Hacker News Reader

Comment by faeriechangling | Hacker News Reader

I don’t know who is invested in Intel that's betting on anything but a high risk high reward underdog story comeback. Intel already was undercompensating to the point of driving the talent out of their company to begin with. These wage cuts are a sign for investors to flee faster because management lacks coherence.

Management seems to be betting on a strategy of promising the world to gain government subsidies while in reality cutting costs and letting the business wither and die and I don’t know if that justifies the valuation.

View on HN

I wonder if they just became terminally risk averse.

The thing that made them big is pretty obvious: high-margin premium priced x86 CPUs, which is a position they were able to hold for decades due both to manufacturing excellence and being able to shepherd the standard. They simply can't do it anymore-- execution has stumbled, and the market for x86 is gradually weakening, especially in spaces outside the cut-throat, lower margin, consumer/corporate desktop "it must run Windows and my line-of-business software/Call of Skyrim" vortex.

They could have been using decades of margin to build and consolidate positions in other segments of the market, just to be sure that the next clever new device comes out with an Intel chip, even if it's non-x86. They didn't.

Rather than, say, taking the ARM licenses they ended up flogging and building a competitive mobile CPU on their leading process, they invented weird Atom chips that ended up in like 3 phones.

Okay, flash is huge now, but it's a low margin game maybe not worth their investment. But they chucked out Optane, which still had some possibility as a "halo product" and for some commercial price-is-no-object markets.

It remains to see how sticky Arc graphics will be; they got a boost by the tight market conditions last year where people would grudgingly take any card they could afford. Will people come back to them out of explicit preference? I also wonder if the Larrabee/Xeon Phi adventure set them back or ahead-- trying to make a GPU architecture out of x86 was an interesting take on "GPUs are hard to program", but did it yield a lot of useful knowledge?

> I wonder if they just became terminally risk averse.

> They could have been using decades of margin to build and consolidate positions in other segments of the market.

That seems pretty simplistic. Optane and Larrabee were bold efforts. (Heck, so was Itanium). It's not like they didn't try throwing money and R&D at things. But that isn't always enough.

> Rather than, say, taking the ARM licenses they ended up flogging and building a competitive mobile CPU on their leading process, they invented weird Atom chips that ended up in like 3 phones.

They made a decent effort with Xscale too, to be fair.

> They made a decent effort with Xscale too, to be fair.

Sure, but they threw it away about a year before the iPhone came out.

Yeah, they did try a lot of bold efforts, but they've pretty much all been mis-steps. They always had factions that were "x86 forever" and when things weren't going great those factions would come in and say "hey, why are we working on this thing that's not our x86 bread-n-butter?" and they'd either kill it or they'd only half-heartedly keep it going until the underinvestment killed it.

But weren't they forced to purchase StrongARM/XScale from DEC as part of their hyperthreading patent infringement settlement?

Think they should have either really stuck it out with Itanium, put more effort into compiler/tooling development or kept up with Xscale.

The line-of-business software is quickly turning into a web browser accessing an internal server cluster, and just about any chip can run that nowadays.

servers also have chips though

Yep. And Amazon has gone as far as designing their own data center chip because Intel was selling a lackluster product for the price.

A lot of server workloads also aren't about maximizing flops/Watt or iops/Watt, but rather setting up DMA between the disk controller and the network card and having a low idle power, to get the most network throughput per Watt. A lot of the high-end computation is also about handing off data to the GPU and having a low idle power.

It took Intel way too long to get the power consumption on the Atom processors down (not ops/Watt, but total Watts), back when Atom had a chance in the phone market. (A difficult task, which is why I bought ARM Holdings stock shortly after Android came out because I realized (1) nearly everyone was going to soon have a smartphone and (2) whether iPhone or Android won, it was going to be very difficult for Intel to get competitive in the smartphone market fast enough to prevent ossification of the market around ARM. Unfortunately, SoftBank took ARM private and closed out my trade.)

Now, the same dynamic of the low-end x86 processor eating its way up the stack to threaten the high-margin mainframe and RISC servers is playing out with low-end ARM eating its way up the stack and displacing a lot of uses for the high-margin x86 server chips.

> Unfortunately, SoftBank took ARM private and closed out my trade.

Uh, probably fortunately. Softbank bought ARM for $32b 7 years ago, and fat chance it's worth that inflation-adjusted now.

Man, remember when Intel Xscale was the hot shit for anything mobile? They completely owned the high-end market and then threw it away just a year before the iPhone released.

They had a cloud business too are one point - might have gotten rid of it around 2008?

> Intel already was undercompensating to the point of driving the talent out of their company to begin with.

Yep. Intel salaries were already on the low side. To some extent they could get away with this in the past because many of their locations were outside of silicon valley in places where there really weren't a lot of (or any) other companies where you could work at the same level. For example, say you worked at their Folsom, CA or Hillsoboro, OR locations - if you were going to change to a different company that pretty much meant you were moving. But now that a larger proportion of jobs are remote that strategy for keeping salaries low isn't going to work.

If you’re predicting a major decline of globalism, then betting on a domestic firm in an extremely economically important sector that is currently extremely reliant on globalism doesn’t sound that crazy.

What stops the Americans from simply stealing (or buying) TSMC or Samsung or whatever technology with how deep their fingers are into those companies and starting up fabs as needed? Even if the world "Deglobalizes" overall does that really mean that America maintaining its influence over countries like Taiwan and Japan and South Korea is not an option if the issue is important enough? America also has AMD/Nvidia for the chip design side of the business.

Intel could be valuable to America but the problem is the damned word could and it's not essential.

There are many reasons why Koreans are angry with the US, and some are related to semiconductors.

After Intel sold its NAND division to SK Hynix, South Koreans were upset because the US government was trying to impose sanctions on Chinese chip factories. Koreans believe that Intel sold it to South Korea after consulting with the US government, knowing that Intel's Dalian fab in China would be affected. It is some kind of insider trading or fraud.

Then passing the IRA and pressuring Samsung Electronics and SK Hynix to build semiconductor factories on US soil is not an allied attitude.

My understanding is that the semiconductor supply chain is hyper-globalized with many steps in the process being so enormously capital-intensive that only a single firm can produce them at scale, the firms producing these various steps are widely distributed geographically, and each step is nearly useless if any step is missing.

And I wouldn’t just be worried about something simple and direct like “the United States losing influence over Taiwan’s semiconductor industry.” The concern is more like “the United States losing its global influence and thus its ability to almost unilaterally enforce a particular global order including (among other things) relatively safe global ocean freight.”

Sounds like you read Peter Zeihan too. :-)

As a matter of fact, he published a video today talking about this topic in some detail. You can find it on YT with the title "Semiconductors: China's the Odd Man Out".

Multinationals have no loyalty to any country anyway. Giving TSMC big tax breaks to build top fabs in the US to make AMD chips is equally as good as having Intel be a success from a national security perspective.

That's a wishfully pessimistic and cynical point of view. Multinationals are made of people and like people, some of them have more less turncoat attitudes. TSMC would probably not abandon Taiwan, it knows it's essential to its defense.

Nothing about my comment implies they’d have to. The US just wants supply chain security and availability. That can be done by TSMC factories in the US regardless of where else they have factories.

not really. the US cares a lot more about Taiwan if they are making the chips for the US. a sizable part of Taiwan's national defense strategy is to have a benevolent monopoly on cutting edge chip production so their allies think it's worth it to fight a war with China if necessary.

Yes, that is the current state. I was entertaining a future scenario above. TSMC is currently building more capacity outside of Taiwan, and the US is making investments that will likely continue this.

But they're not doing investments that will entirely replace Taiwan's relevance. The bulk of their top tier technology will remain in Taiwan, and I would be very surprised to see that change. Moving some production to the US to alleviate worst-case scenarios for Washington, yes, but moving enough that Taiwan's defense becomes irrelevant? I really don't think so.

I mean, if you were TSMC's CEO/Board, knowing full well that your country of birth is highly dependent on your presence for its defense posture, would you do allow this to change?

I had high hopes that Gelsinger would be able to pull it off. But seeing this makes me think that he's getting too much pushback from other factions within Intel and the institutional investors.

I invested when they announced their ARC GPUs at the height of the chip shortage. Took them so long to deliver it seems like they lost their chance.

Unfortunately, what they delivered was also extremely underwhelming. I expected something that would at least be a mid-range GPU, but they ended up releasing trash tier casual gamer cards. I do not believe that was their intent. AMD and Nvidia are just too competitive today.

Given that AMD has been competing with Nvidia for so long and has had so much trouble gaining market share and that they are always competing on price rather than performance, I reckoned that Intel would suffer the same fate with ARC.

How is Intel an underdog next to AMD? I don't know many business machines running AMD.

Intel are screwed. Many of its biggest customers have started just making their own chips instead. The bleeding edge of compute appears to be falling to Nvidia (who now make both the chips and the network that connects them). Their latest GPU project is going the way of their other GPU projects. They don't even have the biggest/baddest fab capability thing going on any more. I have no idea what's going on in there, but I can imagine, and it's not pretty.

Other than Apple who else is making their own chips?

Amazon. And I was assuming Microsoft but it turns out they are using ampere. Heh, Google too. I guess it's starting to work out for ampere (but Google make their own tensor chips).

> How is Intel an underdog next to AMD? I don't know many business machines running AMD.

I don't know what a "business machine" is, but you probably aren't intimately familiar with organizations that have been directly purchasing large amounts of processors or processor containing devices over the last several years. For many such entities, AMD has made way more sense than Intel.

I meant workstations, laptops, servers, etc. My org has 1000s of endpoints and 10s of servers across multiple campuses. All of it running Intel.

Then your org is over paying for mediocre HW, at least for servers, arguably also for laptops (Dev machines for sure) At least for the last two years.

I can count on one hand the number of devs we have. We're not a software tech company. vPro is a standard for device management here.

Intel has been shitting the bed for a decade+ now

not quite, Haswell was only 9.5 years ago :) fma+avx2 is a pretty nice architecture boost.

Investing in this giant companies instead of providing funding to spur new ones is just dumb

Building a fab is very expensive. It's unlikely for some small new entrant to be successful, unless they happen to invent something better than EUV lithography or can somehow build all their machines from scratch.

Basically at this point, the big players are TSMC, Intel, and maybe Samsung. No one else can afford what it costs to be constantly upgrading their factories and doing the necessary basic research to get high quality and consistent yields.

Reply on news.ycombinator.com