What happened to Intel?

Looking at times before Gelsinger as well. Thing that had slipped my notice: Intel invested billions in ASML’s development of the machinery for the current generation of chip manufacture, but then opted to not adopt it…

In a way, it all comes down to reversing one mistake, the ultimate bad bet — the one where Intel funded the technology that its competitors used to leap ahead.

Over a decade ago, Intel spent billions investing in Dutch multinational ASML, which is today the most important company in chips. It’s the only firm in the world that manufactures machines capable of pulverizing a ball of tin, using high-power lasers, such that it emits an extremely tight wavelength of ultraviolet light to efficiently carve circuits into silicon wafers, a process known as EUV.

Intel initially believed in the tech, even carving out a $4.1 billion stake in the company, then decided not to order the pricey machines. But Taiwan’s TSMC did — and went on to become the undisputed leader in silicon manufacturing, producing an estimated 90-plus percent of the world’s “leading-edge logic chips.” Samsung ordered machines, too.

Gelsinger was not shy about calling Intel’s choice “a fundamental mistake” in our 2022 interview. “We were betting against it. How stupid could we be?”

So Gelsinger decided to embrace EUV, while simultaneously giving its technology departments a blank check to leapfrog TSMC. “I said, ‘You have an unlimited budget, and you are going to deliver five nodes in four years. We are going to get back to unquestioned process leadership.’”

Gelsinger wanted to build the capacity to actually produce those chips in volume, too, investing tens and even hundreds of billions of dollars in new factories in the United States and finally offering up its chipmaking services to competitors to secure business for those fabs.

Five nodes in four years?

That’s been Intel’s catchphrase ever since. Literally, it means that Gelsinger promised his company would deliver five different generations of product, each with a smaller size of transistor than the last in just four years. If you’ve ever heard of Moore’s Law, you know that shrinking down the circuits to fit more transistors is one of the key ways that chips improve, but historically, Intel would only shrink its chips every other year — known as the “tick-tock” release cycle. Five shrinks in four years was seen as wildly ambitious.

2 Likes

The five nodes in four years was a nice rallying cry for the team, but I see it as mostly showmanship. With Pat gone the UBS conference was done by Zinsner and the new IFS head Naga Chandrasekaran. An interesting comment from Naga… “And the second half of 2025, our milestone is certifying the node (18A)…” I always thought when Pat said manufacturing ready the node was certified, but he was talking about some other milestone that was a full year ahead of certification. That was probably his biggest issue was overpromising and under delivering.
Intel_7 was really just the old 10nm++
Intel_4 was a new node and Intel’s first EUV node
Intel_3 was just Intel_4+ Pretty much similar to Intel_4
Intel_20A was a major step ahead, but canceled
Intel_18A is really Intel_20A+ (the high density library is identical to 20A)… and is really a H2 2025 program.

So I really just give Pat credit for one node (Intel_4) in four years.
To be fair, TSMC takes similar small steps and the + increment on a process node at Intel could be called a full shrink by TSMC.

In an interview Ann Kelleher mentioned one of their biggest failings was not setting up Intel 10nm (TSMC/Samsung N7 equivalent) to be retrofitted with EUV. Both TSMC and Samsung use EUV on this node now.

All that is as they say “water under the bridge” now. Intel starts checkout silicon on their third EUV factory in Q1 2025, so things are turning around. Intel_18A looks very good and promising if it hits its targets.

Seeking Alpha has a transcript of the UBS conference and gives a refreshing look at the changes going on at Intel.

3 Likes

You have a link on Seeking Alpha?

I was looking at the claims of 10% yields on 18A and thinking to myself “Yeah, but it’s still a long way off, they have time to get it working.” I wonder whether 5 years from now we will look back and see that Gelsinger, other than overpromising and maybe overbuilding, had the right general idea…

1 Like

Bad management. For a few decades, Intel “played it safe” and didn’t reach deep enough into the future, and didn’t take enough risks to push their technology forwards. They did this because they were the clear leader in the space with the second place company being way behind them. Meanwhile, other companies (most notably TSMC) leapfrogged them in technology a few times and then they became the leader in the space. Now we will see if Intel can recover. It will be very difficult. First off, we will see who they choose as a new CEO.

1 Like

Yep. They were dominant over AMD… until they weren’t. By which I mean, while AMD was stumbling around with the Bulldozers etc… Once they got it together with the Zen designs, it was going to be hard fought. Of course, by coincidence they botched up their processes right around then too, and left AMD/TSMC with the field unexpectedly to themselves…

And they misjudged what “the space” was, missing both mobile and GPUs. AMD goes after the heart of the business they did play in, TSMC takes away their fab dominance, and GPU and mobiles come around on the flanks and pull ahead, and… no obvious way out.

Gelsinger wasn’t playing it safe, but he was swinging wildly. Let’s see if the 18A process works out, and the foundry gets customers… then maybe they can start to dig themselves out of the hole.

I don’t think Intel misjudged the space at all. I think they misjudged their capabilities. Intel’s space has been big iron for decades. They bring in the most revenue and the biggest margins. Extra capacity flowed downhill to the next highest margins in order: small servers, workstations, gaming computers, desktops/mobile, etc. There was an inflection point in there where desktops went from higher margins to mobile becoming higher margins. AMD was drastically superior on laptops because their architecture was designed for better yields and flexibility whereas Intel was committed to servers, so that helped AMD, but I didn’t see it coming, and I don’t think it hurt Intel very much as the combination of ever larger server CPUs and processes no longer being the best.

As for GPUs, there were two GPU manufacturers worth mentioning: Nvidia (the best and biggest) and ATI (plucky little competitor, like AMD to Intel). AMD paid up (at the time I thought overpaid) for ATI, and Intel tried to develop its own graphics because that was their corporate mentality and it had yet to fail them. But eventually AMD figured out how to combine ATI + x86 to create APUs, which were low margin but cheap and relatively high volume, which fit AMD’s foundries at the time. So that provided a backbone of revenue as AMD owned all the gaming consoles and eventually that market grew as tablets and Chromebooks came along.

There were other things going on at the physics level that others here can explain better than I can, but basically the Zen design spread out hotspots. There’s be a hot chiplet here and a hot data transfer line over to a hot controller unit there, and more chiplets doing the same, so since Zen came along, you had about the same amount of heat generated but over a much larger surface area, so overall they ran cooler. Cooler allows overclocking which makes things faster. And with MANY small chiplets per wafer, yields (the number of good die candidates per wafer burned) rose significantly. Say there are 64 chiplets on a wafer with ten defects. The worst case is you get 54 good chiplets, a very high yield. If some of those defects on on the same chiplet or on the interstitial areas between chiplets, yields get even better. Whereas Intel’s server centric process was generating ever denser (hotter) chips, and far fewer CPUs per wafers, so heating kept on being more of a problem and yields dropped. Intel was still making money hand over fist because nobody could compete well with them in the highest revenue server sector.

Then Zen servers were developed. If you make a lower chiplet Zen device, you drop it straight into a smaller form factor product and ship it. But for servers, you probably need to have it running for two years or so before anyone buying a mission critical server would consider buying from you. AMD grew revenues on high volume small margin small revenue products, and the added capacity that APUs gave laptops made laptops more profitable. Why would anybody buy a desktop for an employee when they could get the same work on a laptop that employees could take home or on business trips?

Those revenues put AMD into the black while AMD’s servers will being validated. ARM also put new pressures on Intel’s servers, because all those years of price hikes meant nobody really was completely loyal to Intel. NV was the first to figure out how to really make GPUs improve servers, as did AMD. Intel’s graphics weakness meant they had to buy other folks’ GPUs to help speed up their servers’ performance. The process failure was hard to envision from Intel’s long history and from what I read self inflicted.

What advantages does Intel have these days? They are sitting on a ton of cash, but also likely to face shareholder lawsuits. They do have backwards compatibility that IMO is more likely to work than with servers from other manufacturers for the customers that want to expand their data center 3with new servers that can talk to the old servers, but that market is shrinking with every non-Intel server purchased. Every time a better total cost of operation (ICO) server comes along it makes more sense to slash power usage and just punt the old server and put in something that will save money every month.

I don’t think Gelsinger was swinging wildly. He was process focused. But so is TSMC, and with Intel no longer close to competitive (which is the ugly truth nobody wanted to discuss when Gelsinger came back) they have to do MUCH better than TSMC to catch up. I think Intel needs to give up on its Foundry business, close the fabs, lay off the employees, fix the process gap (from what I read here, they’ve already fixed much of the architecture gap) and hope they have enough savings to make that happen. That will cut the losses substantially, leading to a much leaner company. But that mandate has to come from a board of directors that doesn’t want to lose their Intel money or especially their Intel stock value, a board that allowed Intel to deteriorate so far and wants to protect their resumes by not admitting how bad things have gotten. Now that TSMC hates Intel, I see few paths to viability that do not start with fixing Intel’s process issues, which is exactly what Gelsinger was trying to build.

Until Intel admits how bad things have become and starts a complete turnaround, things are only going to get worse. I believe Intel has enough smart people trying to protect their retirement portfolios based on their Intel stock options. Good ideas will come, and maybe already have come. But I lack confidence in Intel’s board to recognize any company saving idea as worth following if it in any way implies that maybe the board shares a large chunk of the blame.

Fool on!
Roleplayer

1 Like