# AMD Revenue per Gigawatt: What the Hardware Is Worth — retained record

> Retained reddit record from reddit:AMD_Stock.

Canonical: https://highsignal.app/data/records/5deb49d208804224

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AMD_Stock)  published 2026-09-26 13:30 ingested 2026-09-27 02:39

[original source ↗](https://www.reddit.com/r/AMD_Stock/comments/1wqqm3d/amd_revenue_per_gigawatt_what_the_hardware_is/)

## retained text



```
Hey everyone,

I run an independent research firm called Northwise, we've been bullish in our AMD coverage since around the 80s not so long ago (congrats everyone) and and we've finished a new AMD model this week.

I wanted to share the operating work here, particularly what changed in our revenue assumptions and how we account for the costs and dilution that come with the growth.

The biggest change is moving away from a fixed revenue-per-gigawatt assumption. We previously used roughly $25B. The new Base forecast starts at $24.92B in 2026 and rises to $38.70B in 2030. 

**What does AMD actually earn from a gigawatt?**

Our calculation is GPUs delivered per gigawatt of IT capacity, multiplied by their selling price, plus attached AMD CPUs and networking. Buildings, cooling, electrical equipment and partner assembly sit outside that revenue calculation. We also distinguish IT capacity from total facility power, which includes cooling and other overhead.

For 2027, we assume approximately 290,000 GPUs per IT gigawatt at an average selling price of $96,371. That produces $27.95B of GPU revenue. Attached AMD processors and networking bring the total to approximately $30.26B.

By 2030, the model assumes fewer GPUs per gigawatt, 260,000, but a higher average selling price of roughly $139,000. The argument is that more capable hardware can earn more per unit of power. The obvious challenge is whether customers receive enough additional useful output to justify paying those prices, especially as competing systems improve.

Getting the revenue timing right matters too. A gigawatt installed in 2027 does not generate another complete hardware sale in 2028 just because it is still running. We model new installations separately from replacement and upgrade purchases. And if a customer uses AMD equipment through a cloud provider, we count that deployment once, rather than counting both companies' announcements as separate demand.

**Where the operating forecast lands**

These are our scenario forecasts:

|2030 forecast, USD billions|Bear|Base|Bull|
|:-|:-|:-|:-|
|Revenue|190.55|387.75|468.97|
|Normalized operating income|50.90|131.33|181.84|
|Free cash flow|44.30|96.42|132.81|

Free cash flow here is after cash capital expenditure, but before acquisitions, financial investments and financing flows. Normalized operating income includes employee stock compensation, while excluding acquired-intangible amortization and specified nonrecurring items. Future customer-warrant accounting marks are not forecast; the ownership dilution is modeled separately.

The Base is an aggressive forecast for those unfamiliar with AMD's real opportunity. It requires approximately 66.6% annual revenue growth from our 2026 estimate through 2030. 

The cases group sixteen combinations of demand, execution, pricing and CPU growth assumptions; they are not a statistical confidence interval. Even Bear describes a much larger AMD, and should not be read as the worst possible outcome but rather an aggregation of downside scenarios.

EPYC is also a substantial independent assumption. Of our $75.12B Base EPYC revenue in 2030, only about $9.19B comes from CPUs attached to the modeled accelerator deployments. The remaining $65.93B requires broader server demand. Success selling GPUs does not, by itself, validate that CPU forecast.

**Three things that can change the shareholder outcome**

* **Cash collection and inventory.** In 2027, Base normalized net income is $26.74B, but free cash flow is $12.40B. Manufacturing commitments, inventory, receivables and capital expenditure absorb cash before the larger earnings figure can accumulate.
* **Customer and employee dilution.** The model ties customer warrant claims to purchases and other disclosed conditions. Base reaches roughly 252 million cumulative customer claims at the end of 2030, with a further allowance for Meta's remaining early-2031 window. Despite approximately $81.5B of modeled repurchases during 2026 through 2030, ending economic diluted shares still rise from about 1.658B to 1.868B. Those future repurchases are our assumption.
* **Pricing versus manufacturing costs.** In a separate sensitivity, GPU selling prices 15% below the reference path, with unit costs unchanged, reduce 2030 free cash flow from $90.13B to $68.59B. Those figures are weighted across all sixteen outcomes, so their starting point differs from the Base column above. Selling more equipment is much less rewarding if customers capture the pricing benefit while AMD still pays the same suppliers.

For those interested [the report is here](https://northwiseproject.com/research/amd-revenue-per-gigawatt). The full financial model is completely public and free. Our price targets, valuations, and the downloadable Excel workbook are reserved for our members, so the link is to my own research business, but the numbers discussed above are all in the free section.

I will be around this weekend if anyone has any questions or disagrees with a part of the model. 
```



document fetched 2026-09-27 00:33 · canonical https://reddit.com/r/AMD_Stock/comments/1wqqm3d/amd_revenue_per_gigawatt_what_the_hardware_is
