multiple, required annual growth remains approximately 36%–37%.

The revenue implication is demanding. At a sustainable 20% FCF margin, producing $47 billion of FCF requires $235 billion of revenue. At 25%, required revenue is $188 billion; at 30%, it is $157 billion. TTM revenue is approximately $41.3 billion. The market therefore requires some combination of fourfold revenue growth, mid-to-high-20s FCF margins, major accelerator share, strong EPYC and networking growth, broad ROCm adoption, warrant economics that are more than offset by customer purchases, and preservation of a premium terminal multiple.

The base DCF assumes FCF of $10.5 billion in FY2027, $13.0 billion in FY2028, $16.0 billion in FY2029, $19.5 billion in FY2030 and $24.2 billion in FY2031. A 10% discount rate, 24 times terminal FCF multiple, $9.9 billion of net cash and 1.96 billion success-case diluted shares produce approximately $220 per share.

The bear case assumes FY2031 revenue of $70 billion, a 14% operating margin, 14% FCF margin, $9.8 billion of FCF, an 18 times terminal multiple and 1.75 billion shares, producing present value of $87. The bull case assumes $180 billion of revenue, 32% operating margin, 28% FCF margin, $50.4 billion of FCF, a 30 times multiple and 2.02 billion shares, producing $523. Probability weighting the $87 bear case at 20%, $220 base case at 55% and $523 bull case at 25% gives approximately $270 per share.

The operating thesis would weaken materially if Data Center growth falls below 25% for two consecutive quarters before the MI450 ramp is established, consolidated gross margin falls below 50% without a clearly temporary charge, or normalized FCF margin remains below 15% despite Data Center contributing more than half of revenue. A delay of more than two quarters in MI450 production would also undermine the expected product cadence.

The software and competitive thesis would fail if ROCm remains confined to a small number of heavily supported customers, or if Nvidia or custom ASICs establish a decisive performance-per-dollar advantage. Repeated roadmap delays or the departure of key technical executives would indicate that AMD's historical execution advantage is deteriorating.

The customer-economics thesis would fail if more than 160 million customer warrants vest without at least $5 billion of incremental sustainable annual FCF. Concentration would also become unacceptable if one customer exceeds 20% of revenue or 25% of receivables. These thresholds test whether commercial validation is producing durable economics rather than subsidized volume.

Balance-sheet and supply warning signs include inventory growth exceeding revenue growth by more than 20 percentage points followed by material write-downs, or net cash becoming more than $10 billion of net debt because of capacity, investment or lease commitments. Export-control charges above $1 billion in a year would show that geopolitical restrictions are causing recurring rather than exceptional losses.

A severe TSMC or Taiwan disruption would invalidate near-term production assumptions regardless of AMD's liquidity. Finally, a material accounting restatement, internal-control weakness or regulatory enforcement action would overturn the current integrity assessment. These are not routine volatility markers; they are observable conditions under which the long-term case should be reconsidered.