Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

AMD's data center revenue nearly doubles to $8.55bn as MI450 orders build

AMD reported third-quarter revenue of $13.38bn on 3 November, with data center sales up 97% to $8.55bn, or 64% of the total.

TechnologyNewsGrace OkonkwoPublished: 4 November 20264 min readSources 28
AMD's data center revenue nearly doubles to $8.55bn as MI450 orders build

For cloud buyers, the practical change is a second accelerator supplier worth benchmarking. AMD's data center unit went from 47% of company sales a year earlier to 64%. Oracle Cloud Infrastructure has committed to deploy AMD's Instinct MI450 at scale from 2026. What nobody has disclosed is what that commitment is worth, or how fast memory supply turns orders into working capacity.

AMD's data center quarter

AMD's July-September revenue came in at $13.38bn, up 45% from $9.25bn a year earlier. Data center sales did the heavy lifting, rising 97% to $8.55bn.

Adjusted earnings per share were $2.09, ahead of the $1.95 analysts polled by LSEG expected. Net income nearly doubled to $2.43bn, and non-GAAP operating income rose 75% to $3.92bn.

The guidance matters more than the quarter. AMD told investors to expect about $17bn in fourth-quarter revenue, plus or minus 3%. That would be roughly 66% growth year over year and 27% from the prior quarter. Chief executive Lisa Su went further, saying AMD expects companywide revenue growth above 80% in 2027, with data center above four-fifths of total sales. That is a claim about a business mix that does not exist yet.

Not everything grew. Client and gaming revenue fell 8% to $3.7bn, and gaming dropped 40% year over year.

Supply is the other half of the guidance. AMD has pledged over $10bn to Taiwanese packaging and substrate firms including ASE and Unimicron through 2029.

Oracle's 50,000-GPU bet

The clearest sign the pitch is landing is Oracle. Oracle Cloud Infrastructure has committed to deploying 50,000 AMD Instinct MI450 GPUs starting in 2026, according to Startup Fortune. The outlet calls it the largest single hyperscaler commitment to AMD silicon to date, and the first time a major cloud provider has offered a publicly accessible AI supercomputer built on AMD chips rather than Nvidia's.

The MI450 is built to cluster 72 chips into one rack-scale system with up to 432GB of HBM4 memory per GPU. Oracle has said it plans to expand those deployments further in 2027. AMD also began volume shipments of its Instinct MI455X accelerators in the quarter, with Anthropic, Meta, Microsoft, OpenAI and Oracle deploying initial Helios systems, Su said.

Nvidia still holds more than 90% of the AI accelerator market, according to industry estimates cited by Startup Fortune. Fifty thousand GPUs does not change that. It does mean every rack Oracle buys from AMD is a rack it is not buying from Nvidia.

The memory squeeze

Memory is the constraint AMD does not control. Apple chief executive John Ternus called the rise in memory chip prices "unprecedented" on a conference call on Monday, and predicted fourth-quarter memory costs would be higher again. Apple shares fell 2.75% in after-hours trading after the company warned the prices would eat into margins.

The picture outside legal channels is starker. Smuggled HBM3E units in China changed hands for about $4,000 each, according to a Shenzhen broker quoted by The Information and relayed by Tom's Hardware and The Decoder. A year earlier the same units went for several hundred dollars. Each AI chip needs six to eight high-end memory chips, so memory alone can cost $24,000 to $32,000 per GPU.

Samsung and SK hynix together control more than 80% of the global HBM market, according to The Decoder.

Enforcement is tightening. South Korea's customs authority had uncovered trade violations worth $560m by May, already above the previous full-year record. A third of that total involved strategic goods such as chips.

Power, politics and the next quarter

Capacity also depends on grid connections. Drax struck a deal with Brookfield and Radiant to develop a datacenter of more than 1 gigawatt at its North Yorkshire power plant site. The project could attract more than £20bn in investment, with first-phase operations from the first quarter of 2028. That first phase would deliver up to 100MW of compute capacity, and Drax hopes to power the site directly by the early 2030s. Drax produces almost 5% of the UK's electricity by burning woody biomass. Analysts cited by The Guardian call it the country's single largest emitter of carbon dioxide.

Politics adds a second brake. Decision Desk HQ projected Democrats would take control of both the House and Senate after Tuesday's elections. In Texas, Democrat Gina Hinojosa defeated Republican governor Greg Abbott after campaigning for a moratorium on new data centers.

Investors are not pricing that in yet. AMD shares closed 3.65% higher at $483.90 on Tuesday, then rose another 7.5% to around $520 in after-hours trading.

Inside the company, the mix is shifting fast. Chief financial officer Jean Hu said AMD's data center AI business was expected to surpass its server CPU business in revenue in the fourth quarter. Su said AMD now expects server revenue to double year over year in 2027 and data center AI revenue to grow faster than it projected 90 days earlier.

Comments 0

Sources

28
  1. 01AMD's Data Center Revenue Nearly Doubles as AI Chip Demand AcceleratesEN
  2. 02Apple warns of 'unprecedented' rise in memory chip pricesEN
  3. 03AI chip smugglers exploit trade control loophole to bring banned HBM from South Korea into ChinaEN
  4. 04UK's most polluting power plant could power country's largest AI datacentreEN
  5. 05Huawei tells customers to find their own AI chip memory, effectively outsourcing procurement to the black marketEN
  6. 06Midterm shock: US-fueled datacenter boom faces political testEN
  7. 07AMD tops third-quarter estimates as AI chip demand lifts data center salesEN
  8. 08Accelerating Plug-And-Play Prototyping In Chiplet-Based DesignsEN
  9. 09Beyond The Chip: How AI Is Reshaping Semiconductor Design And ManufacturingEN
  10. 10Bosch gives chips their own division, eyeing €5bn in sales by 2030EN
  11. 11Is There Room for Small-Scale OSATs in India’s Chip Packaging Boom?EN
  12. 12Accelerating engineering for the physical worldEN
  13. 13Marvell Snaps Photonics Chipmaker Lumentum to Bolster AI ConnectivityEN
  14. 14Global AI Accelerator Market - 2026 Edition: Size, Share, Trends, and Forecast to 2030 | Generative AI Infrastructure ExpandsEN
  15. 15Scientists confirm global warming is accelerating faster than ever recordedEN
  16. 16Li Auto launches 2026 i6 with unchanged starting price, upgraded in-house chip and batteryEN
  17. 17Infineon Rolls Out AI Data-Center Power Chips as Wall Street Splits on the Payoff - ad-hoc-news.deEN
  18. 18China’s CXMT defends bet on risky new DRAM architecture to circumvent US chip curbsEN
  19. 19AI Chip Burn-in Test Market: Reliability Testing for AI Accelerators, GPUs, and High-Performance ChipsEN
  20. 20As AI cyber risks grow, Chinese chip expert urges shared ethics standards with US rivalsEN
  21. 21AI, fuelled by advanced chips, set to power semiconductor manufacturing: industry veteranEN
  22. 22MediaTek Says Its Second AI Chip Customer Is Not Chinese - www.tokenpost.comEN
  23. 23Chinese AI chipmaker Iluvatar CoreX says new GPU matches Nvidia’s H200 as chip gap closesEN
  24. 24FuriosaAI raises $620M for inference chips for data centers - GamesBeatEN
  25. 25Oracle Cloud to Offer Google TPUs and New AI ChipsEN
  26. 26Huawei’s Tau Scaling Law roll-out reaches base stations in blow to US chip curbsEN
  27. 27ByteDance accelerates robotics commercialisation with new recruitment driveEN
  28. 28South Korea urged to guard chip profitability as China’s CXMT and YMTC gain market shareEN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Grace Okonkwo

Grace Okonkwo

AI, models and technology

Grace Okonkwo covers AI, models and technology for FLASH24, working from primary sources such as model cards, API documentation and benchmark papers rather than vendor summaries. She checks training data provenance, evaluation conditions and reported scores against the underlying datasets before any figure reaches print. She interviews researchers and engineers directly, tracks release calendars from major labs, and compares successive model versions on the same tests. Her own self-hosting, home-network and documentation-reading habits feed straight into that desk, since she tests tools on her own hardware first. She does not publish benchmark claims without a reproducible method.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.