Aug 4, 2026 earnings call — Helios AI platform ramp and server CPU growth
Record revenue of $11.5 billion, led by data center revenue of $6.7 billion.
Revenue $11.5B (est $11.3B) · EPS $1.66 (est $1.62) · Next-quarter revenue outlook $13.0B
Key quotes
"At our advancing AI event, we launched Helios, our rackscale AI platform combining EPYC Venice CPUs, MI450-Series GPUs, Pensando networking and ROCm software."
"Our sixth-gen EPYC Venice family is purpose-built for this expanding range of workloads and delivers one of the largest generational performance gains in EPYC history."
"We delivered another outstanding quarter with record revenue and profitability as adoption of our leadership products continued to expand."
Analyst Q&A 10
Data center GPU vs CPU growth
Q (Thomas O'Malley · Barclays): When will data center GPU business cross over CPU business and which will contribute more dollars into calendar year 2027?
A: Management said both data center AI accelerators and server CPUs will grow a lot. They see server CPU growing over 80% in the second half year-over-year, and data center AI TAM is larger with Helios ramping in 2027 driving substantial growth. Jean said gross margin is driven by business mix, they made progress in H1 2026 and Q3 guide, and Embedded recovery should help gross margin in 2027.
"So look, we are certainly seeing a very, very strong compute market demand over both the data center accelerator as well as the server CPU." — Lisa Su (CEO·CHAIR)
Which grows more Q3: CPU or GPU?
Q (Timothy Arcuri · UBS): Which part of data center drives the growth from Q2 to Q3 — CPU or GPU? And can AMD grow revenue faster than the 40% TAM CAGR through 2030?
A: They said server CPU will grow substantially in Q3 and Q4. The Q3 segment is guided to grow sequentially double digits, both server and data center AI will grow, and Helios ramp just starts end of Q3 and steps up in Q4. They also said they expect to grow faster than the TAM and that the TAM view is greater than 40% and AMD will grow greater than that on overall revenue.
"So we would -- we're guiding the Q3 segment to grow sequentially double digits." — Lisa Su (CEO·CHAIR)
Gigawatt line-of-sight and HBM costs
Q (Vivek Arya · BofA): How many gigawatts of compute does AMD see for 2027 and what revenue per gigawatt should we expect? Also, does 50% more HBM expose AMD to memory cost inflation and margin risk?
A: They said the server CPU portion of data center will grow by over 70% year-over-year and the overall segment will grow by over 100% in 2027. They have strategic customers OpenAI, Meta, Anthropic ramping and said they have supply to more than meet guidance; Helios begins in Q3 with step-ups in Q4 and Q1 and will ramp through 2027. Revenue per gigawatt is in the "double-digit billions" range. On HBM, they have good visibility into HBM allocation for 2027, see memory benefits as workload-dependent, and can modify memory footprint if customers need.
"The way I would describe it is we have supply to more than meet the guidance that we've talked about and the upside potential is there." — Lisa Su (CEO·CHAIR)
Server CPU units, ASPs and Helios ramp
Q (Joshua Buchalter · TD Cowen): What unit and ASP assumptions are baked into server CPU guidance, and when you said Helios is ahead of forecast, was that volumes or yields? How will Helios gross margins evolve as it ramps?
A: They said Q2 saw double-digit growth in both units and ASPs, with unit growth higher, and expect both units and ASPs to keep growing as they add capacity across wafers and backend. On Helios, being ahead of forecast referred to overall volumes and demand; yields should improve over the next few quarters as the Q3 ramp continues.
"So Josh, when I was talking about Helios was ahead of our initial forecast, it was as it relates to the overall volumes." — Lisa Su (CEO·CHAIR)
Size of sandbox (agentic) servers
Q (Aaron Rakers · Wells Fargo): How large is the middle "sandbox" or agentic AI server bucket in the $220 billion TAM, and will client (PC) CPU grow in 2026 and into 2027?
A: They said when thinking about the $220 billion TAM in 2030, agentic AI sandboxes are expected to be the largest and fastest-growing piece, though smallest today, with that expansion occurring over the next 3, 4, 5 years. On client CPUs, first half 2026 was strong, they expect 2026 year-over-year growth, and they remain optimistic for 2027 with new AI-centric PC products and rising enterprise content.
"when you think about the $220 billion TAM in 2030, we actually see this agentic AI or these agentic sandboxes being the largest piece of the TAM." — Lisa Su (CEO·CHAIR)
What does "acceleration in H2" mean?
Q (Stacy Rasgon · Bernstein): The press release says data center accelerates in the second half while Q2 grew 107% and Q3 guide is lower; how should we interpret 'acceleration'? And what does 'more' mean in server >70% and total doubling?
A: Jean said "acceleration" means second half versus first half growth. Lisa clarified the server CPU is expected to grow more than 70% and the overall data center business is expected to be well over 100%, driven by strong data center AI strength and the Helios ramp.
"And we do expect the overall data center business to be well over 100%." — Lisa Su (CEO·CHAIR)
Customer diversity and readiness
Q (James Schneider · Goldman Sachs): How many customers will contribute to MI400/Helios in Q4 and early '27, and are there data center readiness constraints (land, power, shell) that could stop you from hitting targets?
A: They said large frontier model companies like OpenAI, Anthropic, Meta will consume through multiple cloud providers and there are additional smaller customers, so diversity should be good going into Q4 and Q1. They said they do not see anything that would give them pause about hitting targets, though the range depends on bringing on more capacity and they are building the whole supply chain and working with operators.
"I don't think we see anything that would give us pause about hitting the targets." — Lisa Su (CEO·CHAIR)
Instinct revenue cadence and gross margins
Q (Christopher Muse · Cantor): If data center guide implies Instinct revenues around $30 billion, is that too low? And how should we think about Instinct gross margins beginning 2027 and exiting 2027?
A: Lisa said the $30 billion estimate is probably too low and the data center AI number should be well over 100%, with Helios growing quarter-by-quarter. Jean said 2027 gross margin will be determined by the pace of the server CPU ramp and the data center AI ramp; they are optimistic about managing the MI450 ramp and expect server growth to offset AI mix, with other levers like embedded and client improving gross margin as well.
"when we think about the gross margin for 2027, as I said earlier, it's really determined by both the pace of server CPU business ramp and the data center AI ramp." — Jean Hu (CFO)
Server CPU supply constraints
Q (Joseph Moore · Morgan Stanley): Is server CPU supply constrained now, will that tightness persist, and can the supply chain support next year's CPU growth? Also, does ramping on 2-nanometer help vs 3-nanometer tightness?
A: They said the server CPU supply chain is tight right now because much of the demand was unforecasted in the first half of the year. For 2027 they expect demand to be better forecasted and the supply situation to be better than '26, and they feel good about supporting over 70% year-over-year growth. On process nodes, bringing up a new node is challenging, but their chiplet approach uses fewer wafers on the new node, helping ramp, and they are managing the whole supply chain.
"I would say the server CPU supply chain is tight right now, and it has been tight for the first half of the year because much of this demand was unforecasted." — Lisa Su (CEO·CHAIR)
Fast inference growth and OpEx
Q (Atif Malik · Citi): How will sales grow in the fast inference market with the Cerebras partnership, and how should we think about OpEx growth versus the 40% TAM?
A: Lisa said the inference market is growing substantially this year into next and fast inference is becoming more relevant; the Cerebras partnership with Helios Plus gives a good solution expected to start becoming available in Q4 in the Cerebras cloud and extend into 2027. Jean said they will continue to invest OpEx given the opportunity but will manage OpEx increases to be less than top-line revenue growth so they can drive operating leverage and higher EPS, noting EPS will be significantly higher than $20 from their Analyst Day.
"And we would expect that solution to be start becoming available in Q4 in the Cerebras cloud and extend into 2027." — Lisa Su (CEO·CHAIR)