The strife of Brian: Why doomed Intel boss's ex86 may not be the real reason for his hasty exit
Had the board just had enough of Krzanich?
Comment The sudden and shocking resignation of Intel CEO Brian Krzanich this week over a long-ago affair with a subordinate – banned under company rules – has led to much mirth among Register readers.
"Sounds like he took 'Intel Inside' a bit too literally," quipped SVV. "Talk about taking the wrong branch," a computer architecture guru whispered privately to us. It's not just AMD and Xeon buyers who've been screwed, others joked.
Besides the snickering and tittering in the industry, there was one feeling among Chipzilla staff: disbelief.
Intel announced on Thursday Krzanich had resigned over "a past consensual relationship" with an underling, which broke the company's non-fraternization policy between managers and staff.
Those with inside knowledge of the chip manufacturer we spoke to this week questioned that explanation. Although flings within the ranks are frowned upon, such dalliances happen simply as a result of human nature – yet very, very rarely lead to firings or resignations.
A spokesperson for Intel declined to comment beyond the silicon giant's official statement.
As one well-placed insider told The Register on Thursday, inter-office affairs are not uncommon at Intel, even among senior management. "Otellini was notorious for shagging around, and even boasted about it," the source alleged of onetime Intel president and CEO Paul Otellini, who died last year.
Intel's internal rules allow for relationships between staff who don’t work directly together. However, there is a non-fraternization code for managers – basically, if you're overseeing someone directly or indirectly, and have a measure of control over their job and career, then you're not supposed to get entangled. But it still happens – and is so commonplace, the biz's HR department has a set of procedures for dealing with such situations.
If a boss comes clean to HR about hooking up with an underling, there is the usual stern reminder of policy, and executives move the amorous pair into separate teams so that there is no direct or indirect link. If a manager fails to 'fess up, and HR gets wind of the secret liaison, the situation is more serious. Interviews without coffee, and couples separated at work, yet it is virtually unheard of for someone to be forced out of the business over it.
"You only get in trouble if you try to hide it," one mole told us. "If you shag someone in your reporting structure, you go to HR, and then you as a boss get taken out of the decision making process for aspects of that person's career. I have friends who married their admins, and I've not heard of anyone getting fired for this – either the #metoo movement really made a difference, or this was just a pretext."
Keep it in the family
Let's not forget Krzanich's own wife, Brandee, worked for Intel before they were married. Krzanich, now 58, started work at Intel in 1982, and climbed the corporate ladder until falling off it this week. He married Brandee in June 1998 in Florida. She was a process engineer at Intel from May 1996 to August 1998 before leaving to get her MBA at Harvard. Today, they have two daughters.
It's not known how closely they worked together, if at all. In 1996, Krzanich was given a fabrication plant in Chandler, Arizona, to manage, and may have been her superior, directly or indirectly. Either way, this obviously wasn't a problem back then, and the non-fraternization policy for managers wouldn't be put in place years until later – in 2011.
Judging from conversations we've had, who Brian Krzanich had a relationship with, and when, isn't what triggered his sensational departure: it was that he broke the rules. Rules the chip giant up until now has not rigorously nor robustly enforced, from what we can tell.
Trouble at t'mill
The willingness of the Intel board to accept – or seemingly demand – Krzanich's resignation may instead have to do with the state Intel finds itself in.
Ostensibly, the company is as blue chip as it gets. It dominates the professional IT space, and is still the default option for your common or garden PC buyer. When it comes to data center processors, Intel is king and has near monopolistic levels of market share – roughly 99 per cent of compute CPUs in data centers are Intel badged. But that may change sooner than some expect, a point Krzanich rather embarrassingly made himself earlier this month.
Meet TLBleed: A crypto-key-leaking CPU attack that Intel reckons we shouldn't worry aboutREAD MORE
At the E3 games show in Los Angeles, Krzanich sat down with Romit Shah from trading firm Instinet, and made a rather startling admission. He conceded that AMD's Epyc family of server processors was going to, to some degree, erode Intel's considerable data center market share.
"Mr Krzanich did not draw a firm line in the sand as it relates to AMD’s potential gains in servers; he only indicated that it was Intel’s job to not let AMD capture 15-20 per cent market share," Barrons reported.
AMD and other chip slingers are scrapping over their one per cent server market share, hoping to balloon their position by offering processors cheaper than Intel's eye-wateringly expensive Xeons that are good enough for customer workloads. Yes, AMD once had about a quarter of the data center market, back when the Opteron was the best chip on the block. Then AMD stumbled.
Now Krzanich indicated his rival was back, ready to chew gum and kick ass, and all out of gum. Epyc could be a major headache. And meanwhile, Intel is stuck in a quagmire of 10nm fabrication, haunted by Spectre and Meltdown vulnerabilities, shaken by Nvidia's unstoppable rise with chips for AI and graphics, reeling from the Xeon Phi train wreck, lumbered with sometimes wobbly Puma modem chips, and was unable to get its fad tech – drones, wearables, etc – off the ground.
I wish I could quit you, x86
Intel took a major wrong turn when it came to chip manufacturing a few years ago: it has suffered massive setbacks getting its 10nm process ready for primetime, and the technology is years late. It cannot get the node up to sufficient yield due to going down a dead end, technologically. It is believed Chipzilla is struggling with multi-patterning and its reliance on cobalt.
As a result, the Xeon goliath lost its edge against TSMC and Samsung – they can make chips just as dense with transistors as Intel – and TSMC, Samsung, and GlobalFoundaries are going to give it a run for its money at 7nm.
Process sizes are now a little meaningless: Intel's 14nm is on a par with the industry's 10nm, and Intel's 10nm meets the industry's 7nm. However, while others are gearing up to ship 7nm parts, Chipzilla's 10nm factories have been stuck in neutral for years. Effectively, rather than confidently drive the leading edge in the semiconductor fabrication world, it has allowed its competition to catch up.
During a conference call with financial analysts for Intel's latest quarterly earnings, Krzanich admitted that its 10nm processors won’t be ready for volume production until 2019, and even then refused to say if this would come in the first or second half of the year.
AMD, on the other hand, hopes to emit 7nm parts later this year via TSMC and GlobalFoundries, if all goes to plan. Clearly Intel has not sorted out its production problems, and is in serious danger of falling behind.
Let's look at the other market sectors Chipzilla has tried to fall back onto. Sure, its FPGA business is cooking nicely, thanks to its Altera purchase. However, its attempts to break into the truly mobile market with the Atom largely failed, and Arm still rules in that land.
Similarly in another growth area of the chip market – AI systems – Intel is struggling. After failing to come up with its own designs, in 2016 Intel spent $350m on much-hyped deep-learning startup Nervana Systems to kick start itself in the sector.
However, progress there has been slow and uneven. Nervana's first commercial AI ASIC won't be out on the market until 2019 at the earliest, and Intel's excuse that Xeons will do the job in the meantime is gaining little support among machine-learning software experts. It's also decided to have another crack at the discrete GPU market, although won't have a model available until 2020 at the earliest. It also stumped up $15bn for self-driving machine-learning biz Mobileye, an investment that will take a while to pay off.
As for Intel's efforts in the Internet of Things market, results have been mixed. Lately, its IoT division is pointing in the right direction, financial-wise, thanks to flogging chips and software to retail and video markets.
We cannot deny Intel's data center group is performing strongly, and it is flinging tons of gear at cloud and telecoms providers, but such is to be expected of a monopoly player. The same goes for its PC processor lines: no declines, no big rises, and with average selling prices nudging upwards, more cash through the tills. Intel is undoubtably a moneymaking machine.
But how much longer will that last?
Its silicon photonics didn't live up to the hype. And some analysts and journalists have grown sick and tired of Intel's PR department dumbing down its briefings, often leaving out useful technical details, and instead catering for credulous yet influential YouTubers.
Stock aching, what a man
Much has been made of Intel's share price under Krzanich's reign. The share price has virtually doubled since he became CEO in 2013, from $26 to $53 apiece, so investors should be happy. Yet, it's how that share price was reached that has some worried.
Under Krzanich's rule, Intel has been increasing dividends to shareholders, which, although makes the stock more attractive, costs the company every quarter. Intel has also been carrying out a massive share buyback program, allocating as much as $25bn to the task by some reports.
This financial chicanery used to be illegal in the US as it was seen as a form of price manipulation, however, when the Reagan administration sought to deregulate the financial markets, buybacks were essentially legalized in 1982. Since then they have become a wildly popular form of share support for companies.
Almost all big American corporations buy their own shares. However, the scale of Intel's buyback has raised some eyebrows, and has led to questions about how much the shares will be worth if Intel stops buying.
Then there are also Krzanich's personal share dealings. As CEO, Krzanich has to hold a certain number of Intel shares, and in December 2017 he sold every scrap of stock he didn't have to keep. He had initiated the trade in October that year.
The timing of this looked very unfortunate. In January this year, a few months after that stock offload, The Register broke the news of the Spectre and Meltdown processor flaws that were found in the processors of Intel and others. The CPU designers had known about this since the middle of last year, and were holding off going public until software mitigations and patches were finalized and ready to roll out.
It seems highly unlikely that Krzanich wouldn't have known about these flaws. Intel insisted that the timing of the sale wasn't in any way related to the security disclosures, and we note that Chipzilla's share price recovered from its wobbles in January caused when the details of the Spectre and Meltdown design flaws came to light. However, some staff admitted privately that the optics weren't good.
Too many straws for the camel's back
All of this may add up to why the board was so keen to accept Krzanich's resignation.
After all, Intel isn’t that straight-laced normally. One employee told us that they had heard rumors of people being fired for inter-office relationships but never seen it with their own eyes. That the CEO was forced out seems to stretch the bounds of credulity.
This is Silicon Valley we are talking about. Google cofounder Sergey Brin is still in his job despite, apparently, having a fling with one of his Glass coworkers. Bill Gates married one of his juniors, and no one even thought his job should be up for grabs. HP CEO Mark Hurd ejected amid a probe into alleged work-related sex harassment, and Larry Ellison immediately rehired him as co-CEO of Oracle.
Which suggests our Brian was chucked under the bus by the board. While it's true he's been a very naughty boy, with regard to the floppy drive in his pants, one has to wonder if there wasn't more to the strong-armed resignation. ®
With thanks to Matt Weinberger for the ex86 pun.