AMD now expects the CPU market to top $200 billion by 2030, and TrendForce forecasts that global server CPU shipments will climb more than 40% in 2027. Both data points landed within ten days of each other in October 2026. They point at the same shift: AI inference and agentic workloads are pulling processors back into the center of data-center spending, after two years in which GPUs took most of the attention.
Lisa Su, AMD’s chair and CEO, told the OCP Global Summit on October 12 that inference now represents about 60% of the market going forward. Ten days later, TrendForce put numbers on the server side. It sees shipments up 31% in 2026 and 43% in 2027, after just 9% growth in 2025. AMD’s EPYC 9006 Venice family, with its 256-core flagship, sits at the center of that story.
This analysis covers what AMD and TrendForce actually said, what the data supports, where Arm-based chips and NVIDIA fit, and what buyers should watch next. Where a number is a forecast, a vendor claim or a single-source report, the text labels it that way. The AMD CPU market forecast is a belief about 2030. It is not revenue, and it is not a guarantee.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
AMD’s CPU Market Forecast: What Su Told the OCP Summit
Su delivered the opening keynote at the 2026 OCP Global Summit on October 12. TrendForce captured her remarks on-site and published its write-up on October 13 under a headline that called 2026 the “Year of Inference” and noted AMD lifting its 2030 CPU forecast to more than $200 billion. The core of her statement was a revision history. AMD has now raised its CPU total addressable market three times in nine months.
Su put it this way, as reported by TrendForce: “We have actually raised the TAM on this three times in the last nine months because we’ve seen that much growth in the overall CPU demand, which we now believe will grow at approximately 50% CAGR over the next 4 or 5 years to over a $200 billion TAM by the time you get to 2030.”
Three details matter here. First, the figure is a total addressable market, so it describes the size of the pie and says nothing about AMD’s slice. Second, the roughly 50% compound growth rate is Su’s stated belief, not a measured result. Third, three upward revisions in nine months tell you the company’s own models keep undershooting demand signals. That can mean real momentum. It can also mean the forecasts are chasing a narrative, and investors should hold both readings at once.
The keynote was not only a forecast. AMD used the same stage to introduce Helion, an agentic-AI rack blueprint, and to confirm that Helios rack-scale systems were moving into production ramp. Those announcements show how AMD intends to turn the market forecast into revenue, and later sections cover them.
Inference at 60%: Why Agentic AI Needs More CPUs
The second Su statement from the keynote explains the first. In the same TrendForce write-up, she said: “This is the year where we believe inference is now about 60% of the market going forward. And it’s not just inference; it’s really the agentic workloads that have really taken off.”
Training a frontier model is a GPU story. Serving it to millions of users, and letting software agents call tools, run code and open sandboxes, spreads work across many more kinds of silicon. An agent that browses, writes a file and executes a script spends much of its wall-clock time in ordinary CPU territory: scheduling, memory management, networking and process isolation. More agents mean more of that glue work, and the glue work lands on server CPUs.
TrendForce’s October 22 analysis makes a similar argument. According to the firm, inference and agentic-AI services are reviving server CPU demand and could alter the traditional balance between CPUs and GPU or ASIC accelerators in data centers. Su added a caution against reading that as a CPU-only future. In remarks TrendForce shared on Substack, she said: “There’s no one size fits all when it comes to compute.”
That line fits AMD’s pitch. The company sells EPYC CPUs, Instinct GPUs, Pensando networking and ROCm software, and it wants cloud buyers to purchase the whole rack. A market in which CPUs, GPUs and networking all grow suits that portfolio better than one in which a single chip type wins. It also explains why AMD’s server push now gets as much airtime as its accelerator roadmap.
TrendForce’s Server CPU Shipments Forecast: 31% in 2026, 43% in 2027
TrendForce published its server CPU release on October 22 and updated it on October 28. The headline says shipments will grow over 40% year over year in 2027. The accompanying TrendForce Insights piece, “The Inference Era Is Also a CPU Story,” gives the precise figures. The table below collects the numbers that matter.
| Metric | Figure | Type | Source |
|---|---|---|---|
| Server CPU shipment growth, 2025 | 9% | Baseline | TrendForce Insights, Oct. 15 |
| Server CPU shipment growth, 2026 | 31% YoY | Forecast | TrendForce, Oct. 22 |
| Server CPU shipment growth, 2027 | 43% YoY | Forecast | TrendForce, Oct. 22 |
| Arm-based share of server CPUs, recent | Above 10% | Estimate | TrendForce, Oct. 22 |
| Arm-based share of server CPUs, 2027 | About 30% | Forecast | TrendForce, Oct. 22 |
| AMD server CPU share, 2026 | About 27% | Estimate | TrendForce, Oct. 22 |
| Cloud provider capex, 2027 | Above $1.5 trillion | Forecast | TrendForce, Oct. 14 |
| AI server shipment growth, 2027 | Approaching 36% | Forecast | TrendForce, Oct. 14 |
Every row labeled Forecast or Estimate is a projection by one research firm. Real shipments can miss in either direction, particularly when supply constraints bite. The shape of the data still carries a message. A segment that grew 9% in 2025 is expected to grow 31% a year later and 43% the year after.
What the 2025 baseline hides
Single-digit growth in 2025 means the server CPU market was a slow, replacement-driven business before agents arrived. That matters for AMD’s TAM math. A jump from 9% to 43% growth is a regime change, and regime changes are exactly when forecasters revise estimates upward several times in a row. Su’s three revisions look less odd against that backdrop.
Reading Two Forecasts Together: Dollars Versus Units
AMD’s number and TrendForce’s numbers measure different things. AMD talks about a market value in dollars, reaching past $200 billion by 2030. TrendForce counts units shipped, with the detailed forecast ending in 2027. You cannot divide one by the other to derive an average selling price, and anyone who does is stacking assumptions on top of two separate datasets.
You can, however, check that the directions agree. Both point up, and both name inference and agentic workloads as the cause. When an AMD executive and an independent research firm reach the same direction from different starting points, the odds that the trend is real improve. The odds that the exact magnitudes are right do not.
The wider spending backdrop supports the direction. TrendForce’s October 14 release projects cloud provider capital expenditure above $1.5 trillion in 2027, with AI server revenue growing over 90%. Shipment growth for AI servers is expected to approach 36%. Revenue growing faster than units implies higher prices per server, which tells you buyers are paying up for dense, expensive systems. Server CPUs ride inside those systems.
For investors and procurement teams, the practical takeaway is to treat the AMD CPU market forecast as a ceiling scenario and TrendForce’s shipment forecast as a base case. Neither is audited, and both depend on hyperscalers continuing to fund AI buildouts at the pace they have set in 2026.
AMD’s EPYC Lineup Today: Genoa, Turin and Venice
TrendForce says Genoa and Turin remain AMD’s primary shipment drivers in 2026. Venice, the sixth-generation EPYC 9006 family built on Zen 6, entered small-scale mass production this year. AMD’s Venice production ramp uses TSMC’s 2-nanometer process. The older platforms still carry the volume, and Venice is the ramp that will matter in 2027.
| Platform | Status as of Oct. 29, 2026 | What the record supports |
|---|---|---|
| EPYC Genoa and Turin | Shipping | Named by TrendForce as AMD’s primary shipment drivers in 2026 |
| EPYC 9006 Venice | Small-scale mass production | 31 SKUs, from the $700 EPYC 9016 to the $14,904 EPYC 9996 at 1,000-unit pricing |
| EPYC Verano | Announced, not released | Shown Oct. 12 as a future 2027 product |
| Helios rack systems | Production ramp | Customer deliveries and workload validation under way |
| Helion agentic-AI rack | Planned for November 2026 | Open blueprint, still in development with partners |
The Venice price list is public. TechSpot’s coverage of the lineup shows the family spanning an 8-core part at $700 to a 256-core, 512-thread flagship at $14,904. AMD reportedly says Venice extends the prior EPYC 9005 generation from as many as 192 cores to 256 and lifts stated memory bandwidth 2.5 times against its chosen comparison. Treat those as vendor claims until independent testing appears. Reports also say the first SP7 systems arrive in the fourth quarter of 2026 and the SP8 platform follows in the first half of 2027.
TrendForce also says AMD plans additional launches in the first half of 2027 but does not name the products. That gap is worth remembering. Anyone claiming to know the exact 2027 EPYC roadmap beyond Verano is guessing.
Verano, Helion and Helios: AMD’s Rack-Scale Pivot
On October 12, AMD introduced the Helion agentic-AI rack, an open blueprint based on an Open Compute Project design and powered by AMD compute. Availability is planned for November 2026, and configurations are still under development with ecosystem partners. Helion is meant to support CPU-only hyperscale builds as well as racks that pair EPYC with Instinct GPUs. One report says those GPU configurations use the Instinct MI355X, though that detail comes from a single source.
Helios is further along. AMD said Helios rack-scale systems are entering production ramp, with customer deliveries, deployments and workload validation under way. The system integrates Instinct GPUs, EPYC processors, Pensando networking and ROCm software. TrendForce expects 2027 customer orders to center on NVIDIA’s GB300 and VR200 systems and AMD’s Helios series. For a deeper look at the accelerator side of that contest, see this comparison of Intel’s D-series against the Instinct MI355X.
Verano is the next CPU step. Su revealed it on October 12 as a 2027 product. AMD reportedly describes it as its first EPYC with an LPDDR memory interface, which would target power and bandwidth efficiency. The company has not released it and has not published pricing. Analysts should read the Verano reveal as a roadmap signal, not a purchasing option.
The strategic logic is straightforward. Selling a complete rack carries more revenue per customer than selling a CPU, and it raises the customer’s switching costs. AMD’s own description of its portfolio, spanning CPUs, GPUs, FPGAs, adaptive SoCs, NPUs, networking IP, rack-scale systems and ROCm, shows a company positioning itself as a full-stack AI infrastructure supplier.
Meta and Oracle: Two Customers Shaping AMD’s Demand
Two October 27 announcements show where the demand is coming from. AMD and Meta announced an expanded strategic partnership covering EPYC CPUs, Instinct GPUs, Helios rack-scale systems, networking and software. AMD said Instinct GPUs already run in Meta production infrastructure and that Meta will deploy a custom Instinct GPU tuned for its workloads. The announcement also says AMD will invest $1 billion in Meta’s equity offering, an unusual step that ties the supplier’s balance sheet to a customer’s financing.
The same day, AMD and Oracle expanded their collaboration. Oracle said 6th Gen AMD EPYC processors will power its upcoming OCI Compute E7 instances, and AMD Pensando technology will support Oracle’s Acceleron converged NICs. Oracle’s choice is a useful data point because it shows a hyperscale cloud adopting the new EPYC generation for general instance types, not only for AI-specific clusters.
Beyond that, AMD has other public ties worth tracking. One report says Microsoft announced plans in July to deploy next-generation AMD Instinct and EPYC processors under an expanded long-term partnership. Another says AMD plans to acquire World Labs, announced September 28. Both items come from single-source reporting, so confirm them against AMD’s newsroom before relying on them.
Arm Closes In: Cloud Chips, NVIDIA Vera and the Arm AGI CPU
TrendForce’s release carries a warning for x86 vendors. Arm-based processors recently passed 10% of server CPU shipments, and the firm expects them to approach 30% in 2027. It names AWS, Google, Microsoft, NVIDIA, Arm, Huawei, OpenAI and Alibaba as sources of forthcoming or expanding Arm CPU supply. One report puts x86 at nearly 90% of the market in 2025, which would make a 30% Arm share a dramatic reversal in two years.
The Arm AGI CPU
Arm’s own server silicon is the newest entrant. On October 28, Oracle said it plans to bring the Arm AGI CPU to Oracle Cloud Infrastructure in fall 2027, and Meta’s Yosemite YV5 design is incorporating the chip into an OCP rack-scale design. Arm’s materials cite independent Signal65 testing that found up to 45% lower sandbox creation latency compared with AMD’s EPYC 9965. That is a vendor-published result, and it tests one narrow task. Agent sandboxes are exactly the workload Su called out, so the claim aims directly at AMD’s growth story.
NVIDIA’s Grace and Vera
TrendForce says demand for NVIDIA’s Grace CPUs is expanding and expects a substantial ramp of Vera CPU shipments in 2027. At OCP, ASUS unveiled the XA P2N-E2 server powered by NVIDIA Vera for agentic AI workloads. No official standalone Vera price or general availability date appears in the sources, so 2027 volume remains a forecast.
Competitive Comparison: AMD, Arm Vendors and NVIDIA in Server CPUs
The table summarizes where each camp stands based on the October reporting. It contains no invented benchmark scores, because the sources do not provide independent head-to-head results.
| Camp | Position | Next milestone | Main risk |
|---|---|---|---|
| AMD EPYC | About 27% server share estimated for 2026 | Additional launches planned for H1 2027 | TSMC capacity limits additional orders |
| Cloud-provider Arm CPUs | Above 10% of shipments recently | Roughly 30% share forecast for 2027 | Designs stay in-house, so merchant sales stay small |
| Arm AGI CPU | In production, per TrendForce | Oracle Cloud Infrastructure in fall 2027 | Vendor-published benchmarks only |
| NVIDIA Grace and Vera | Grace demand expanding | Substantial Vera ramp expected in 2027 | No public price or availability date for Vera |
| Intel | No Intel-specific figure in the TrendForce material cited here | Not covered by the October releases | Share pressure from AMD and Arm |
AMD’s stated advantage is breadth. It sells merchant CPUs to every cloud, while the Arm designs from AWS, Google and Microsoft mostly serve their own platforms. That means the Arm share gain does not translate cleanly into lost AMD sales. Some of it comes at the expense of Intel and older installs, and some of it comes from new demand that did not exist before. For a longer look at the two x86 rivals, see our AMD versus Intel CPU comparison.
One AMD claim deserves a flag. Based on its own June 10 modeling, AMD says its EPYC 9965 delivers 2.37 times the rack-level CPU inference throughput of a cited NVIDIA Vera baseline. That is an AMD model, not an independent test, and Arm’s Signal65 citation pushes the other way on a different workload. The honest summary is that nobody has published a neutral, apples-to-apples agentic benchmark yet.
Supply Limits: TSMC Capacity and Pricing Pressure
Demand forecasts mean little if foundries cannot deliver the wafers. TrendForce estimates AMD’s server CPU share at about 27% in 2026 but warns that TSMC capacity constraints could limit AMD’s ability to accept additional orders. That sentence matters more than the share number. It says the company that sells the chips may be unable to sell all of the chips customers want.
Capacity limits help rivals. If AMD cannot ship extra Venice volume, hyperscalers will lean on their own Arm designs, which sit in separate capacity allocations. They will also stretch older Genoa and Turin deployments. Both outcomes slow the pace at which AMD converts a rising market into rising revenue.
The same squeeze shows up across the hardware market. TrendForce published separate October items on a reported 20% CoWoS price hike at TSMC and on NVIDIA’s memory choices for its Rubin Ultra platform. The details of those stories sit outside this article, but they reinforce the theme: AI buildouts are absorbing leading-edge manufacturing and packaging capacity, and buyers outside the hyperscalers feel it in lead times and prices.
What It Means for Desktop and Laptop Buyers
Most readers do not buy EPYC chips. The server shift still reaches consumer hardware through wafer allocation and through AMD’s roadmap priorities. On AMD’s own product pages, the newest Ryzen 9 X3D desktop chip on sale is the Ryzen 9 9950X3D, and the newest Radeon RX desktop flagship is the Radeon RX 9070 XT. Nothing in the October announcements changes those listings.
For laptops, the competitive picture sits in a separate contest. Our look at Intel Panther Lake against AMD Ryzen AI 400 covers that. Buyers waiting on the next desktop generation should treat reports of a delay with caution, since the coverage of a Zen 6 desktop delay into 2027 rests on a leak and AMD has not confirmed it.
The practical advice is simple. Buy desktop and laptop hardware when you need it, based on current prices and your workload. A booming server market does not make consumer chips cheaper or scarcer by itself, but capacity pressure at the foundry level is one more reason not to assume steep discounts are coming.
Historical Context: From 9% Growth to 43%
AMD announced its fourth-generation EPYC family, codenamed Genoa, in November 2022, according to one report. Turin followed as the EPYC 9005 series on Zen 5. Through 2025, the story in servers was steady share gains by AMD against Intel inside a market that grew slowly. TrendForce puts 2025 shipment growth at 9%.
The agentic-AI boom changed the arithmetic in 2026. TrendForce now expects 31% growth this year and 43% next year. AMD launched the 6th Gen EPYC 9006 family at its Advancing AI 2026 event in July, according to one report, and the company published pricing for all 31 processors shortly afterward. By October, Venice had moved from a roadmap codename to a priced, partly shipping product.
That arc, from slow replacement cycles to forecast-busting growth in about eighteen months, is why Su’s three TAM revisions feel less like salesmanship than like catching up. Readers who follow the broader category can find related coverage in our hardware section.
Five Predictions for the Server CPU Market in 2027
These are analyst judgments based on the October record. They are not facts, and none carries a guarantee.
- Forecasts get revised again. With growth jumping from 9% to a projected 43%, the chance that research firms raise or lower their 2027 figures before year-end is high. Expect at least one more revision from both AMD and TrendForce.
- Helion ships in some form by early 2027. AMD planned November 2026 availability. Blueprint products with ecosystem partners tend to slip a few months, so the first real deployments probably land in the first half of 2027.
- Arm’s share gain reads as a floor, not a ceiling. With Oracle committing to the Arm AGI CPU for fall 2027 and Meta designing it into a rack, the pipeline supports TrendForce’s roughly 30% projection.
- Foundry capacity caps AMD’s upside. If TSMC constraints persist, AMD’s 2027 share gain will come in below what demand alone would suggest, and the gap will go to Arm designs and extended use of older EPYC platforms.
- An independent agentic CPU benchmark arrives. The vendor claims from AMD and Arm conflict. A neutral test suite for agent sandboxes and tool execution will emerge because buyers need one.
FAQ: AMD’s CPU Market Forecast and Server CPU Shipments
What is AMD’s CPU market forecast?
AMD CEO Lisa Su said at the OCP Global Summit on October 12 that AMD expects the CPU market to exceed $200 billion by 2030. She said AMD has raised the figure three times in nine months, and that it expects demand to grow at roughly 50% a year over the next four to five years.
How fast will server CPU shipments grow?
TrendForce forecasts global server CPU shipments to rise 31% year over year in 2026 and 43% in 2027, up from 9% growth in 2025. These are forecasts, not reported shipment results.
Why is inference driving CPU demand?
Su said inference is now about 60% of the market going forward, and she pointed to agentic workloads as a growth driver. Agents spend time on scheduling, sandboxing and tool execution, which run on CPUs rather than accelerators.
What share of the server CPU market does AMD hold?
TrendForce estimates AMD’s server CPU share at about 27% in 2026. That is a research-firm estimate, not an AMD-reported figure, and TrendForce notes TSMC capacity limits could cap AMD’s ability to take more orders.
Is EPYC Venice available yet?
Venice, the 6th Gen EPYC 9006 family, entered small-scale mass production in 2026, and AMD has published prices from $700 to $14,904 for its 31 listed processors. Genoa and Turin still drive most shipments. Reported plans place SP7 systems in the fourth quarter of 2026 and SP8 systems in the first half of 2027.
How big will Arm’s share of server CPUs get?
TrendForce says Arm-based processors recently exceeded 10% of shipments and expects them to reach about 30% in 2027. Growth comes from cloud-provider chips, NVIDIA’s Vera and Arm’s own AGI CPU, among others.
Should I wait to buy a desktop CPU because of this news?
No. The server forecasts do not change consumer pricing directly. AMD’s newest X3D desktop chip on sale is the Ryzen 9 9950X3D, and nothing announced in October alters that lineup.
