Your browser is not supported. Please update it.

17 July 2015

Intel’s handset quest has failed – it looks ahead to IoT, and back to memory

Intel has been girding its loins for almost a decade to take the impact of a decline in PCs, the sector which accounts for half its revenues. Its efforts to create a new business that would compensate for the decline of PC chip revenues have, logically enough, centered on the mobile phone, as this has taken on many of the capabilities of the PC and started to usurp its place. But the attempt to become a mobile giant has failed. And just as Intel has been making some progress, smartphone growth is over too, and it needs to seek out a new source of rejuvenation.

In fact, it has three – highlighted by CEO Brian Krzanich on this week’s second quarter earnings call – all of which are demonstrating more immediate results than the long, hard slog to get into smartphones. One is the internet of things (IoT), where Intel aims to push its silicon everywhere along the device and equipment chain. The second is far more closely related to its existing strengths in servers – chips for large-scale cloud platforms and associated storage.

And the third is memory, especially NAND flash – which presents a neat full circle. Intel’s first commercial product, in 1969, was a static RAM memory, and it was chiefly known as a memory supplier until Gordon Moore, as CEO, took the momentous decision to refocus the business almost entirely on its microprocessor offerings.

Now memory is important again, and certainly current CEO Krzanich needs to think in terms as radical as his illustrious predecessor did. He has already shown signs of willingness to take new directions, notably with his decision to go into the foundry business and form new alliances in China. Now he has made it clear where he sees his three main sources of growth in the next few years – and none of them is the smartphone.

His chosen three do feed off one another to some extent, with the rise in IoT devices feeding the need for huge big data engines and more efficient memory. That, combined with Intel’s steady progress in carrier network and IT infrastructure, gives it a promising ability to play in many areas of the emerging cloud and IoT platforms.

But meanwhile, has the smartphone adventure all been a waste of effort and of many billions of dollars? Intel lost over $7bn on its mobile activities in 2013-2014 alone, and now it appears to have missed the boat. According to Gartner figures, semiconductor sales are in a two-year slump, and this is mainly down to weak demand for PCs, but also for smartphones.

So that bubble has burst, as seen in the Q215 results. The company no longer breaks out mobile and wireless revenues, but includes them in a combined client computing group – a logical move, given that the opportunities for Intel lie in converged form factors like Chromebooks, but also a potential way to hide the worst of the mobile performance from prying eyes. Overall, in Q215, that client group saw sales fall 14% year-on-year to $7.5bn. This is challenging for a company so heavily tied to PCs still – the client group generated more than half of its total Q2 revenues of $13.2bn. The giant’s total net income was $2.7bn.

Yes Intel has made progress in tablets and established a strong position in Chromebooks. Tablet sales were sufficiently strong for it to break them out – up 11% year-on-year in unit terms to 9.9m. However, the figures clearly show that tablets and smartphones are still not fully compensating for PC decline, and these hybrid form factors have low volumes, compared to the PC market it is used to. There is still the opportunity to take a reasonable share of the devices – whatever shape or size they may be – which will replace the PC in the coming years. But the prospect of being a significant challenger to Qualcomm and MediaTek in the handset space is surely over, unless Intel can leverage its new Chinese alliances, with Spreadtrum and Rockchip, extremely effectively. These center around the Atom/modem system-on-chip for smartphones, SoFia, but in recent weeks there have even been shadows over that. Intel was reported by DigTimes to have delayed the LTE version of SoFia – critical to any chance of gaining significant smartphone market share, and to its Chinese plans – until April 2016, which would push out the revenues from that product, and risk being fatally overtaken by Qualcomm and MediaTek.

With other chips reportedly suffering delays, there are question marks over whether Intel’s software capabilities are keeping pace with its famously advanced manufacturing processes, and whether it is able to turn out new chips, at high volume and low price points, at the speed required by the consumer devices sector. This may be worsened by the challenges of moving to the 10nm process. CEO Brian Krzanich said Intel’s chips are likely to stay in the current leading edge process, 14nm, for longer than usual for a new node, and the company will slightly delay the purchase of some 10nm equipment. “The lithography is continuing to get more difficult as you try to scale,” he pointed out.

This statement, of course, has bigger implications – Moore’s Law, the observation (by Intel’s Gordon Moore) which has governed the pace of chip and device design cycles for decades, is slowing down. Originally, Moore said the number of components per integrated circuit would double every year, though in 1975 he revised that to two years. Intel has said that the pace slowed from 2012 (the 22nm mode) and on the results call Krzanich said “our cadence today is closer to two and a half years than two”. This has been predicted by Moore himself, who said earlier this year that his law would die within the coming decade.

That is because of the increasing difficulty of packing more transistors in, but there are also economic considerations – a ‘second Moore’s Law’, less often cited, covers the way that the cost of semiconductor manufacturing fabs will increase exponentially over time, with each new generation of process technology. At some point, it becomes unaffordable to sustain that cost in capex on such a regular basis, especially if the industry sees diminishing returns from the process advances. The slowdown is “a natural part of the history of Moore’s law”, said Krzanich, and Intel says the 2.5-year cycle will hold at least until the 10nm node emerges in late 2017. Instead of its usual ‘tick tock’ pattern – a process advance one year, a new microarchitecture the next – Intel will shift to a ‘tick tock tock’ cycle, with two architecture advances for each process upgrade. So it will introduce an enhanced 14nm microarchitecture, and associated processors, called Kaby Lake, in the second half of 2016, building on its current new architecture, Silverlake.

Of course, since Moore’s Law is not an actual physical law, but an observation/prediction, other companies may be able to sustain a two-year cycle for longer and so steal a march on Intel. IBM is certainly attempting this in intellectual property terms with its recent display of a 7nm process in its labs. Intel itself has not resigned itself entirely to a longer cycle. Krzanich said: “We’ll always strive to get back to two years”, but added that the most important thing was to have realistic expecations. “The feedback from customers said, ‘We want you to be predictable’,” he said. “That’s as important as getting to that leading edge.”

Given the second Moore’s law, it will be increasingly hard for any firm to outdo Intel on process advances, and this will be another factor shifting innovation and growth away from the constant quest for rapid-fire improvements in microprocessor performance and power consumption. Parallel architectures, gaining performance by using large numbers of mainstream processors together, will become increasingly important to achieve processing gains, while commercially, many firms will focus on applications, some of them in the IoT, which do not necessarily need the smallest geometries.

That will be as true for Intel as many others – as it tries to leap these hurdles, the PC business continues to look bleak. Krzanich did his best to be optimistic about PCs, especially the impact of Windows 10 which, in combination with the 14nm Skylake processor, will “bring excitement to client computing”, as he put it. “Over the long haul Windows 10 will be a boost to the PC market and will take some of the Windows 8 negativity out,” he added. However, few expect W10 to drive a major refresh cycle in the same way as previous releases, especially given Microsoft’s promise of free upgrades to current W8 users.

These patterns explain Intel’s excitement about IoT chips, the newest source of device-side growth potential and a market where Intel has made a far more confident start than it did in mobile – and without a single powerful incumbent like Qualcomm blocking its path. The IoT segment, while still small, saw a 4% increase in sales to $559m in Q215, and CFO Stacy Smith predicted that it would jump again in Q3 – by 8% quarter-to-quarter.

However, it will be a long time before the IoT comes anywhere near to balancing out the revenues lost from PC downturn. According to Gartner forecasts, the IoT will account for $43.5bn in semiconductor sales by 2020, but that will be only just over 10% of the total value of the chip market. IoT devices will have to remain simple in order to hit the required price points and so will be designed to use a few cheap chips. For instance, analyst Michelle Reitz said that wearables have little silicon content and will only account for 1% of chip sales by 2019, even though their shipments will have grown to about 500m units by then (bad news for Intel, which has been investing heavily in start-ups making smart eyewear and watches). The real revenue potential of the IoT is in software and services, and (better news for Intel) the additional data center capacity it will drive.

Nevertheless, Intel is looking to IoT activity, along with the other two areas selected by Krzanich, to cushion it from the effects of PC decline, not the smartphone into which it has plowed a decade of effort for just that defensive purpose. Smith said Intel now gets 40% of revenue and 70% of operating profit from this trio of rising stars and that will enable it to see only a 1% fall in full year revenue, when the PC market will be falling more steeply than expected, in the high single digits.

Intel’s aggressive expansion in the networking infrastructure market is starting to pay off. It has been investing in the technologies which carriers will deploy to handle high speed wireless and wireline broadband, from high end switch-chips to Cloud-RAN to gateways. In Q2, network infrastructure and cloud servers were the primary drivers of growth in the data center group, which reported a 10% year-on-year revenue hike, reaching $3.9bn.

A jewel of increasing luster in this area is Intel’s Wind River subsidiary, which has evolved from a real time operating system provider to create software frameworks for virtualization, SDN and the internet of things. This week, it announced that its Titanium Server telecoms infrastructure software has been adopted by Nokia as an NFV Infrastructure component in the Finnish firm’s recently announced AirFrame carrier IT platform. This shows how Intel will often infiltrate difficult markets not just with its server chips – as IT technologies spread into the networks themselves – but with the products of its software acquisitions, such as Wind River and, in security, McAfee.

By contrast with the service provider space, the chip giant said it was not looking for higher enterprise growth in the second half of the year, after a leap in Q115, but believed software defined networking and storage would deliver buoyancy. Nevertheless, Krzanich warned on the analyst call that “the enterprise market is weaker than expected, it’s more macroeconomic driven”.
That, and PC weakness, contributed to a weakening of the full year forecast, especially as Intel will embark on the “steep part of the curve” in terms of spending on the 10nm process technology. It now expects about a 1% fall in full year revenue compared to 2014, whereas previous forecasts were for flat sales. Krzanich insisted: “We’re looking in the long haul, don’t get caught by a quarter.”

 

 

Intel, and US, should not dismiss a Tshinghua bid for Micron too hastily

Memory was a highlight of Intel’s Q215 results, and particularly its joint venture with Micron for NAND flash memory, which delivered 40% growth in that business. Micron is now reported to be the subject of a takeover bid by China’s Tshinghua Unigroup, which also owns Spreadtrum and in which Intel has a stake. The immediate reaction might be to see this as another potential blow to the US giant, but on deeper reflection, it might actually be an opportunity for Intel to extend its tentacles more deeply into the Chinese ecosystem.

Nobody has yet confirmed any offer or talks between Tshinghua and Micron, but the issues the report raises will certainly come up repeatedly as the chip market consolidates and the Chinese government and vendors look to establish a powerful homegrown industry.

Tshinghua, according to the Wall Street Journal, has presented Micron with a letter outlining plans to make a $23bn bid. Such an acquisition would further shift the technology balance of power towards China, but would face high regulatory barriers.

In some ways, the report reflects the same trends which have driven other recent bids, such as Intel’s for Altera or Avago’s for Broadcom. There is the need to consolidate, especially for any companies targeting the stagnating PC sector and even the maturing smartphone space; and the need to acquire cutting edge technology for the growth markets in connected devices and post-PC form factors.

And though Tshinghua’s bid has been widely cast in the context of east/west trade rivalries, in fact, it reflects the reality that the device chip market has to be global in scale in order to sustain growth, and that incumbent vendors – Korean and Japanese as well as American or European – need to cultivate alliances with their Chinese challengers, not build protectionist walls.

Intel – which has a significant joint venture with Micron centered on advanced NAND flash memory (critical for mobile devices) – has also made an investment in Tshinghua, which has brought it a manufacturing and distribution deal for its SoFIA smartphone system-on-chip. If Tshinghua buys Micron and Intel retains the JV beyond its renewal date of 2018, that will just bolster the alliance, creating a cross-border powerhouse which will contribute far more to Intel’s chances in the mobile and IoT space than old-fashioned resistance to the rise of Chinese challengers.

Current chip giants still have many cards to play, in terms of entrenched customer bases, R&D resources and manufacturing process technology, over Chinese rising stars. The Chinese companies’ aces are still largely in their hands, and western alliances (whether acquisitions or partnerships) will be important to allow them to play those cards.

Companies like Tshinghua’s own Spreadtrum, a mobile SoC and modem supplier, still lag behind Qualcomm and others in terms of advanced technology and customer relationships. China consumes one-third of the world’s semiconductors but produces only 12% of them. There is a chasm to be crossed, and the US giants can either pull up the drawbridge, or improve their own position in China – and the emerging markets, such as Indonesia, which will naturally turn to Chinese suppliers too – by creating a fully global industry.

So we shall see how Intel, and other powerful lobbyists, respond to the Tshinghua bid for Micron. If it is passed, it would fill a major gap in China’s semiconductor capabilities, which are currently poor when it comes to advanced memory. It could also inject added resource into Micron, which despite being the second largest US chipmaker with $16.8bn in 2014 revenues, has experienced disappointing growth in recent times, and has fallen behind Samsung in developing some key memory technologies for the mobile and IoT markets.

A new injection of resources and energy could be good for Micron and the memory industry as a whole, but unfortunately for memory buyers, the deal is not likely to be regarded in that light by US regulators. It is sure to be case in traditional terms – loss of a US company; a further slide of US power and profit towards China. If this were merely a story of deals between commercial companies, it might be easier, but China’s attempt to build a world leading chip industry is heavily government backed and therefore seen to be a matter of national interest for both countries.

Last year, the Chinese government announced a $20bn fund to develop its semiconductor business, a move perceived, in the context of increasing US-Chinese trade and security tensions, to be a direct threat to US industry – never mind that, in the background, Intel, Qualcomm and others have long accepted that their growth will rely on expanding their own activities in China, often through partnerships. And China is not just thinking about jobs at home – it is envisaging global reach (Micron has no Chinese fabs and that would not necessarily change very quickly under the new ownership as it has only recently been investing in factories in Japan and Singapore, and other non-US locations).

Chinese government funds will surely come into play to fund the Micron acquisition, since memory is seen as a strategic requirement – according to the Wall Street Journal, officials have also been in talks with Powerchip of Taiwan about a possible purchase.

This is the fourth Chinese bid for a US chip firm so far this year, with deals for Omnivision and ISSI still awaiting US approval.