- The Old Money Machine Dies
- Winners and Losers in the Coming Semiconductor War
Bold call: by 2025, up to 40% of listed companies operating in mature (and not so mature) sectors will be ripped apart by the coming tech takeover. Any survivors who fail to adapt will fade into corporate history…
Quite simply, the impending impact of a cluster of technologies — sensors, big data analytics, the cloud and especially AI — means that companies who don’t ‘get it’, and adapt over the next ten years, starting now, will be gone.
Company life cycles are already shortening — the average age of S&P 500 companies was 40 when the index was formed, it is 12 years today and falling.
Consider what’s underway at the very heart of tech land, the semiconductor industry.
The End of Predictable Progress
The $350 billion global semiconductor industry is shrinking almost as fast as the geometries of the transistors it fabricates.
Since 2013, a third of the Philadelphia SOX index has vanished through bankruptcies and largely defensive M&A, and a further 20%-30% are expected to go by 2018.
It’s not a question of too many companies competing for too few ‘sockets’ as growth stalls. The industry is at a highly strategic inflection point.
Since 1991, the International Roadmap for Semiconductors (ITRS) has tracked Moore’s Law – the oft quoted prediction of price/performance of semiconductors doubling every two years–and the operation of the standard industry model of making mostly general-purpose microprocessors and memories and selling them in huge volumes. These roadmaps helped convert Moore’s Law into a self-fulfilling prophecy.
But this year’s ITRS report slammed on the brakes and then exited the highway…
It calls time on ‘more Moore’, partly because we are reaching the limits of what can be etched. As geometries approach atomic scale, on-chip ‘thermal budgets’ get breached and quantum weirdness abounds — causing the cost of new chips and fabrication to soar.
As the quip among the technorati goes, ”We run out of money before we run out of physics.”
But also because there’s a fundamental shift away from general purpose computing.
Rather than making chips and letting applications follow as happened for fifty years, says the ITRS, the industry now needs to start with the applications — from smart phones, robots, cars, and data centres on the Cloud– and work downwards to figure out exactly what chips the customer needs, embedding ever more “bang per watt”, sensors and security in their devices.
A Sudden Shift in Power
This betrays a deep anxiety in the semiconductor industry.
For 70 years programmed computing has been based more or less on the Von Neumann model of the calculating engine supported by enough memory for stored instructions. The next decades will be about memory systems and machine learning in an exploding online world that’s less about raw computing, more about ‘search’ and vast sensor-fed datasets.
More and more design of high quality silicon and chipset architecture will be done by or with the heavy participation of the mega-customers themselves. And these will take the form of ‘exclusive’ applications, even algorithmic specific chips and SoCs .
Facebook is testing its own chip designs based on the workings of the neocortex and deep learning under the leadership of Yann LeCun, its superstar head of AI.
Apple is the stand-out among the ‘captive, fabless’ chip operation. Its ARM-based A Series application management chips, the core of iPhones, far outperform rival offerings from such ‘merchant market’ heavyweights as Intel, Qualcomm and Nvidia.
And last year Amazon acquired low power chip specialist Annapurna for $300 million and announced this year that it will sell Annapurna chips, just as it sells books and diapers in the merchant market. It already counts Asus and Netgear among its customers.
The industry will be increasingly shaped by a consolidating customer base of big players — Apple, Alphabet, Amazon, Huawei, Baidu, and Tier 1 automotive system suppliers such as Continental and Delphi — needing proprietary solutions to stay ahead and build moats around themselves.
The industry shake-out and revamp is not nearly done.
A major consequence of this power shift is that the future lies more and more with ‘fabless’ chip design companies such as ARM, Broadcom and NXP and ‘non-captive’ foundries such as TSMC, Global Foundries and UMC knocking out the chips.
‘Captive’ foundries at Samsung, Intel and SK Hynix are for hire to outsiders as well. Even during the current slow growth phase there’s a shortage of foundry capacity to fabricate at 20nm geometries and below–with TSMC, Intel and Samsung alternating to share pole position.
China Bulks Up
This is especially significant given China’s total lack of indigenous device manufacturers, as it drives to grow its consumer electronics sector by going global. China accounts for half global chip consumption but supplies less than 10% from indigenous sources.
It is on an intensive drive, under fiat from the Politburo and the State Council, to upgrade the country’s semiconductor capabilities and autonomy by whatever it takes — foreign acquisitions, winning bids for scarce foundry capacity, screwing down the Intels and Qualcomms of the world to very tough ‘technology transfer’ deals and hefty investments in return for access to China’s huge market.
3 Big Short Term Calls
#1 By 2018, half the companies that made up the SOX index in 2013 are likely to be gone
After the frantic $100 million plus of M&A last year there is more shake-out M&A to come. Linear Tech, Analog Devices, Maxim and Marvell are on most lists to be swallowed, with a now small cap AMD on its way to Chapter 11.
This is indicative of how much disruption we will see as we gravitate towards an AI-directed society. Moore’s Law has mapped the steady exponential progress of technology for decades, but as we abandon traditional hardware and cede control to intelligent machines, huge legacy companies will be exposed.
In the semiconductor industry, Intel has a clear need to rebalance its portfolio, as does Texas Instruments. We’ll see aggressive moves by Broadcom, and the Chinese. The latter, led by SOE Tsinghua Unigroup, acting as a ‘white glove’ for the Politburo, already made passes at Micron and Western Digital last year but withdrew because it feared problems with the US authorities.
The running assumption is that China has at least $150 billion earmarked to invest in upgrading its chip sector. Meanwhile, Tsinghua is close to acquiring Taiwan SoC design company Media Tek. Other Chinese State interests such as Hua Capital — which acquired OmniVision — will be scouting for foreign semiconductor assets, as will Huawei, Foxcom, and Baidu.
#2 The Foundry Sector is a Sellers’ Market
The complex geopolitics of the world’s scarce advanced foundry capacity and fierce bidding for that capacity make the foundry sector a sellers’ market. Apple has chosen TSMC over Samsung to make it’s A Series 8 chips suggesting that TSMC is the sector leader.
An intriguing narrative could develop over Abu Dhabi owned Global Foundries — which acquired fabs from AMD and IBM. Due to the fiscal impact of low oil prices, Abu Dhabi may need to dispose of Global Foundries in part or as a whole. Chinese interest?
#3 NAND, 3D Flash and Xpoint are the Future
The main product line interest will centre on the new fast/high density 3D ‘flash’ and NAND memory chips, and soon the much heralded Xpoint technology due out later this year — a collaboration between Micron and Intel.
These are technologies that are fit for purpose for the new online world of searching massive volumes of fast moving data. Volatile, veteran US memory maker Micron is the ‘star turn’, already very much in Tsinghua Ungroup’s sights, and it would make an obvious fit for the diversifying Intel.
The Old Money Machine Dies
In days of yore I worked for Regis McKenna, the legendary Silicon Valley PR and marketing consultant, who was Steve Jobs’ Svengali and played a major role in kicking off the microprocessor revolution.
In a discussion I attended between Regis and the Intel founders, yes, that included Gordon Moore of Moore’s Law, I heard that the chip that defined and enabled an age came about by default.
A Japanese calculator company Busicom wanted a faster logic chip and hired a fledgling Intel to design it. Busicom rejected the Intel design and Intel didn’t know what to do with the chip.
Regis suggested they place a splashy ad in electronics publications announcing a new era of integrated electronics – a ‘micro-programmable computer on a chip’.
It was nothing of the kind of course but they came and did things and the rest is history.
In the 1990s at the height of the PC boom Intel became what Goldman Sachs called the ‘greatest money making machine on the planet.’
Now the world has moved on.
The next stage in technological evolution will be about intelligence, rather than brute force.
The Great Reveal is Coming…
I see a new ecology emerging — a ruthless, cutthroat ecology that thrives on the competing machine intelligence of the Tech Titans.
In fact, I believe these companies have yet to show their true identity.
Amazon isn’t about e-commerce. Google isn’t search. And Facebook isn’t social networking.
In my next briefing, I will look at how these companies will soon reveal their true selves, what that will mean for society and your investments.