Intel's thinking in 2011: Semi revenue required to support one leading edge fab

[Xebec]

I always found this chart interesting from Intel in 2011. At the time it seemed plausible to me as an outsider.

These were still early mobility days (iPad released 2010, iPod still ~28% of Apple revenue*), Intel's own 22nm volume was a year away (Ivy Bridge 2012), and TSMC's 28nm was also about a year away, at least for GPUs.
Then I recall around early 2013, a lot of rumors came out that "Apple bought a chip fab", followed by "Apple leaves Samsung for TSMC" a year later.

With all of the Intel discussion going on, I thought this was relevant because it shows Intel's strategic thinking assumed that TSMC was on the way to bowing out of the leading edge, and that Samsung would be Intel's main competition. (The chart implies that AMD and other Intel competitors would lose access to advanced fabs except through Samsung).

If the concept of this chart is correct, then Apple really is the fulcrum that is playing out between Samsung, Intel, and TSMC.

What would a modern version of this chart look like? How much revenue is required to "support" a new 3nm or beyond fab today?

*iPod numbers: https://www.statista.com/chart/10469/apple-ipod-sales/

 

[tooLongInEDA]

I always found this chart interesting from Intel in 2011. At the time it seemed plausible to me as an outsider.

These were still early mobility days (iPad released 2010, iPod still ~28% of Apple revenue*), Intel's own 22nm volume was a year away (Ivy Bridge 2012), and TSMC's 28nm was also about a year away, at least for GPUs.
Then I recall around early 2013, a lot of rumors came out that "Apple bought a chip fab", followed by "Apple leaves Samsung for TSMC" a year later.

With all of the Intel discussion going on, I thought this was relevant because it shows Intel's strategic thinking assumed that TSMC was on the way to bowing out of the leading edge, and that Samsung would be Intel's main competition. (The chart implies that AMD and other Intel competitors would lose access to advanced fabs except through Samsung).

If the concept of this chart is correct, then Apple really is the fulcrum that is playing out between Samsung, Intel, and TSMC.

What would a modern version of this chart look like? How much revenue is required to "support" a new 3nm or beyond fab today?

View attachment 2277

*iPod numbers: https://www.statista.com/chart/10469/apple-ipod-sales/

The chart was a little misleading even then (this is clearly some time pre-2015).

Samsung and Intel both include significant memory and other business lines at this time. TSMC is pure logic foundry. But then equally the TSMC revenue is split over multiple process nodes and fabs.

But it does accurately reflect the Intel narrative at the time that is was inevitable that only they would survive at the leading edge, this would all result in Intel defeating ARM and dominating mobile, etc.

Looking backwards, we may judge that while they didn't convince all of us (though the story certainly seemed credible at the time), they were far too successful in convincing themselves. And thus ruling out considering other possible futures.

In a similar way, you do seriously have to wonder when Pat Gelsinger came up with the IDM 2.0 strategy just which scenarios his army of Intel MBAs financially modelled for the range of possible outcomes. Because the current outcome (declining sales and margins with delayed IFS ramp) doesn't seem to be one they were at all prepared for. One can argue that it as unlikely seen from 3 years ago. But it clearly wasn't impossible.

 

[MKWVentures]

Intel knew this then and they knew it in 2021. You cannot build a leading edge fab unless you KNOW you can fill it with revenue producing wafers.

Intel hoped it would have revenue to support a new fab in 2025. They do not. You cant tool out a fab that will only have 15-20K wafers per month of revenue best case. This has been true for 50 years. Hence 20A was cancelled. This is why everyone should outsource unless they are growing silicon needs at the rate of 1 fab every 2 years. If you look at scenario spreadsheets, it is obvious what is financially possible and not possible.

Pat thought he could get 2 fabs of foundry in 2026. He currently has 0.25 Fabs best case .... Now we need to see the consequences.

 

[siliconbruh999]

I think if they take a margin loss the can simply cover many Markets if they take loss to their margin big if

Intel knew this then and they knew it in 2021. You cannot build a leading edge fab unless you KNOW you can fill it with revenue producing wafers.

Intel hoped it would have revenue to support a new fab in 2025. They do not. You cant tool out a fab that will only have 15-20K wafers per month of revenue best case. This has been true for 50 years. Hence 20A was cancelled. This is why everyone should outsource unless they are growing silicon needs at the rate of 1 fab every 2 years. If you look at scenario spreadsheets, it is obvious what is financially possible and not possible.

Pat thought he could get 2 fabs of foundry in 2026. He currently has 0.25 Fabs best case .... Now we need to see the consequences.

I think Intel's need can cover 2 fabs fab at max easily it's the extra fabs that are the issue and i bet Germany is getting cancelled Israel cancellation was a good decision