Just as it’s no longer enough to be just a mobile operator or a TV provider, but carriers must support a whole raft of services, so the chip designers which enable their networks and devices have to move to broad platforms too. This is sparking the biggest wave of consolidation among chip suppliers for many years. Just weeks after news of Freescale’s plan to buy NXP, Avago announced its shock acquisition of Broadcom last week, and was quickly followed by Intel’s confirmation that it would take over Altera for $16.7bn.
All of these deals, and a long string of smaller ones over the past couple of years, reflect the overriding need for chip suppliers to achieve huge scale, as margins in all branches of networking and devices are squeezed, and to assemble the broadest platform possible.
It must be broad in terms of technology as high end products become more complex and diverse – to secure a significant position in smartphones, for instance requires all kinds of different specialized processors for the high end models and low cost SoC (system-on-chip) offerings for the cheaper ones; while a convincing wireless infrastructure platform requires not just SoCs but FPGAs (field programmable gate arrays).
A large player’s portfolio must also be broad in terms of target markets – for instance, cellular giants like Qualcomm are offsetting the falling prices in their core products by expanding into newer areas like the internet of things (IoT), while infrastructure chip providers are keen to develop end-to-end offerings, from cloud servers to base stations and small cells, so that they can bid for a bigger share of a large vendor’s business, even if the price of each individual element is falling.
This is the context behind the most recent mergers. Intel is adding Altera to a series of other purchases in the telecoms market, including Infineon’s wireless unit, LSI’s Axxia division and Mindspeed’s Picochip small cell activity. The world’s largest chipmaker was
in talks with Altera, the FPGA leader, back in March but apparently its offer price was spurned. However, Altera shareholders may have been more keen on the deal after the firm reported poor quarterly results on April 23, with Q1 revenues down 6% year-on-year and net income reduced too.
The deal reflects Intel’s urgent need to reduce its reliance on stagnating markets, notably PC processors, and consolidate its strength in servers (still 98% share, but under the first credible attack in years, from ARM). Those servers are no longer just about enterprises – Intel is pushing its server processor technology into the carrier network and wireless infrastructure spaces, riding on the interest in virtualization, including Cloud-RAN. It has made several acquisitions recently, including those of Axxia and Picochip, aimed at improving its capabilities in base stations, network processors and C-RAN (even at the cost of introducing ARM-based platforms into the fold).
Altera makes programmable FPGA chips for base stations and other infrastructure, and leads this segment, ahead of Xilinx and Lattice, especially in the high end sub-sector which comprises over half the market. In 2013, the company became the largest customer for Intel’s fledgling foundry business, signing a deal for the US giant to manufacture new top end parts with its cutting edge 14nm FinFet process. At the time, there was also considerable speculation that Intel would need to expand its FPGA expertise, probably via acquisition, in order to take a lead in base stations, and particularly in Cloud-RAN.
This is because FPGAs are increasingly needed, alongside general purpose processors, to satisfy the demands of performance-intensive infrastructure such as cloud servers and carrier networks. FPGAs can run specific tasks very quickly and are increasingly seen in data center platforms used by Microsoft and Baidu alongside Intel chips. Altera has been the leader in penetrating this high growth segment, and if Intel acquired it, there would be the chance to secure those revenues for itself, and also to integrate CPU and FPGA more closely, to gain competitive advantage against rivals.
In particular, that would cut off a valuable source of FPGAs for ARM-based competitors, though that in turn could boost Xilinx, and possibly make it a target for acquisition too (its shares leapt 6% on the reports of the Intel talks with its arch-rival).
As well as servers, FPGAs are an important element of base stations, and of the emerging virtualized version, Cloud-RAN. Intel has been prototyping a base station platform built around its Xeon processor with accelerators optimized for signal processing. Prototypes based on FPGA chips are being used in China Mobile’s huge C-RAN market trial, which aims to virtualize network functions on an x86 server in order to control a large number of small and macro cell sites in a flexible, centralized way.
At this year’s Mobile World Congress, Altera and Intel were part of China Mobile’s demonstrations of C-RAN based on the NFV specifications. Altera signed a strategic collaboration with the China Mobile Research Institute (CMRI) in 2014, focused on the future needs of 5G with regards to virtualization and FPGAs.
Intel has also talked about producing a full base station platform, though to date it has mainly worked with partners to get x86 processors into cell site equipment, as seen in its alliances with NSN to create the RACS/Liquid Apps offering, and Cisco/Ubiquisys for the Smart Cell.
The Altera deal could pull all these strands together, capitalize further on the important Axxia takeover, and give Intel a strong position when the C-RAN market gains scale (which, despite major trials by China Mobile, Telefonica and others, is unlikely to be until the industry gets close to its ‘5G’ stage, around the end of the decade). CEO Brian Krzanich said Intel would ship integrated products starting in late 2016 for servers and some unspecified embedded systems, containing x86 and FPGA die in a single package. Those will be followed “shortly” by products that merge the two types of chip on an SoC.
The deal could also, according to Krzanich, improve the company’s cost base and competitiveness, by moving Altera’s manufacturing to Intel plants. As noted above, that process had already begun under a foundry deal, and this aspect of the acquisition does highlight the differences between Krzanich and his predecessors. A CEO with his roots in chip manufacturing, he is determined to leverage Intel’s famous processes not just to advance its own products, but to create a foundry business.
However, to fulfil the potential of the Altera acquisition, Krzanich will have to be different from his forerunners in another way too – in his ability to steer a merger effectively and achieve the targeted results. Intel’s track record in M&A is miserable, and although it often makes good choices from a purely strategic viewpoint – Wind River for virtualization and IoT, McAfee for chip-level security – its ability to impress shareholders with the tangible returns on its purchases is poor.
“Given the recent track record, we wouldn’t be optimistic that Intel’s purchase will result in a slew of new products in the market,” Bill Kreher, an analyst at Edward Jones, told Bloomberg. “You only have so many chances to change the direction of your company. This was Intel’s bullet,” he added, pointing to the negligible effect that the Infineon Wireless deal has had on Intel’s handset market position or profits.
In hard financial terms, Altera brings just $1.93bn in sales (and fairly stagnant revenue growth) to add to Intel’s $55.9bn (though these revenues are in difficult areas for Intel. In 2014, 44% of Altera’s sales were in the telecom and wireless segment, and it will double its new parent’s presence in embedded chips for communications systems). Its profits have declined from $782m in 2010 to $472m last year. Many analysts believe that, at about 25 times Altera’s EBITDA last year, the offer price is overvaluing the smaller firm (the multiple is similar to that in Avago’s deal for Broadcom, which was also seen as highly priced, though in Broadcom’s case, the valuation is more justified).
“We view the acquisition of Altera as a good strategic asset for Intel, albeit at a very high price,” Ross Seymore of Deutsche Bank wrote in a client note. “By our math, the deal appears ~3% accretive in 2016…the accretion is meaningfully less than if Intel had used the same amount to repurchase its shares (~10% accretion)”. Others point out that, since Altera has been efficiently managed and has a non-overlapping product set, there will be few cost efficiencies to be made.
Intel claims it could generate 7% revenue growth from the Altera takeover, a forecast Wall Street observers questioned. Krzanich’s reasoning is that the combination of Xeon and FPGA will be worth far more than the sum of its parts.
He insists that merging the chip platforms greatly increases the addressable market, which for standalone FPGA server coprocessors is currently worth about $1bn. He said the combined offerings would have twice the performance of current Xeon chips and that, by 2020, 30% of workloads in the cloud would require this capability, with web giants looking to run multiple, very different workloads on the same servers (facial search and encryption were the examples he chose). In addition, the workloads will change over time – an integrated chip is “ideal” for that scenario, he insisted. “That is quite a bit bigger than a billion dollars,” he said, but refused to put a more precise figure on the target market.
For many analysts, though, the server FPGA market is too fledgling to inspire great confidence – and Intel has form in getting carried away by a technology vision which then fails to translate into huge numbers of dollars. After all, Intel was talking about integrated microprocessor/FPGA products five years ago, when it combined an Atom Stellerton chip and an Altera FPGA on a single board, targeting some embedded systems. The resulting products are now listed as end-of-life.
But CFO Stacy Smith insists the timing is better now, and that Intel is on a learning curve, saying: “You can think of Stellerton as an evolutionary path — what we are looking at with Altera now is significantly more evolved.”
And Krzanich feels there will be other, as-yet unexplored opportunities for a company which leaps in early into integrated chips (and after all, that is something few will have the ability to do). In the internet of things, there may be space for co-packaged chips in specialized areas such as industrial automation or connected car control, which currently rely on high performance ASICs or dedicated chips. In the IoT, “we are still trying to understand if co-packaged parts make sense…we haven’t finished that analysis, it’s so workload specific,” said Krzanich on the call. “An integrated part is significant, the question is will a co-packaged part be any good.”
As with its purchases from LSI, Infineon and Mindspeed, Intel is acquiring some ARM-based products with Altera, and insists it will continue to support those, while not significantly changing its new unit’s business model (that was “one of the things we bought,” said Smith). It also confirmed it would continue to make FPGAs for its original foundry customer, Achronix, and move it to the 14nm process.
One interesting comment came from Handel Jones, principal of International Business Strategies, who told EETimes: “We thought Broadcom was a better acquisition – that would have given them additional capabilities.” Broadcom may indeed prove to be the one that got away for Intel, since it is to be acquired by Avago Technologies for $17bn in cash and $20bn in stock, to create the world’s sixth biggest semiconductor vendor. The company claims this will create the world’s leading diversified communications semiconductor company with revenues of $15bn.
This deal, while unexpected, is more characteristic of the current wave of chip mergers, in that cost efficiencies and economies of scale are at its heart, unlike in the Intel/Avago deal. The acquirer pledges that the merger will generate at least $750m in annual cost savings within just 18 months, increasing the efficiencies which are increasingly necessary in the consolidating chip segment. However, the proposed cuts will mean job reductions, including on the engineering side.
In the first quarter Avago had a gross margin of 51% but it yielded net income of $596m, on revenues of $1.66bn, while Broadcom had gross margins of 54.4%, a net income of $390m on revenues of just over $2bn. Avago operates in diverse markets like fiber optics, Gigabit Ethernet, LEDs, light sensors, and RF chips, and has recently acquired both Emulex, which took it more deeply into networking chips, and LSI Logic, which added enterprise storage silicon and more networking to the mix. In both cases, it made significant layoffs, and it sold LSI’s Axxia unit on to Intel.
Henry Samueli, Broadcom’s co-founder and chief technologist, will become CTO of the enlarged firm, which will retain the Broadcom name. In an analyst call, he said: “Broadcom leverages a lot of central IP development across product lines since we are so SoC(system-on-chip) focused. We will be able to incorporate some of that in the Avago model because there are some SoC products there but their non-SoC businesses operate very independently.” He added: “After the deal closes we will see where we can leverage the centralized method and where we can’t.”
It is not impossible that Broadcom might receive a counter-offer, with Intel and Qualcomm both potential candidates, but Wall Street sees the Avago price as a good deal, and the takeover has the strong support of Broadcom’s top brass.
Avago and Broadcom supply chips to the same customers, and the same products, in many cases, but the headline messages around the proposed deal – which is unlikely to close before next April – are so far mainly about efficiencies rather than new opportunities, and it is early days for the companies to discuss in detail how they might leverage the overlaps in their client bases.
Hock Tan, CEO of Avago, said there was “almost zero” overlap between the two companies’ product lines, but added that, once the deal is completed, he will review the Broadcom portfolio in detail to identify any businesses which he wants to sell off – a process with which he is familiar, having offloaded some units of his last big acquisition, LSI, to Intel and Seagate. And there may be a challenge, as well as an opportunity, in bringing together Avago’s and Broadcom’s respective networking portfolios in a creative way – Broadcom is a major player in Ethernet switches and network processors while Avago recently acquired interconnect and storage products from Emulex, LSI and PLX.
Most analysts were positive about the proposal, acknowledging that in an industry experiencing slow growth rates (5-7% a year), acquisitions are the main way to grow for larger players. Other recent mergers include NXP and Freescale, plus Cyprus buying Spansion for $1.6bn, MicroSemi acquiring Vitesse Semiconductor for around $389m, and Lattice Semiconductor buying Silicon Image for $600m.
Broadcom may regret that it exited the smartphone baseband business so quickly, having acquired Renesas Mobile and then backed away just months later. One of Avago’s growth areas is integrating its FBAR filters into combined RF front end modules. These are important components as LTE smartphones increasingly need to address a large number of frequency bands while keeping power consumption down.
The transaction has been unanimously approved by the boards of directors of both companies, as well as a special committee of the independent directors of Broadcom.
Avago Technologies was the new name for Agilent Semiconductor Products Group, after that was created in 2005 by a buyout, funded by private equity groups Kohlberg Kravis Roberts and Silverlake Partners. Agilent itself was the semiconductor arm of Hewlett-Packard.
NXP prepares to sell base station RF power unit:
Ahead of the completion of its deal to merge with Freescale, NXP has signed a deal that will facilitate the sale of its RF power business to Jianguang Asset Management for $1.8bn. This unit is one of the leaders in high performance RF power amplifiers for base stations, though it is also expanding into other applications like industrial lighting and automotive electronic ignition systems.
Under the agreement, the business unit, with its 2,000 NXP employees and its management team, are to be transferred to an independent company incorporated in the Netherlands, which will then be 100% acquired by JAC Capital. All relevant patents will be transferred, as well as a manufacturing operation in the Philippines.
“The creation of a new company focused on the RF power amplifier market is a ground breaking transaction for JAC Capital and a great deal for our customers. Although we would have expected a higher valuation in a regular disposal, JAC Capital’s ability to support continued growth and development of the business and its ability to sign and close a transaction rapidly was a key factor in enabling the best outcome for our customers and shareholders, as well as supporting the closure of the merger with Freescale Semiconductor,” said NXP’s CEO Richard Clemmer.
Project Loon:
While Google’s wireless play in the USA, Google Fi, has only just begun, its balloon-based Project Loon has been running test for quite some time now. The latest update provided from Google says that a single balloon can spend more than 100 days in the air before it is landed, refurbished, and relaunched, and that while it is in position, it’s able to provide 10Mbps connections and positional accuracy at less than 500 meters to users on the ground – within an area footprint equivalent to Rhode Island (3,140 square kilometers).