Tuesday, July 13, 2021

Intel Did A Boeing

Two years ago, Wolf Richter noted that Boeing's failure to invest in a successor airframe was a major cause of the 737 Max debacle:
From 2013 through Q1 2019, Boeing has blown a mind-boggling $43 billion on share buybacks
I added up the opportunity costs:
Suppose instead of buying back stock, Boeing had invested in its future. Even assuming an entirely new replacement for the 737 series was as expensive as the 787 (the first of a new airframe technology), they could have delivered the first 737 replacement ($32B), and be almost 70% through developing another entirely new airframe ($11B/$16B). But executive bonuses and stock options mattered more than the future of the company's cash cow product.
Below the fold I look at how Intel made the same mistake as Boeing, and early signs that they have figured out what went wrong.

William Lazonick and Matt Hopkins chronicle Intel's version of the Boeing disease in How Intel Financialized and Lost Leadership in Semiconductor Fabrication:
In the years 2011-2015, Intel was in the running, along with TSMC and SEC, to be the fabricator of the iPhone, iPad, and iPod chips that Apple designed. While Intel spent $50b. on P&E and $53b. on R&D over those five years, it also lavished shareholders with $36b. in stock buybacks and $22b. in cash dividends, which together absorbed 102% of Intel’s net income (see Table 1). From 2016 through 2020, Intel spent $67b. on P&E and $66b. on R&D, but also distributed almost $27b. as dividends and another $45b. as buybacks. Intel’s ample dividends have provided an income yield to shareholders for, as the name says, holding Intel shares. In contrast, the funds spent on buybacks have rewarded sharesellers, including senior Intel executives with their stock-based pay, for executing well-timed sales of their Intel shares to realize gains from buyback-manipulated stock prices.
What did spending $81B on buybacks cost Intel? It is true that both Intel's leading competitors, SEC (Samsung) and TSMC bought back stock, but there was a big difference in why they did:
The purpose of SEC’s stock buybacks in 2002-2007 and 2014-2018 was to increase the voting power of the founding Lee family, thereby consolidating its strategic control over resource allocation against the threat of corporate raiders.[17] It is clear from SEC’s remarkable history that the Lee family has used its strategic control to allocate profits to investments in world-class productive capabilities. So too at TSMC under the leadership of Morris Chang. At 50% of net income over the past decade, TSMC’s dividend payout ratio was 2.8 times that of SEC and 1.4 times that of Intel. The sole purpose of TSMC’s stock repurchases between 2003 and 2008 was to buy out the ownership stake of Philips.[18]
Two problems:
  • Intel prioritized buybacks over dividends, the competition did the reverse, only using them for strategic goals.
  • Intel distributed essentially its entire net income to shareholders, where TSMC distributed about half and SEC under a third.
Despite Intel having very low retained earnings:
in recent years the company has made substantial allocations to P&E and R&D, even as it has distributed almost all its profits to shareholders.[20] But Intel has been able to tap other cash flows to make, simultaneously, large-scale productive investments and shareholder payouts. For the decade, 2011-2020, these other cash flows included depreciation charges of $87b., long-term debt issues of $45b., and stock sales (mainly to employees in stock-based compensation plans) of $12b.
Table 1 shows that Intel's investments in P&E and R&D for 2011-202 totalled $$236B, while the "other cash flows" they quote totalled $144B and retained earnings totalled $50B. I believe the remaining $42B is explained by Footnote 20:
Note that R&D is accounted for as a current expense, so only increments to R&D spending must be financed out of profits or other sources of funds such as depreciation, debt, or equity issues.
The $81B Intel spent on buybacks in that decade could instead have raised their investment in P&E and R&D by over one-third. Instead, according to Table 3, about $105M or 0.1% of it ended up in the pockets of Intel's first three non-engineer CEOs:
Intel’s buybacks reached $10.6b. in 2005, the year in which Otellini, Intel’s first non-engineer CEO, took over. Buybacks declined to an average of $1.7b. in 2008 and 2009 in reaction to the financial crisis, but then were jacked up to as high as $14.3b. in 2011. The following year, buybacks were $5.1b. as Otellini departed as CEO with $40m. in total compensation, 82% of it stock-based.

With Krzanich as CEO, buybacks peaked at $10.8b. in 2014. He raked in $40 million in total pay (79% stock-based) in 2017 but was ousted in mid-2018 for having a “consensual relationship” with an Intel employee.[21] In early 2018, news outlets alleged that Krzanich engaged in insider trading, based on non-public information of security flaws in Intel’s CPUs, as he sold all his Intel shares except for the minimum 250,000 he was required by contract to hold.[22]

With Krzanich’s exit, the new CEO was Robert Swan, an MBA who had spent his career in finance at a number of companies, including GE, TRW, Northrup Grumman, eBay, and General Atlantic, before joining Intel as CFO in 2016. From 2018 through 2020, Swan averaged just under $13m. in total remuneration, of which 57% was stock-based. In the Swan years, annual dividends were 19% higher and annual buybacks 186% higher than in the Krzanich years.
Boeing switched CEOs from engineers to financiers when McDonnell-Douglas' Harry Stonecipher:
“bought Boeing with Boeing’s money.” Indeed, Boeing didn’t ultimately get much for the $13 billion it spent on McDonnell Douglas, which had almost gone under a few years earlier. But the McDonnell board loved Stonecipher for engineering the McDonnell buyout, and Boeing’s came to love him as well. This was in no small part because Stonecipher cast himself as the savior of Boeing and knew just how to exploit a bad situation to get his way.
Stonecipher’s other big cultural transformation was focused on maligning and marginalizing engineers as a class, and airplanes as a business. “You can make a lot of money going out of business” was something he liked to say.
The end result of focusing the company on the stock price instead of the product was the 737 MAX disaster, the 787, KC-46 Pegasus tanker, and space program debacles, and now the long delay in certification of the 777X.

Similarly, the end result for Intel of focusing on the stock price instead of the product was:
In 2020, both TSMC and SEC transitioned from 7nm to 5nm process, and in 2021 both are making investments to commercialize 3nm. TSMC went from zero 5nm revenues in 2Q20 to 8% in Q320 and 20% in Q420.[7] SEC is intent on closing the technology gap with TSMC by allocating $28b. to capital expenditures in 2021, about the level of its 2020 plant and equipment (P&E) investments.

For its part, TSMC has announced plans to spend $100b. in total on P&E and R&D over the next three years, including $30b. in 2021, up from $17.2b. in 2020. TSMC will construct a $12b. 5nm facility in Arizona and is also considering the state as the site for a $25b. 3nm fab.[8] Most of this new capacity is slated to fabricate Apple’s M-series processors.[9]

Intel still leads the global semiconductor industry in total revenues. But, as an IDM, Intel manufactures almost all its CPUs at 14nm, and its 10nm capacity has been stuck, with limited output, since 2018.[10] Meanwhile, Apple is abandoning Intel processors for its Mac computers, turning instead to TSMC to fabricate Apple’s own designs.[11] Intel itself already contracts with TSMC and UMC to produce 15-20% of its non-CPU chips. Moreover, later this year, TSMC will commence production of intel’s Core i3 processors, inside advanced laptops, at 5nm.[12]
Intel's competitors' announced plans are to invest 5% and 11% more than Intel's average for the last 5 years. And they've been getting a lot more bang for their buck than Intel over those years.

Earlier this year Intel appears to have realized that they'd been suffering from the wrong kind of CEO, and hired back engineer Pat Gelsinger as CEO. Perhaps it was an indication that Intel would no longer "make a lot of money going out of business" when in early May Reuters reported that Intel will ‘focus’ less on buying back company stock — CEO and that:
The new CEO said in March that Intel will spend up to $20 billion to build two new factories in Arizona, greatly expanding its advanced chip manufacturing capacity.
Note that, on TSMC's figures for their Arizona fabs, "up to $20B" isn't enough for two 5nm fabs, or one 3nm fab. So despite fewer buybacks Intel still isn't planning to compete with the foundry leaders. Though one of Gelsinger's early announcements was:
Intel Foundry Services (or IFS) is one prong of Intel’s strategy to realign itself with the current and future semiconductor market. Despite having attempted to become a foundry player in the past, whereby they build chips under contract for their customers, it hasn’t really worked out that well – however IFS is a new reinvigoration of that idea, this time with more emphasis on getting it right and expanding the scope.
Ian Cutress reported on an early stage of IFS development in Intel Licenses SiFive’s Portfolio for Intel Foundry Services on 7nm
Today’s announcement from SiFive comes in two parts; this part is significant as it recognizes that Intel will be enabling SiFive’s IP portfolio on its 7nm manufacturing process for upcoming foundry customers. We are expecting Intel to offer a wide variety of its own IP, such as some of the x86 cores, memory controllers, PCIe controllers, and accelerators, however the depth of its third party IP support has not been fully established at this point. SiFive’s IP is the first (we believe) official confirmation of specific IP that will be supported.
But on the other hand:
Intel’s financial report for Q121 shows that between February 22 and March 27, with Gelsinger at the top, Intel executed $1.5b. in buybacks. Perhaps the new CEO was focused on other things during his first month and a half in office. For Intel as for other major U.S. corporations, the addiction to buybacks is hard to kick.

The 19 publicly listed corporate members of the U.S. Semiconductor Industry Association that signed a letter to President Biden in February,[25] asking the government for financial support for their industry, did buybacks of $540b. (2020 dollars) from 2001 through 2020, with IBM, Intel, Qualcomm, and TI accounting for 84% of these repurchases. In 2016-2020 alone, these 19 companies squandered $148b. (nominal) on buybacks—almost three times the $50b. in financial aid that the Biden administration has offered the SIA.

Our policy recommendation for the Biden administration is simple: As a condition for giving the U.S. semiconductor industry $50 billion in infrastructure assistance, put a ban on SIA members doing stock buybacks as open-market repurchases.
Right, otherwise the $50B is going to end up in buybacks. So much easier than productive investments, which actually require thought, planning and time to come to fruition.

But why is is even legal for companies to manipulate their stock price in this way, allowing their stockholders to avoid tax by converting current income (dividends) into capital gains (buybacks)? The answer is the Securities and Exchange Commission’s Rule 10b-18:
Rule 10B-18 is considered a safe harbor provision. A safe harbor is a legal provision to reduce or eliminate legal or regulatory liability in certain situations as long as certain conditions are met. If the company abides by the four conditions of Rule 10B-18 when it is repurchasing the shares, the SEC will not deem the transactions in violation of anti-fraud provisions of the Securities Exchange Act of 1934.
Clearly, Congress understood in 1934 that companies buying back their stock allowed management to both avoid tax and artificially pump their stock price. But Reagan's SEC thought that avoiding tax and pumping the stock market was a good thing, hence Rule 10b-18. This short-term thinking motivated companies to choose their CEOs for their skill in financial rather than product engineering, with the consequent erosion of US technology competence. In the short term it is easy to “make a lot of money going out of business”. In the long term you're out of business but the CEO is enjoying a well-funded retirement. Repeal of Rule 10b-18 should be a priority for Biden's SEC but I wouldn't hold my breath.


David. said...

It looks like Intel is getting serious about the foundry business - see Intel in talks to buy GlobalFoundries for about $30 billion - WSJ.

David. said...

Jeffrey Burt's Intel Sees Path Back To Chip Process Performance Leadership is an account of Intel's roadmap:

"At its Intel Accelerated event July 26, Gelsinger and Ann Kelleher, general manager of technology development, are outlining a processing and packaging roadmap that they argue will bring Intel to process performance parity in 2024 with other leaders and clear leadership a year later. The idea is to leverage the technologies available now – such as the SuperFIN manufacturing process and Foveros 3D integrated circuit – and bring on newer innovations over the next three years to remain competitive now while aiming for the leadership position down the road."

There's a lot of detail but the bottom line is that Intel plans to invest less than either TSMC or Samsung, so it isn't clear that this is a path to a "leadership position down the road".

David. said...

Sam Rutherford reinforces the message that Intel's plan to achieve "leadership" isn't realistc in TSMC Will Start Making 2nm Chips As Intel Tries to Catch Up:

"Taiwan Semiconductor Manufacturing Co.’s new foundry will produce 2-nanometer chips. Construction on the plant in Hsinchu, southwest from Taiwan’s capital of Taipei, is expected to start as soon as early 2022.

This isn’t great news for Intel, which yesterday said it would reach process parity by 2024 and leadership by 2025. TSMC’s 3nm tech is reportedly expected to be put into production in late 2022—meanwhile, Intel will be rolling out 7nm chips toward the end of 2022 and into 2023. Now that TSMC has approval to start work on a plant that will produce 2nm chips, which are expected to debut in 2023, well—Intel has its work cut out for it."

David. said...

Simon Sharwood reports that Cloudflare says Intel is not inside its next-gen servers – Ice Lake melted its energy budget:

"We evaluated Intel's latest generation of 'Ice Lake' Xeon processors," Howells wrote. "Although Intel's chips were able to compete with AMD in terms of raw performance, the power consumption was several hundred watts higher per server – that's enormous."

David. said...

Intel CEO Pat Gelsinger:

"reiterated that Intel has plans to build at least two new semiconductor factories in Europe with investments of as much as 80 billion euros ($95 billion) over the next decade."

TSMC plans this much in new fabs in the next three years with Samsung planning double this in the same period..

David. said...

The Economist's long article entitled on paper Gelsinger's opening gambit views Intel as following the lead of Satya Nadella at Microsoft in opening up to competition:

"The American semiconductor giant has long guarded its core chipmaking business as jealously as Microsoft did its OS. After years of product delays, misplaced technology bets and changing management, it is ready for some fresh air. “Our processes, our manufacturing, our intellectual property through our foundry services [producing processors for other chipmakers]: all will now be available to the world,” professes Pat Gelsinger"

David. said...

Dan Robinson's TSMC allocates a third more on capital spending in 2022 – that's a fab-ulous $44bn on new plants and other things:

"The financial outlay is estimated to jump to as much as $44bn this year, up from $30bn in the previous fiscal, and was outlined today when the company reported results for its calendar Q4 with revenues up 5.8 per cent year-on-year to $15.74bn and net profit up 16.4 per cent to $6bn.

The top line was driven by demand for semiconductors manufactured with TSMC's 5nm process technology. This contributed to 23 per cent of TSMC's total wafer revenue in the quarter, while those made using its 7nm process accounted for a further 27 per cent. These advanced technologies therefore made up half of TSMC's total wafer revenue during the quarter."

David. said...

Ben Thompson's The Intel Split is a must-read. I'm not even going to sample it, you have to read it.

David. said...

Agam Shah reports that Intel R&D spending surges after years of neglect as Gelsinger pledges to make Chipzilla great again:

"The US giant spent $15.19bn on research and development in fiscal 2021, more than 20 per cent of the company's $74.7bn revenue. That was about a 12 per cent increase from research and investments in 2020, and it was largest year-over-year increase since 2012, when R&D spending went up by 20 per cent."

This isn't likely to be enough, but at least Intel is trying.

David. said...


"provides evidence from the US and Denmark that managers with a business degree (“business managers”) reduce their employees' wages. Within five years of the appointment of a business manager, wages decline by 6% and the labor share by 5 percentage points in the US, and by 3% and 3 percentage points in Denmark. Firms appointing business managers are not on differential trends and do not enjoy higher output, investment, or employment growth thereafter.
Exploiting exogenous export demand shocks, we show that non-business managers share profits with their workers, whereas business managers do not. But consistent with our first set of results, these business managers show no greater ability to increase sales or profits in response to exporting opportunities. Finally, we use the influence of role models on college major choice to instrument for the decision to enroll in a business degree in Denmark and show that our estimates correspond to causal effects of practices and values acquired in business education - rather than the differential selection into business education of individuals unlikely to share rents with workers."