IBM vs. Global Foundries: Is This the End of Outsourcing Manufacturing?

[Daniel Nenni]

His consulting clients must be horrified by this one:

IBM vs. Global Foundries: Is This the End of Outsourcing Manufacturing?

Global Foundries was primarily created around what once was AMD’s and IBM’s microprocessor manufacturing capability. It was formed with the fundamental

techspective-net.cdn.ampproject.org

Global Foundries was primarily created around what once was AMD’s and IBM’s microprocessor manufacturing capability. It was formed with the fundamental belief that a company focused on manufacturing, with their overhead, could be more cost-effective at building parts because it could get economies of scale far more significant than an internal effort at either AMD or IBM. This move seemed to make a ton of sense on paper, and these moves were used to suggest Intel retained its manufacturing capabilities.

But this has not proven to be the case. First, AMD moved away from Global Foundries because the firm refused to develop the advanced process AMD needed to be successful. Then, more recently, IBM sued Global Foundries for breach of contract because IBM alleges Global Foundries refused to comply with the contract IBM has with the company. The contract also gave Global Foundries IBM’s manufacturing capability.

This latest litigation appears to validate Intel’s decision not to outsource manufacturing if only to assure their manufacturing capability isn’t eliminated accidentally in the process, as it seems to have been with IBM and AMD........

Wrapping Up​

Global Foundries was created to optimize the supply chain for microprocessors for AMD, IBM, and others. They made commitments that appear to now be broken and unreachable, putting into question the very reason why the firm was created. As a result, they have potentially become untrustworthy as a supplier. Their very reason for existing is in question because it appears they cannot fulfill their primary mission of providing parts to the firms that initially created the company.

And if a company can’t or won’t meet the needs of its founding customers, any other company with far less clout is likely to be equally disappointed, making it far less likely they will use Global Foundries and far more likely they’ll follow AMD to TSMC. We may be seeing the beginning of the end for Global Foundries because their technology is becoming increasingly outdated. Their large customers are not only fleeing from them, one is suing them for breach, making it look like Intel made the right move by keeping manufacturing in-house. The Global Foundries model may be non-viable.

It also showcases that control may be more critical, particularly during a pandemic, than the cost savings from outsourcing. Because if that company you outsourced to can’t or won’t execute, you are in worse shape than you likely were had you retained your manufacturing capability.

Rob Enderle: As President and Principal Analyst of the Enderle Group, Rob provides regional and global companies with guidance in how to create credible dialogue with the market, target customer needs, create new business opportunities, anticipate technology changes, select vendors and products, and practice zero dollar marketing. For over 20 years Rob has worked for and with companies like Microsoft, HP, IBM, Dell, Toshiba, Gateway, Sony, USAA, Texas Instruments, AMD, Intel, Credit Suisse First Boston, ROLM, and Siemens.

 

[coldsolder215]

"Hence the extraordinary growth of a class, or rather, of a stratum of rentiers, i.e., people who live by 'clipping coupons' [in the sense of collecting interest payments on bonds], who take no part in any enterprise whatever, whose profession is idleness. The export of capital, one of the most essential economic bases of imperialism, still more completely isolates the rentiers from production and sets the seal of parasitism on the whole country that lives by exploiting the labour of several overseas countries and colonies."

Lenin would say "no".

 

[count]

I think this is more a case against private equity than against the fabless model.

GF was a financialized company built by MBAs using spreadsheets to trade around and sell to other investor types as opposed to engineers building a company for the long term on the basis of solid technology. Of course it failed.

 

[Daniel Nenni]

I think this is more a case against private equity than against the fabless model.

GF was a financialized company built by MBAs using spreadsheets to trade around and sell to other investor types as opposed to engineers building a company for the long term on the basis of solid technology. Of course it failed.

This is the one that caught my eye:

"It also showcases that control may be more critical, particularly during a pandemic, than the cost savings from outsourcing. Because if that company you outsourced to can’t or won’t execute, you are in worse shape than you likely were had you retained your manufacturing capability."

When you chose a foundry, professional lives are on the line which is why TSMC beats the "trusted foundry" drum at every event.

But to the point, AMD was falling behind which is why they dumped the fabs, which was a great move. I seriously doubt AMD would have done any better with FinFETs than GF did so what choice did AMD really have? And partnering with TSMC was a great move. And now Intel is following AMD to TSMC........

 

[BobbiMac]

"Hence the extraordinary growth of a class, or rather, of a stratum of rentiers, i.e., people who live by 'clipping coupons' [in the sense of collecting interest payments on bonds], who take no part in any enterprise whatever, whose profession is idleness. The export of capital, one of the most essential economic bases of imperialism, still more completely isolates the rentiers from production and sets the seal of parasitism on the whole country that lives by exploiting the labour of several overseas countries and colonies."

Lenin would say "no".

Huh! Speak American please.

 

[hskuo]

This is the one that caught my eye:

"It also showcases that control may be more critical, particularly during a pandemic, than the cost savings from outsourcing. Because if that company you outsourced to can’t or won’t execute, you are in worse shape than you likely were had you retained your manufacturing capability."

When you chose a foundry, professional lives are on the line which is why TSMC beats the "trusted foundry" drum at every event.

But to the point, AMD was falling behind which is why they dumped the fabs, which was a great move. I seriously doubt AMD would have done any better with FinFETs than GF did so what choice did AMD really have? And partnering with TSMC was a great move. And now Intel is following AMD to TSMC........

Intel could be the last milestone to conquer for foundry business in tsmc. I expect eventually there will be 2-3 giants in semiconductor manufacturing. tsmc leads the logic and Samsung/Micron leads the memory.

 

[tooLongInEDA]

Huh! Speak American please.

I believe that's a - quite intentional - verbatim quote from Lenin (possibly translated from Russian). Clearly you miss that authentic dense, almost unreadable prose and the 3 hour politburo speeches in the same vein !

But coldsolder215 had a serious point here. An increasing proportion of wealth in countries like the USA and UK does seem to come from rentier activity and private equity deals are often part of this. It would be interesting to know if this sort of behaviour suppresses or increases economic growth and innovation. Of course, that might require some serious brainpower and hard work from our governments and public servants ...

Back on topic - GF has been a huge disappointment. I guess they are finding a niche behind the leading edge. It does increasingly seem in technology that once a company has dropped off the leading edge, there's almost never any way back. with very occasional exceptions (e.g. Apple, perhaps AMD).

 

[SemiChemE]

There's way more to the story than this article implies. The reality is that AMD's spin-off of GLOBALFOUNDRIES (GF) was a brilliant move for AMD. If you remember, in the 2009-2012 timeframe, they were trailing Intel both on the process side and on the chip design side and losing market share rapidly.

Spinning off the Fabs into GF allowed AMD to focus on chip designs, both in the processor and graphics spaces and to become competitive in the console gaming space. It also gave them access to the deep pockets of Mubadala (an Emirati Sovereign Wealth Fund), which enabled construction of Fab 8 in Malta, NY.

In the meantime, GF set about integrating AMD's fabs (Fab 1 and Fab 8) with those of Singapore-based Chartered Semiconductor to attract additional foundry customers, which would help fund future advanced node development. Unfortunately, this turned out to be too complex to succeed on every front.

GLOBALFOUNDRIES' 32nm HKMG technology node (SOI), which was the backbone of the AMD Bulldozer Family, was more than a year late, compared to Intel's 32nm technology. Further, between design flaws and lack of proper compiler optimization, Bulldozer under performed expectations, putting AMD in serious distress and constraining GF's prospects. At the same time, Intel was able to transition to Finfets at the 22nm node, way ahead of the rest of the industry and leading to a scramble among the remaining players (IBM/GF/SAMSUNG and TSMC) to catch up.

GF was successful in bringing up a 28nm node in both Fab 1 and Fab 8 for a foundry customer and producing some 32nm chips for IBM, but AMD would struggle mightily for the next several years. GF's 20nm planar node was late, failing to make it into production and their internal 14nm finfet development stalled out.

However, in 2014, they were able to license Samsung's 14nm finfet technology. Here GF was likely helped by Apple's demand that Samsung have a second source supplier. And, while little of Apple's business actually found its way to GF (partially due to direct chip competition between Apple and Samsung), it gave them a foot in the door with the Finfet technology that ultimately powered AMD's resurgence with their ZEN line.

GF's 14nm tech allowed AMD to be competitive with Intel. As intended, the spin off allowed AMD to focus on design, while others worked on technology development. It also allowed GF to partner with Samsung and pool resources among multiple foundry customers to support the 14nm bring up and subsequent 7nm (and 22FDSOI) development. The foundry business also provided a path to keep the the now older Fab 1 loaded, as a money-making asset, rather than an idle drain (What would happen to older AMD fabs in the past).

Now it's true that GF was unable to deliver a viable 7nm technology in a competitive timeframe, but it did put AMD in a position to migrate to TSMC, while also providing a business model to keep the two former AMD fabs productive and profitable (running "older" technologies, which will likely continue to be viable for another 15-20 years or more if history is any guide). In other words, the GF spin off has been a big success for AMD.

Now, the deal has been much worse for IBM. In 2015, they seemed to give GF a sweetheart deal (They paid them $1.5B to take their two fabs and many experienced engineers). The intention was for GF to develop a 7nm technology node, but at the time GF was one of only four companies in the world undertaking such a task. The other three companies (Intel, Samsung, and TSMC) each had revenues well in excess of $30B/year, compared to <$6B for GF, and research budgets to match. Further, GF would leapfrog the 10nm node. This was David vs. three Goliaths and it was clear to everyone that a bet on GF would be a longshot.

By 2018, GF actually had functioning 7nm chips, but ramping yields to acceptable levels was still months away. It's hard to say for sure, but given intel's continued yield struggles at both 10nm and 7nm, GF's tech may have been pretty competitive with them. However, TSMC was already sampling product to GF's major customer, AMD. At that point the writing was on the wall. IBM would never be able to provide enough business to justify continued development and to fill the fab without AMD. It made more sense for GF to cut it's losses and focus on the business it could win (14nm, 22FDSOI, and various derivatives, like RF). IBM's bet went south and back in 2018 they seemed to have accepted it. The recent lawsuit may let them claw back some of their losses, but that's for the courts to decide depending on the details of what I'm sure are long and detailed contracts.

Today, GF has made a significant turn around. The decision to focus on older technology nodes and derivatives appears to have been the right one. They are able to address about 70% of the Logic semiconductor market. It's not going away any time soon, and there are opportunities for it to be very profitable if managed properly.

Sure, GF has lost the leading edge. That's exactly what they said back in 2018, but they are on firm ground and the future looks bright.

 

[smeyer0028]

Possibly related to your post. The AMD engineers who
have designed the Zen architecture that hasIPC rates as
good or better than Intel's mostly came from IBM when
it left the hardware market. I think IBM finally exited the
server market in 2014.

 

[count]

There is a rumor that Intel is looking at buying GF.

Intel in Talks to Buy GlobalFoundries for $30 Billion: Report

Intel reportedly is considering buying GlobalFoundries as part of efforts to expand U.S. chip manufacturing.

www.thestreet.com

There is only one way this makes any sense for Intel, and that's if the long term goal is to spin out the fab business and if that was the plan all along. Intel could in theory, buy GF, put it's own fabs in a combined foundry business, and then spin the whole thing out again 3 years down the road. It would be a nightmare to execute but makes sense on a spreadsheet, and as I said earlier, GF is a a company built on spreadsheets and financial models.

 

[benb]

Looking at this industry from 2010 to present, these are the interesting things that didn’t happen:
- 450nm fabs
- A leading edge logic process for high performance compute.
- Non-volatile memory scaling stopped (scaling vertically now)
Semiconductors have been downsizing and retrenching since 1999. The last big hurrah was Y2K. This year, semis will grow 26%, the first time that has happened in decades. The survivors of the retrenching have some breathing space, which may permit a more growth-oriented approach, which may permit addressing these unmet needs.

It would be interesting if GF and Intel join forces, develop a novel HPC process (non-finfet, or perhaps a radically modified form of finfets), and become the HPC foundry.

 

[tooLongInEDA]

Looking at this industry from 2010 to present, these are the interesting things that didn’t happen:
- 450nm fabs
- A leading edge logic process for high performance compute.
- Non-volatile memory scaling stopped (scaling vertically now)
Semiconductors have been downsizing and retrenching since 1999. The last big hurrah was Y2K. This year, semis will grow 26%, the first time that has happened in decades. The survivors of the retrenching have some breathing space, which may permit a more growth-oriented approach, which may permit addressing these unmet needs.

It would be interesting if GF and Intel join forces, develop a novel HPC process (non-finfet, or perhaps a radically modified form of finfets), and become the HPC foundry.

What is this "Semiconductors have been downsizing and retrenching since 1999" ?

A quick search reveals the W/W annual revenues increased from $204bn in 2000 to $481bn in 2018.