Intel's Gelsinger grades his chip flip a hit, but AMD exec thinks it's more silicon slip Intel CEO Pat Gelsinger might be giving himself a "passing grade" in his turnaround efforts since moving into the top office at the chipmaker, but an exec at arch-rival AMD isn't being nearly so generous. Gelsinger returned to Intel in March 2021 and has since tried to reignite revenue growth and regain technology leadership. …
AMD was fabbing their latest [released] chips on TSMC's 5nm process.
Intel have just rolled out their 7nm process (Intel 4) which they claim has performance circa that of a 4nm process. If true, it is going to be better than the existing generation of AMD chips.
However, TSMC's roadmap was 3nm in December last year, of which Apple brought the entire production run, and 3nmE(nhanced) starting volume deliveries in this quarter. Test examples will have been around for a while, so AMD are probably now sitting on an increasing pile of new 3nm(enhanced) chips and waiting for the optimal point to roll them out, which i'd imagine would be about a month after Intel releases their chips in quantity as the press furore over "New! Shiny! Intel takes the Performance lead from AMD!" starts tailing off and gets replaced with "New! Shiny! AMD retakes the performance lead from Intel!". At least, that's what i'd do if I was running AMD.
If you compare TSMC's process map to Intel's then AMD can be expected to be in the lead for a couple of years. TSMC's next process node is 2nm, with intel rolling out a 3nm process next year which will at best match TSMC's 3nm process briefly, before losing the process advantage to the TSMC 2nm process, and they might deliver their response to that around the same time it's deployed. In theory Intel then expects to leapfrog TSMC's processes after this.
However, TSMC has a very good historic track record on underpromising and overdelivering, and Intel has a very good historic track record of overpromising and underdelivering. We'll see in ~2026.
You’re assuming that TSMC’s roadmap is AMD’s roadmap. Looking at how N5 went, AMD isn’t going to N3 any time soon - leaving aside that Apple pre-booked a whole year of production, AMD tends to wait until the process has a mature high-performance version and they can get it at a good price (i.e. high yield). On that pattern, expect Apple and the other mobile SoC vendors to gobble up the N3 capacity for the next two years, and AMD to move over to it only when TSMC intros N2 to keep the mobile makers happy in 2025.
Yes, 2025. N2 is not coming until 2025. TSMC has been warning customers and the industry at large that the days of a new node every 18 months are over, but the message is taking a while to sink in. It recently announced an extension of its existing processes to 2.5 years, and the delivery date for N2 is now set by TSMC as late 2025. That’s still a good schedule, but Intel’s competing 20A can now overtake it, having been pulled back in by Intel by about six months to mid-2024 (it had slipped out to 2025 from an original early-2024 timeline). The fact that Intel has publicly brought the date for 20A forward again suggests that they have very high confidence in bringing this process to market.
Intel had an almighty clusterfuck on getting “7nm” (now “Intel 4”) up and running, but from what I’ve heard about it, the issues were organisational and financial rather than a lack of technical ability, and the delay has created a backlog of technologies that were blocked by an inability to fabricate them at scale. That fabrication blockage is now clearing, and clearing quicky, so it’s very likely that Intel will meet its roadmap targets - at least up to 18A, after which we’re into another step-change of technology and the game resets again.
I'm not an expert in either finance or fabrication so I may be barking up the wrong tree, but to me, outsourcing a key part of your business like the actual manufacturing sounds like a great idea until it isn't - what if the fab partner pulls a Unity and alters the deal, or they have a catastrophic event, and have to decide which chips to prioritise building?
Fabless has been the only practical route to *survival* for most companies. Not success, survival.
The problem is the semiconductor business cycle, and the capital requirements. Basically, if you go back to the sixties/seventies, there were a dozen fab companies. On each process generation, there’s an early start of the first victor, then the others come in, and eventually there is massive overcapacity. Each time the downturn hits, you end up being forced to sell at a massive loss, losing billions. And you have to continue to annually invest was-billions-then-tens-billions-now-hundred-billion. Because if you don’t, then when the upturn happens it’s just too late.
The overcapacity downturn literally cannot stop, cannot turn into profit, until the one or two weakest fab company runs out of money and goes bankrupt. It’s not technical, or even sales, it’s pure deepest pocket. 12 companies became 10 after an 8-year cycle. Then 8…. 7….6, 5,4,3. Tick, tock. Now we are down to TSMC, Samsung, Intel. That’s it, on leading edge. GloFo still exists *because* it saw the writing on the wall, and got off the train to service the industrial 45nm segment. AMD still exists *because* it went fabless.
This time round, there are three left. In 3-5 years time, there will be just two companies in the next long-lasting node at 3nm. There is simply no universe in which that isn’t true, there can be no profitable semiconductor industry until one of them is removed from the board. And there is no universe in which it is TSMC that goes bankrupt. Which means, on Intel current strategy of betting the house, in five years time *either Samsung or Intel will be bankrupt*. Nobody outside the industry sees it, it’s just unthinkable. But inside the CPU industry, we all know what’s coming.
So you gotta ask yourself….do you really believe in a world where it is Samsung that will disappear in just 3-5 years? Bankrupt means bankrupt. Samsung can funnel funds from smartphones, TVs, fridges, laptops, ships, wind turbines. That’s all going to be gone? Or will it be Intel?
Boss of AMD is correct. Intel are picking a fight it can’t win.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
