AMD put its whole sixth-generation server CPU line on a public price list on 29 September 2026. Not a partial sheet: 31 EPYC 9006 SKUs, 8 to 256 cores, two sockets, US$700 to US$14,904. The interesting part is not the flagship. It is what the per-core math says about everything else in the catalog.

1. On 29 September, AMD priced 31 server CPUs at once
On 29 September 2026, AMD published complete specifications and per-thousand-unit pricing for the sixth-generation EPYC 9006 series, codenamed Venice. It is the first time AMD has published a full price sheet for server processors covering 8 to 256 cores across two socket platforms: 31 SKUs at once, from US$700 to US$14,904.
The specifications: Zen 6 and Zen 6c cores on TSMC's 2nm process, the first AMD server microarchitecture designed from the ground up for data center workloads. The flagship EPYC 9996 carries 256 cores and 512 threads, 1GB of L3 cache and a 600W Default CPU Power rating. AMD's blog says the top configuration supports 16 channels of DDR5, JEDEC-standard MRDIMM up to 12,800 MT/s and PCIe Gen 6.
One detail is easy to miss. From this generation on, AMD has retired the term TDP in favor of "Default CPU Power," the total power consumed across the compute and I/O dies at the stated performance target. TDP is a design ceiling; Default CPU Power is measured consumption, and it is the better number when you size a rack's power and cooling.
2. Two sockets, two different machines
SP7 and SP8 are not premium and budget versions of the same box. They are two platforms for two different jobs:
| Comparison | SP7 (high-end data center) | SP8 (entry / volume) |
|---|---|---|
| SKU count | 9 | 22 |
| Core architecture | Both Zen 6 and Zen 6c | Zen 6c only |
| Core range | 64 cores minimum, up to 256C/512T | 8C/16T minimum, up to 128C/256T |
| Memory channels | 16-channel DDR5, up to ~1.6 TB/s | 8-channel DDR5 |
| PCIe 6.0 lanes | 96 | 128, about a third more than SP7 |
| Power range | 400W to 600W | 130W to 400W |
| Shipping | Q4 2026 | H1 2027 |
Three rows matter directly for purchasing and operations:
- SP8 has more PCIe lanes than SP7. If you need a pile of NICs, NVMe drives or HBA cards, 128 lanes gives you more room. Lane count is not performance, but it caps how many cards you can install.
- SP8 does not ship until H1 2027. Build this quarter and SP7 is the only option.
- SP7 starts at 64 cores. There is no small SP7 this generation, so fewer cores means SP8 or the previous generation.
One more variant: the AMD 9006 LP (Verano) is the first server EPYC with LPDDR5X. It belongs to the Verano platform rather than Venice proper: 72 cores at most, LPDDR5X SOCAMM2 memory, a 112 Gbps xGMI interconnect, second-half 2027. It targets AI rack nodes, the same logic as NVIDIA's Grace/Vera decision to put memory next to the processor.


3. What the price list reveals, and one trap to avoid
The flagship 9996, at 256 cores and US$14,904, works out to about US$58.2 per core. The entry SP7 part, the 64-core 9556, works out to about US$125 per core. More cores means a lower cost per core, a clear nudge toward large customers buying big configurations.
The telling comparison is with the last generation. Turin launched in October 2024 with the 192-core EPYC 9965 at US$14,813. Venice's 9996 adds 64 cores for US$91 more: 64 extra cores for a 0.6% price increase.
One trap, stated plainly: US$14,904 is AMD's price at a 1,000-unit volume, an OEM purchasing basis, not what one customer pays. TechRadar makes the same point. Actual prices depend on order size, discounts and negotiation. Use the figure to compare per-core structure across SKUs, never to budget a single machine.
Performance claims need the same care. AMD's blog claims a 174% uplift (2.74x) against Intel's Xeon 6980P in the CPU-side stages of an agentic AI pipeline. AMD's footnote is explicit: this is internal testing as of 11 July 2026, covering only the CPU infrastructure layers, not GPU inference. That qualifier must travel with the number. On the same basis, AWS Graviton5 scores 1.08x.
Mercury Research puts AMD's server CPU revenue share at roughly 46%. Read alongside the price list, that is a confident position.

4. How much supply is there? Reading the "2027 sold out" claim
Around 30 September, leaker Jukan posted on X that Venice's 2027 allocation is fully committed and AMD is taking orders for 2028. Several outlets relayed it, citing Wccftech and unnamed channel sources.
Channel reports stay unconfirmed until an official announcement, and AMD has not verified this one. Jukan said he doubted the numbers himself, worried about last year's oversubscription-then-cancellation pattern. Treat it as a demand signal worth watching, not a fact.
The grounded figure comes from Morgan Stanley's June 2026 report: about 6.75 million Venice units shipping in 2027, ahead of NVIDIA's Vera CPU at 5.75 million. At the 31-SKU average of US$7,691, that implies roughly US$51 billion in revenue. Still a forecast.
5. What this means for a small business in Guangzhou
A 256-core flagship is a long way from the 8-to-32-core servers most small companies run. Three layers are still useful:
- Knowing the next generation helps you time a purchase. SP7 ships in Q4 2026 and SP8 in H1 2027. If you plan a refresh next year, map SP8's per-core curve now instead of accepting launch pricing later.
- The per-core gap is the number mid-sized buyers should watch. US$58 against US$125 means the 64-core 9556 costs more than twice as much per core as the 256-core 9996. For databases and virtualization that never fill 256 cores, that matters more than peak performance.
- PCIe 6.0 and MRDIMM are buy-early, regret-early technology. TechRadar does not expect PCIe 6.0 in ordinary PCs until around 2030, and MRDIMM, a registered module built for AI memory bandwidth, may never get there. Unless AI training or inference is your workload and memory bandwidth is already the bottleneck, do not pay for either.
One more signal: TechRadar reports that Intel's 128-core Xeon 6980P launched at US$17,800 in 2024, more than AMD's priciest part costs today. AI demand has pushed server CPU pricing toward selling cores and bandwidth rather than percentage gains.
Specifications and performance data here were checked against AMD's blog (23 July 2026) and reporting from 29 September to 1 October 2026. All prices are AMD's per-thousand-unit figures. Sold-out claims and shipment forecasts are predictions, not AMD-confirmed. Base any purchase on the quote you receive and on your own workload. Putting a flagship CPU on a light database load is the most common waste in server buying.

