AMD EPYC and the Server CPU Business: Size and Market Share

Every cloud service, banking app and online shop runs on servers, the powerful computers that fill data centers. Each server has a main processor that runs its operating system and hands the heaviest math to faster specialist chips. For the first years of the AI boom, most of the attention went to those specialist chips, because they train and run AI models. Then the main processor came back into the story. Chipmakers began telling investors that AI agents, programs that work through tasks step by step, lean on it heavily.

Sizing a business built on those main processors is harder than it looks. One research report can give a chipmaker about a quarter of the market while another gives it nearly half. The company itself may publish no sales figure for its server chips at all.

So a server chip business is sized from three kinds of evidence. There is what the company reports in its filings, where sales are grouped into segments, the business units it gives results for. There is what its managers tell investors on calls, and there is what outside research firms estimate about market share. Each one measures something different, and the numbers only line up once you know what each one counts.

AMD's server processors carry the EPYC brand and use x86, the instruction set, or basic command language, that AMD's and Intel's chips share. On BloFin, the AMD perpetual is a futures contract on AMD's share price, so news from the EPYC business reaches a position there only by moving that price.

What is AMD EPYC?

AMD EPYC is AMD's line of x86 server processors: central processing units (CPUs) built for large data centers, company IT and supercomputers (source: AMD 10-K, 2025). The first EPYC chips launched on June 20, 2017 (source: AMD release, June 2017). Their sales sit inside AMD's Data Center segment, with no EPYC revenue line (source: AMD 10-Q, Q2 2026).

A central processing unit handles the general work of a computer: it runs the operating system and most of the programs. A server CPU does that work for the machines in a data center. Because AMD's and Intel's server and PC chips share the x86 instruction set, software written for x86 can run on chips from either company.

Each EPYC chip holds many cores. A core is a processor inside the chip that works through its own stream of tasks. More cores let one server handle more work at the same time. Every one of AMD's EPYC launch releases gives the top core count, which makes it the simplest way to compare one generation with the next.

The other main chip in AI is the graphics processing unit (GPU). It is an accelerator built to run huge amounts of math in parallel, and it does most of the work of training and running AI models. An AI server still needs CPUs to run it. AMD said in May 2026 that the CPU was becoming even more important as AI moves from training models toward running agents (source: AMD release, May 2026). In AMD's words, the CPU coordinates data movement, networking, storage, security and the running of the whole data center (source: AMD release, May 2026). The two kinds of chip split the work of an AI data center like this:

In AMD's reports, EPYC belongs to the Data Center segment. The quarterly filing defines that segment as AI accelerators, server CPUs, GPUs and a range of networking and other chips for data centers (source: AMD 10-Q, Q2 2026). EPYC shares it with Instinct, AMD's GPUs for AI, which drive AMD's AI and data-center story. EPYC has existed for nine years, and each generation has raised what a single chip can do.

AMD EPYC generations from Naples to Venice

AMD has launched six generations of EPYC server processors in nine years, from the 32-core Naples in June 2017 to the 256-core Venice in July 2026 (source: AMD release, July 2026). Based on AMD's figures, the top core count rose eightfold across the six, and a new generation arrived every 19 to 26 months.

Zen is AMD's name for its CPU core design, and each EPYC generation has moved to a newer version of it. AMD's launch releases give each generation's date, core design and top core count, except Venice's Zen 6 core, which AMD chief executive Lisa Su named on AMD's August 4, 2026 earnings call:

The jumps were uneven. A process node names the factory generation a chip comes from, and a smaller number of nanometers (nm) means a newer, denser process. The 2nd Gen doubled the top count to 64 cores in August 2019 and was made on a 7nm process (source: AMD release, August 2019). The 3rd Gen arrived on March 15, 2021 with the same 64-core top, now on the Zen 3 design (source: AMD release, March 2021). Genoa, the 4th Gen, reached general availability on November 10, 2022 with up to 96 cores (source: AMD release, November 2022). Its Bergamo chips followed on June 13, 2023 with 128 Zen 4c cores each (source: AMD release, June 2023). Turin, the 5th Gen, became available on October 10, 2024 with 8 to 192 cores (source: AMD release, October 2024).

AMD's releases name a 7nm process for the 2nd Gen, 3 to 4nm for the 5th Gen and 2nm for Venice. The 5th Gen launch release says those chips use "advanced 3-4nm technology" (source: AMD release, October 2024). AMD gave Venice's 2nm process when it announced the production ramp in May 2026 (source: AMD release, May 2026).

Venice, AMD's 6th Gen EPYC

Venice is AMD's 6th Gen EPYC processor family, built on the new Zen 6 core and a 2nm manufacturing process (source: AMD Q2 2026 call). AMD announced on May 21, 2026 that Venice was ramping production at contract chipmaker TSMC in Taiwan (source: AMD release, May 2026). Ramping production is the step from early samples to steady factory output. AMD launched the family on July 23, 2026 at its Advancing AI event (source: AMD release, July 2026).

The top model, the EPYC 9996, has 256 cores and 512 threads. A thread is a separate stream of instructions that a core works on, and here each core runs two at once (source: AMD release, July 2026).

Lisa Su told AMD's earnings call on August 4, 2026 that Venice was in production now and that the family has more than 30 processors. She said the leading cloud providers were planning deployments beginning later in 2026. Demand for Venice, she added, was "stronger than for any prior EPYC generation" (source: AMD Q2 2026 call). Dan McNamara runs AMD's compute and enterprise AI business. On September 11 he called Venice AMD's broadest launch yet, with server makers, cloud vendors and software firms backing it from day one (source: Goldman Sachs conference transcript).

A rack is the tall frame that holds a group of servers in a data center. Venice also goes into AMD's Helios racks, with 18 Venice CPUs in each (source: AMD release, July 2026). At launch, Meta was validating, or testing, 6th Gen EPYC platforms in its own labs (source: AMD release, July 2026).

How big is AMD's server CPU business?

AMD reports its server CPU sales inside its Data Center segment, alongside Instinct GPUs and networking chips. That segment's revenue was $6.7 billion for the quarter to June 27, 2026 (source: AMD 10-Q, Q2 2026). On September 8, 2026, chief financial officer Jean Hu called server CPUs the largest part of that business (source: Citi conference transcript).

EPYC has two main groups of buyers. Cloud providers rent out computing power over the internet. Su named AWS, Microsoft, Google and Oracle among those using more EPYC, both in their own systems and in the services they rent out (source: AMD Q2 2026 call). Enterprises, meaning ordinary companies, run servers in their own buildings. There AMD reported a fourth straight quarter of record sell-through, which means sales by resellers to the final buyer. Su said cloud and enterprise sales each grew by more than 70% year over year (source: AMD Q2 2026 call). These cloud firms and companies are the EPYC side of AMD's customers, who also include the buyers of its Instinct GPUs.

The quarterly filing names strong demand for EPYC processors and Instinct MI350 GPUs as the main driver of the segment's growth (source: AMD 10-Q, Q2 2026). Data Center is one of AMD's business segments, and the filing reports its revenue and operating income as single figures covering EPYC and Instinct together. What AMD has disclosed about EPYC, and where:

Put together, AMD's own words give EPYC's growth rate, its records and its weight inside Data Center, but no dollar figure, so any EPYC revenue number you meet elsewhere is an outside estimate.

AMD's server CPU outlook for 2026 and 2027

On its August 4, 2026 earnings call, AMD said it expected server CPU revenue to rise by more than 80% year over year for July to December 2026 (source: AMD Q2 2026 call). For 2027 as a whole, it expected growth of more than 70% (source: AMD Q2 2026 call). That is guidance, a company's own forecast of its results, and it came from Lisa Su.

Su also said server CPU supply had been tight in the first half of 2026 because much of the demand was unforecasted. In plain terms, customers wanted more chips than had been planned for. A supply constraint like that caps how fast sales can grow, whatever the demand. She said the 2027 supply situation should be better than 2026 (source: AMD Q2 2026 call). Mercury Research, a firm that tracks chip shipments, said server CPU supply constraints could limit gains in the third quarter of 2026 (source: Tom's Hardware, August 2026).

AMD also sizes the whole server CPU market, meaning the total that every maker's server CPUs could sell for in a year. Jean Hu recalled on September 8, 2026 that AMD's November 2025 analyst day had put that market at $60 billion in 2030 (source: Citi conference transcript). On August 4, Su had given a much larger figure: growth of more than 50% a year, to about $220 billion by 2030 (source: AMD Q2 2026 call). Hu said the largest part would be the agentic AI sandbox, servers that run many AI agents side by side. She called it very small today but said it would be more than half of the $220 billion (source: Citi conference transcript). Matt Ramsay heads financial strategy and investor relations at AMD. At the same event, he said AMD intends to win half of that dollar market across every instruction set, a goal he described as building a "$100 billion server business" (source: Citi conference transcript).

Each of these figures is AMD's own view on the date given, and a later call can change any of them.

AMD's server CPU market share against Intel

AMD's share of x86 server processor shipments was 34.5% in the second quarter of 2026, compared with 27.3% a year before, Mercury Research estimates (source: Tom's Hardware, August 2026). By revenue, Mercury put AMD at a record 46.2% of x86 server CPU revenue in the first quarter of 2026, on 33.2% of units (source: Tom's Hardware, May 2026).

Market share is a company's slice of all the sales in a market, and it can be counted in two ways. Unit share counts chips shipped, so a cheap chip and a costly one count the same. Revenue share counts the money buyers paid, so pricier chips weigh more. Research firms also draw the edges of the market differently, which moves the figure again.

AMD's own statements on share give a direction only. Su said on August 4 that AMD had gained x86 server revenue share over the past year (source: AMD Q2 2026 call). Intel sells its server processors under the Xeon brand, and some counts also include chips built on designs from Arm, which uses its own instruction set. The same company can show very different shares depending on what is counted.

Mercury's broad count also takes in chips aimed at networking and storage gear (source: The Register). Counting only EPYC and Intel's Xeon SP server chips lifts AMD's unit share to 46.4% in the second quarter of 2026 (source: The Register). UBS, an investment bank, counts every server processor, including chips built on Arm designs. On that wider basis its figure for AMD's unit share in the first quarter of 2026 is lower, at 27.4% (source: Yahoo Finance). These are the 2026 estimates for AMD, each with its basis and quarter:

Each figure is right for what it counts, so compare two of them only when they share a basis and a quarter.

Why AMD's revenue share is higher than its unit share

AMD's revenue share runs ahead of its unit share because the average EPYC chip sells for more than the average Intel server chip. The average selling price is the average amount a company receives for each chip it sells. Tom's Hardware put the gap down to AMD's higher average selling prices and the popularity of its high-core-count chips (source: Tom's Hardware, May 2026).

You can work out the size of that price gap from Mercury's first-quarter figures. Say the x86 server market sold 100 chips for $100 in total in the first quarter of 2026. AMD's share, 33.2 chips, brought in $46.20, or about $1.39 a chip. Intel's 66.8 chips, the rest of the market, brought in the other $53.80, or about $0.81 a chip. Divide $1.39 by $0.81, and AMD's average x86 server chip sold for about 1.7 times as much as Intel's, worked out from Mercury's shares (source: Tom's Hardware, May 2026).

Jean Hu said on September 8, 2026 that AMD's CPU growth came from both more units and a higher average selling price (source: Citi conference transcript). The 1.7 times figure is a single quarter's reading, and the next Mercury report can move it in either direction. The gap also shows why a share figure needs its basis beside it. Read alone, 33.2% and 46.2% look like figures for two different companies, yet both describe AMD's x86 server chips in the same quarter.

Intel's Xeon answer to EPYC

Intel still ships most x86 server processors, 65.5% of units in the second quarter of 2026 by Mercury Research's count (source: Tom's Hardware, August 2026). On July 23, 2026, Intel reported its strongest server growth on record and the launch of Xeon 6+, its first server chip made on Intel's 18A manufacturing process (source: Intel Q2 2026 call).

Xeon is the direct rival to EPYC, and it has been losing ground: Intel's 65.5% compares with 72.7% a year earlier (source: Tom's Hardware, August 2026). By Mercury's count, then, Intel gave up about 7 percentage points of unit share in a year, worked out from those two figures.

Intel's managers described their server CPU business on the July 23, 2026 call in these terms:

Intel has also reshaped its plans. Memory channels are the paths between a processor and its memory, and more of them let a chip feed its cores faster. On its January 22, 2026 call, Intel said it had simplified its server roadmap to focus on Diamond Rapids, a 16-channel server processor (source: Intel Q4 2025 call). On the January call, Intel also said it planned to bring forward Coral Rapids, another processor on that roadmap, where it could (source: Intel Q4 2025 call). On the July call, Tan pointed to the server chips code-named Clearwater Forest, Diamond Rapids and Coral Rapids as Intel keeps working to compete better with its rival (source: Intel Q2 2026 call).

For investors, Intel's server CPU results are one input when they weigh AMD versus Intel stock, alongside everything else each company sells.

Arm and custom chips in the server CPU market

Arm-based chips are the third force in server CPUs. Mercury Research estimated Arm's share of server processors at a record 13.6% in the second quarter of 2026 (source: The Register). In March 2026 Arm announced a server chip of its own, the AGI CPU, with Meta as its lead partner and customer (source: Arm).

Arm is a chip design company whose instruction set is an alternative to x86. Many companies license Arm's designs to build their own chips, and on March 24, 2026 Arm added a chip it sells itself. The AGI CPU is the first product in Arm's new line of data center chips, and it was available to order the same day (source: Arm). Meta is co-developing it for its own apps, and the other launch partners include OpenAI, Cloudflare and SAP (source: Arm).

Some server buyers design their own processors as well. AMD's annual report warns that some customers are building their own data center chips, which could affect the market open to AMD's server processors (source: AMD 10-K, 2025). Google's Axion, a processor built on designs licensed from Arm, is one example.

Arm's share also depends on who is counting, and Mercury's and UBS's estimates differ in basis and quarter:

AMD frames the contest as one about chips, whatever instruction set they use. Matt Ramsay said on September 8 that AMD's approach is to build the best server parts, period (source: Citi conference transcript). Dan McNamara said on September 11 that x86 and Arm have no fundamental differences as instruction sets, though lots of software is built for x86 (source: Goldman Sachs conference transcript).

Why server CPU stocks rallied in September 2026

Chip stocks tied to server CPUs rallied in September 2026. Quartz linked part of the move to Meta's Muse AI agent, a program that carries out tasks for its user, climbing to first place in Apple's App Store free chart (source: Quartz). On September 21, AMD closed 9.95% higher (source: Stock Analysis).

Quartz, which credited Bloomberg for the chart ranking, said the app's reception raised investor interest in the makers of the chips needed to run agentic AI workloads (source: Quartz). The reason lies in how an agent works. It moves through a task in steps, such as looking up data, running code and calling other software.

Jean Hu described agentic AI as workflow execution, meaning retrieving data, running tasks and orchestrating them, and said all of that is done on CPUs (source: Citi conference transcript). Lisa Su said on August 4 that agentic AI was creating a new growth vector for server CPUs, in other words a new source of demand (source: AMD Q2 2026 call). Intel's David Zinsner said the ratio of CPUs to GPUs was now almost in parity, or close to one for one, and that it could eventually tilt toward more CPUs by unit count (source: Intel Q2 2026 call).

Lisa Su listed three kinds of server CPU work in AI on the August 4 call (source: AMD Q2 2026 call). A head node, the first of them, is the CPU that runs a GPU server and keeps its GPUs supplied with data:

At a Goldman Sachs conference on September 11, 2026, AMD pointed to 96-core high-frequency EPYC chips for head nodes and 256-core chips for agent sandboxes (source: Goldman Sachs conference transcript).

Intel's shares closed 12.14% higher on September 21 (source: Stock Analysis, Intel). Arm's closed 17.16% higher in the same session, which left AMD's gain the smallest of the three (source: Stock Analysis, Arm). That session was one day in AMD's 2026 stock rally, which had AMD at $630.63 at the close on September 25, 2026 (source: Stock Analysis).

Trading AMD's server CPU story on BloFin

BloFin's contract on AMD's share price is the AMDUSDT Perpetual, with leverage up to 20x. It was listed on May 6, 2026, when leverage ran from 1x to 10x (source: BloFin listing notice). EPYC news, from AMD's results to Mercury's share estimates and Intel's results, reaches the contract only through AMD's price.

Perpetual means the futures contract has no set end date, and each AMDUSDT contract equals 0.01 AMD shares. Holding it does not make you an AMD shareholder, so it carries no dividend or voting rights (source: BloFin notice). Its price follows an index built from AMD prices that third-party data vendors supply. If those vendors send nothing, the index keeps its most recent calculated value (source: BloFin listing notice).

Much of the EPYC story arrives on set dates, and some of it can land outside regular trading hours. BloFin may allow just reduce-only orders in its stock futures outside regular weekday trading hours, and on weekends and market holidays (source: BloFin notice). Those orders lower or end a position and do nothing more. During those hours you may be unable to open a position or add to one.

Leverage adds a second risk, because it lets your position be bigger than the money you put up, which is called margin. BloFin may cut a leveraged position in part or liquidate it in full once it hits a set point (source: BloFin liquidation rules). That point comes when your margin, counting any profit or loss, drops to the level of the maintenance margin plus the cost of closing out (source: BloFin liquidation rules). In this rule, maintenance margin stands for the minimum margin BloFin requires for a position to remain open. At high leverage, a sharp move on a big news day can reach it quickly.

Before you trade around EPYC news, put five things in AMD's server CPU business on your calendar:

1. AMD's quarterly results, and whether server CPU revenue keeps to the growth AMD guided for on August 4, 2026.

2. Mercury Research's next quarterly share estimates, as reported by outlets such as Tom's Hardware and The Register.

3. Intel's quarterly results, and what Intel says about Xeon demand and supply.

4. Venice deployments at the large cloud providers, which Su said were planned to begin later in 2026.

5. The hour each item lands, since reduce-only rules may apply to AMDUSDT outside regular weekday trading hours.

Before any EPYC news day, settle your position size, leverage and stop-loss, the first steps when trading AMD with leverage.

Looking to trade the AMDUSDT Perpetual as AMD's server CPU news lands? To get started, you'll need to first create a BloFin account, fund your account with cryptocurrency, and navigate to the AMDUSDT Perpetual trading page. There, keep leverage low enough to ride out a sharp move on results day.

Frequently asked questions

When did AMD launch its first EPYC server processor?

AMD launched its first EPYC server processors, the EPYC 7000 series, on June 20, 2017, in a range from 8 to 32 cores per chip (source: AMD release, June 2017). AMD had first shown the chip under its codename, Naples, on March 7, 2017, as an x86 design built on its Zen core (source: AMD release, March 2017). It unveiled the EPYC brand on May 16, 2017 (source: AMD release, May 2017). At the launch, Lisa Su said AMD was bringing choice and innovation back to the data center (source: AMD release, June 2017).

How many cloud instances run on AMD EPYC processors?

More than 1,600 EPYC public cloud instance types were available around the world by August 4, 2026, Lisa Su said. The 5th Gen EPYC family, Turin, powered nearly one third of them. An instance type is a server setup of a fixed size that a cloud provider rents to its customers. Su also counted more than 230 5th Gen EPYC server platforms on the market from HPE, Dell, Lenovo, Supermicro and other server makers (source: AMD Q2 2026 call).

What comes after Venice on AMD's EPYC roadmap?

After Venice, AMD has named Verano, another 6th Gen EPYC processor, which it says is tuned for performance per dollar per watt (source: AMD release, May 2026). That measure is the work a chip does for each dollar and each watt of power it uses. After that, AMD says EPYC CPUs on its Zen 7 design are coming in 2028, and a CPU called Ravenna, based on Zen 8, in 2030 (source: AMD release, July 2026). On September 8, 2026, Matt Ramsay named Florence as the step between Venice and Ravenna (source: Citi conference transcript).

How does Mercury Research measure AMD's server CPU share?

Mercury Research estimates each chipmaker's share of processor shipments every quarter, and outlets such as Tom's Hardware and The Register report its figures. Mercury says its broad server count puts AMD at a disadvantage, because Intel now reports its edge and networking chips inside its data center business (source: Tom's Hardware, August 2026). On that view, AMD's real unit share should be higher (source: Tom's Hardware, August 2026). Its Arm estimates carry significant uncertainty, since nobody collects Arm server revenue figures in one central place (source: The Register).

Is AMD raising prices on its EPYC server processors?

AMD has linked its server CPU prices to its own costs. Chief financial officer Jean Hu spoke at a Citi investor conference on September 8, 2026. She said that if wafer prices go up, customers should expect AMD to raise CPU prices as well. A wafer is the thin disc of silicon that a chip factory turns into many chips, so its price feeds straight into what each processor costs to make. Hu's remark describes how AMD would respond to a cost increase, with no specific price change or date attached (source: Citi conference transcript).

Researched and written by the BloFin Academy editorial team with AI-assisted drafting. All facts independently verified. Primary sources include AMD's EPYC launch releases from 2017 to 2026, its annual report for fiscal 2025, its quarterly report for the three months to June 27, 2026, and transcripts of its August 4, 2026 earnings call and two September 2026 investor conferences. They also include Intel's earnings calls of January and July 2026, Arm's announcement of its AGI CPU, Mercury Research data as reported by Tom's Hardware and The Register, UBS data as reported by Yahoo Finance, Quartz, Stock Analysis price histories, and BloFin's AMDUSDT listing notice, stock futures notice and liquidation rules, current as of September 28, 2026.

Nothing in this article constitutes financial advice, and nothing in it is a recommendation to buy, sell or hold AMD in any form. AMD publishes no revenue figure for EPYC, so every size and share number for it is either AMD's own description or an outside firm's estimate, and a later report can revise it or count it another way. AMD's server CPU outlook is guidance its managers gave on August 4, 2026, and results that fall short of it can arrive outside regular trading hours, when AMDUSDT trading may be restricted to cutting or closing positions. On a leveraged position, a move of that kind can leave only the maintenance margin in place, and BloFin may then reduce the position or liquidate it. Past performance does not indicate future results. Do your own research and consider your risk tolerance before you trade on BloFin.