AMD (Trans): FY27 revenue to grow >80%; server biz to double

Below is a Dolphin Research-compiled $AMD(AMD.US) FY26 Q3 earnings call Trans

I. Key takeaways

1. Q4 guide: revenue around $17 bn (±3%). Midpoint implies approx. +66% YoY and +27% QoQ.

a. By segment: Data Center up approx. 50% QoQ. Client and Gaming down over 20% QoQ, while Embedded grows double digits QoQ.

b. Profit and shares: non-GAAP GPM around 55%, OpEx about $3.75 bn, net interest and other income approx. $4 mn, tax rate 13%.

c. Share count: diluted shares about 1.67 bn, including the expected Q4 close impact of the Taalas and World Labs deals. Not impacted by customer warrants.

2. 2027 outlook

a. Revenue: expected to grow over 80%, more than 2x the long-term model given at last Nov Analyst Day.

b. Data Center: server revenue to double (vs. prior view of >70% growth), and DC AI to outpace expectations from 90 days ago. Data Center to exceed 80% of revenue, driving most of the top- and bottom-line growth; 2026 DC revenue projected to more than double vs. 2025.

c. Embedded: growth to accelerate in 2027.

3. Warrant accounting: with MI450 series shipments in Q3, AMD began recognizing the accounting effects of previously announced customer warrant agreements.

a. This quarter: GAAP diluted EPS of $1.46 includes a $1.1 bn non-cash expense in other income and expense, mainly reflecting stock price appreciation and changes in probability of meeting vesting conditions.

b. Going forward: quarterly fair value adjustments are expected to increase and will fluctuate with the share price and other valuation inputs. No impact on non-GAAP results or FCF.

4. Key Q3 financials

a. Aggregate: revenue of $13.4 bn, above the high end of guidance. GPM at 56%, up 190 bps YoY on better mix; OPM at 29%, up over 500 bps YoY.

b. Segment margins: Data Center 34% (25% a year ago); Embedded 41% (33%); Client and Gaming 15% (21%).

c. Cash flow and balance sheet: to meet DC demand, inventory rose approx. $3 bn QoQ to $11.5 bn. FCF was $554 mn, weighed by working capital and capacity investments, and AMD will keep investing in inventory and capacity into 2027.

Ending cash and short-term investments were $15.8 bn, with total debt of $7.1 bn including $4.75 bn of senior notes issued in Aug.

II. Earnings call details

2.1 Management highlights

1. Server CPU: Agents are becoming the next killer app for cloud CPUs, and EPYC share continues to climb.

a. Growth: cloud and enterprise server sales both more than doubled YoY, beating the prior view of >80% growth. Turin now has 540+ public cloud instances; enterprise sell-through hit a record with dozens of new first-time customers.

b. Agents and head nodes: for example, Muse surpassed 10 mn activated users in the first 6 weeks for personal Agents. Head nodes won large deals at multiple frontier labs and cloud providers, with OpenAI selecting Turin as the host CPU for its first in-house accelerator platform.

c. Venice: revenue recognition began in Q3; partners have committed to 240+ platforms (20% more than Turin), with cloud instances to launch in early 2027. A Helion rack supports up to 144 Venice CPUs and nearly 74k threads.

2. Data Center AI: MI455X and Helios racks entered volume production and shipments in Q3, with deployments set to ramp sharply in Q4.

a. Customers: initial cohort includes Anthropic, Meta, Microsoft, OpenAI, and Oracle. From Q4, HPE, Dell, Lenovo, and Super Micro will begin shipping systems.

b. MI430X: selected for Europe’s next-gen flagship LUMI AI supercomputer, going live in H2 2027. Delivers nearly 9x the double-precision performance vs. a competing product.

c. Software and M&A: launched ROCm 10; acquiring Taalas (low-latency inference silicon) and World Labs (spatial intelligence), with both expected to close by year-end.

d. MI500: co-optimizing MI500 with multiple frontier labs for their multi-GW deployments.

3. Client and Gaming: commercial PC penetration is deepening, but memory price hikes are pressuring overall PC demand.

a. Commercial: 8 of the top 10 global banks and 7 of the top 10 healthcare and auto companies use Ryzen Pro PCs.

b. Market: sell-through slowed toward quarter-end, and AMD aligned shipments to sell-through.

c. Gaming: revenue fell 40% YoY, mainly on fewer semi-custom units in the back half of the console cycle. Radeon AI Pro grew sharply QoQ.

4. Embedded: growth accelerated for the fourth straight quarter, and 2026 design wins are on track to exceed $18 bn for a third consecutive record year.

a. Drivers: aero-defense, auto, comms, and industrial led the gains, with DC build-outs boosting control-plane and optical networking demand. Strong double-digit growth is likely over the next few years.

5. Supply and investment: AMD has secured key components needed for high growth in 2027. After 2027, it will step up R&D and co-invest with suppliers in long-term capacity.

2.2 Q&A

Q: With revenue up over 80% in 2027, are wafer, packaging, substrate, and memory supplies secured, and would demand be higher with more supply?

A: Demand exceeds supply, and supply has been planned years ahead while AMD continues to pursue upside. AMD has co-planned deployment with customers and secured wafers, substrates, advanced packaging, memory, and more with supply chain partners.

Server will roughly double on top of a high 2026 base, with strong visibility in DC AI and good progress on supporting infrastructure.

Q: How long will the PC downturn last, why did AMD outperform before, and how will it exit the trough next year?

A: Planning assumes PC unit shipments decline double digits in 2027. AMD had flagged a weaker-than-seasonal second half, which played out; component inflation raised PC prices, dampening more price-sensitive consumer demand and slowing sell-through.

AMD is prudently aligning shipments to keep channel and customer inventory healthy and is focusing on enterprise (still underpenetrated), premium consumer, and desktop channels. PC and DC supply chains are fungible, allowing supply to be shifted based on market signals.

Q: Do anchor customers have financing, land, power, and data center capacity for approx. 3 GW of AMD deployments in 2027?

A: Co-planned with customers, AMD has clear line of sight to which DCs and locations will host the GPUs. Visibility is good on land, power, and facilities for 2027, and Venice, MI455, and Helios competitiveness gives confidence.

AMD did not specifically address customer financing.

Q: With competitors investing in ecosystems and offering residual value guarantees, will that pressure AMD from 2028 onward?

A: The first principle is to invest in AMD’s own growth, with selective investments in partners. With a strong balance sheet, funds are prioritized for R&D, supply chain, and capacity; AMD selectively invests in key partners, for example by locking in DC infrastructure years ahead to ensure new silicon launches are fully supported.

Q: How do frontier labs split workloads between in-house ASICs and AMD GPUs, and where does AMD have a performance edge?

A: Decisions hinge on performance and TCO, where Helios is advantaged. Q3 was the first quarter of revenue on MI455 and Helios racks, with a steep Q4 ramp and continued volume growth through 2027.

The key is to keep proving cross-workload TCO leadership; general-purpose silicon remains more flexible across diverse workloads. AMD is co-designing deeply with frontier labs, and MI500 and MI600 after MI450 are being tuned to their needs.

Q: With HBM prices up, why is revenue per GW still holding at $15–20 bn, and can higher costs be passed to large customers?

A: Partly locked by long-term agreements, with future cost increases to be shared with customers. AMD has LTAs with some suppliers and customers; memory and other components face inflation, and AMD is negotiating to share costs.

The core metric remains tokens per dollar on a TCO basis, and even with inflation, Instinct and Helios offer leading TCO. AMD did not directly confirm the revenue per GW framework.

Q: With 2027 growth far above the long-term model, how will OpEx scale?

A: In 2027, OpEx growth will be less than half of revenue growth, implying significant OPM expansion. 2026 investments are translating into a revenue acceleration, with revenue growing much faster than OpEx.

The Q4 guide continues this pattern and is expected to extend into 2027.

Q: Can AMD hold the 55% GPM floor during the Helios ramp?

A: AMD did not commit to a 55% floor and said GPM is driven mainly by Data Center mix. In Q3, MI450 ramped and DC AI outperformed early-quarter expectations; in Q4, DC AI revenue will exceed Servers, with margin slightly below the company avg but adding substantial gross profit dollars.

With DC exceeding 80% of revenue in 2027 and AI ramping while Servers double, mix differences will drive quarterly volatility; Embedded strength will also support margins.

Q: How does 2027 DC revenue split between GPUs and CPUs, and what is the cadence through the year?

A: Server CPU is roughly doubling, while DC AI will grow well above 2x. These are directional markers only; Q4 is the first quarter where DC AI surpasses CPU at scale, and both should grow steadily through the year.

Q: How should customer warrants be modeled for 2026–2027?

A: Warrant impacts are threefold, with revenue already netted, and non-cash expenses excluded from non-GAAP. First, company-reported revenue is net, reduced by grant-date fair value; second, shares only enter diluted count after customers meet shipment and commercial milestones and warrants vest and are exercised, and AMD will provide clear share guidance for 2027.

Third, quarterly fair value remeasurement is a non-cash expense and does not affect non-GAAP results or FCF.

Q: For the Server CPU doubling, how much comes from units, cores, and ASP?

A: Growth is mainly units and cores, with a smaller ASP contribution. Shipments are rising meaningfully; moving into Venice brings higher core counts for some products, lifting ASPs alongside cores.

Q: How is Q1 seasonality, especially for Embedded?

A: Embedded typically has Q1 seasonality, but should grow significantly in 2027, strengthening each quarter. AMD did not comment on Q1 for Client, Gaming, or Data Center.

Q: By how much will DC AI exceed Servers in Q4?

A: DC AI revenue will be materially above Servers in Q4 on a dollar basis. MI450 sees a major sequential jump from Q3 to Q4; Servers also grow strongly, but AI ramps faster.

AMD did not quantify the gap.

Q: Approximate entry and exit GPM for 2027?

A: AMD did not provide a quarterly GPM path for 2027 and said mix remains uncertain. Margins depend on the DC AI vs. Server split, with both in supply-constrained environments as AMD works to secure more supply; Q4 (AI far above CPU) can serve as a reference point, with potential quarterly swings through 2027.

Q: How does a CPU-only Helion rack differ from traditional racks, and what is the GPU blade attach rate?

A: Helion is configured by use case: all-CPU high-density for cloud, mixed CPU-GPU for enterprises. Venice delivers best-in-class single-socket performance, and a CPU-only build makes for a powerful Agent rack.

The design has been contributed to OCP as a fully open rack, and OEMs, ODMs, and customers have responded positively; AMD did not provide GPU attach rates.

Q: Does a doubling in Servers meet demand, and will there still be shortages next year?

A: There is more demand, and AMD is working with the supply chain to add capacity. The Server TAM is set for approx. 50% CAGR over the next 3–4 years (with opening remarks citing about $220 bn by 2030), driven by traditional cloud and enterprise, Agent CPUs and workloads, and head nodes, and AMD is gaining share across all three.

Ensuring DDR5 and other components are in place is also essential, and customer demand forecasts have been raised materially across nearly all accounts.

Q: What is the share goal for Helios?

A: The goal is to be a major compute supplier to the largest customers and keep increasing share. Customers span frontier labs and hyperscale cloud; near-term focus in Q4 and 2027 is the Helios or MI450 ramp, while AMD is already engaged with its largest customers on MI500, MI600, and beyond as demand keeps rising.

Q: Plans to use Helios to penetrate the enterprise market?

A: Enterprise is important, and AMD has introduced enterprise GPUs. Enterprise AI adoption lags frontier labs and hyperscalers; many enterprise CPU customers are engaging on how to deploy AI efficiently on-prem and in the cloud, and AMD will broaden its portfolio to grow share at large enterprises.

Q: If a key competitor cuts PC prices, will AMD follow?

A: AMD will remain disciplined, focusing on the highest-value and most profitable PC segments. These include enterprise, premium consumer, and desktop channels; exposure to volatile low-end markets is minimal.

PC remains a key endpoint in AMD’s cloud-to-edge AI strategy, and AMD did not directly say whether it will match price cuts.

<End of text>

Risk disclosure and statements:Dolphin Research Disclaimer and General Disclosure