The semiconductor industry’s consolidation continues, with Marvell making a play for Cavium. Marvell has offered $6bn in cash and stock in the latest major M&A move in this sector – only a week after Broadcom made a play for Qualcomm.
If both the acquisitions were to go ahead, Intel would face two far stronger challengers in the server, cloud infrastructure and telecoms hardware markets, since both predators would gain greatly enhanced platforms for these spaces. And the two developers of ARM-based processors for the telecoms space, particularly Cloud-RAN, would be snapped up, creating uncertainty as to whether the powerful platforms which are emerging from Qualcomm and Cavium would be as well supported by the new parents.
So, as Intel tries to defend its data center stronghold and expand in telecoms systems such as Cloud-RAN, it may be able to use a period of uncertainty and adjustment for its rivals to consolidate its own position.
Cavium would be an asset to Marvell, which moved early into the ARM-based server processor space, but has made less headway there. Cavium has a broad range of processors which span low end to high end servers and telecoms infrastructure, including wireless platforms like small cells. It has succeeded in pushing an ARM-based system-on-chip to Intel-like levels of performance with its Thunder X offering, and it has made a point of focusing heavily on Cloud-RAN and other telco applications which are promising to be deployed in scale from 2018, and which are particularly demanding to support, raising the barriers to entry for other ARM players.
A couple of analysts speculated that Marvell might shut down Thunder X, having pulled back, at an earlier stage, from its own ARM-based server processor. The firm has a tendency to focus on keeping costs low, and on products which have already achieved good market share, like Cavium’s most successful offering, the Octeon family of embedded processors. These are third in the market after Intel and NFV’s products, though far behind those two leaders, according to Wheeler, who believes the future of Thunder X is uncertain.
It would be a mistake to axe Thunder X. It is a significant feat to adapt the ARM architecture for large-scale servers, or for demanding use cases like Cloud-RAN. Several attempts, including Marvell’s, have failed along the way, and even Qualcomm has taken several years to get its Centriq server processor family right. Undoubtedly cloud infrastructure in general, and the telecoms network aspect of that in particular, are growth areas in a chip industry which is short of sources of expansion, and with Qualcomm fighting a takeover battle, the timing could be right for Marvell, courtesy of Cavium, to ensure that Intel does not have the space to itself.
More broadly, the deal would increase Marvell’s addressable market to over $16bn, deepen its portfolio in areas where both firms operate, and help it reduce its reliance on its traditional storage controller business. The two companies’ areas of focus are largely complementary – Marvell’s main offerings are in storage controllers, Ethernet switches and wireless connectivity chips; while Cavium also has storage and networking chips, as well as operating in the security, multicore processing and storage connectivity fields. The combined company would have total annual revenue of about $3.4bn.
The deal, which has been rumored for some weeks, would also inject new confidence into Marvell’s shareholder base after an accounting scandal last year led to the resignation of the company’s founders, CEO Sehat Sutardja and president Weili Dai. The new CEO, Matt Murphy, said in a statement: “This is an exciting combination of two very complementary companies that together equal more than the sum of their parts. Together, we all will be able to deliver immediate and long term value to our customers, employees and shareholders.” Marvell would pay $40 per share in cash and 2.1757 Marvell common shares for each Cavium share.
However, while Wall Street was largely pleased with the proposals, analysts pointed out that the projected growth rate for the combined company – an average of 6% to 8% a year – would not be much higher than was forecast for Marvell alone, or indeed for the overall chip business. However, the firms’ executives say they would expect to make savings of up to $170m in the first 18 months, even though there are no plans to axe products, and the increased scale and reduced competition the merger would drive could help Cavium achieve profitability. The firm has suffered annual net losses for the past five years and is expected to post a sixth for fiscal 2017, though its growth rate has been healthy. Revenues rose from $235m in 2012 to $603m last year when it lost $146m, and in the first three quarters of this year, its revenue has risen by 40%, partly because it acquired QLogic in 2016.
By contrast, despite the scandals and a major reorganization, Marvell reported revenues of $2.3bn last year with $21m profit. As well as appointing Murphy as CEO, it also hired a new executive team including Neil Kim, former head of central engineering at Broadcom, to be CTO.
The question marks over the Thunder X, and the broader strategy for the Cloud-RAN space, are likely to remain until the merger is complete (assuming it gets all the necessary approvals, expected by mid-2018). The ARM-based server/cloud processor opportunity is one that analysts, vendors and customers disagree about. It has been a tough ride to get to the point where customers can look forward to a choice of robust platforms from big-name vendors like Cavium and Qualcomm. The first pioneers, such as start-up Calxeda (formerly Smoothstone), were innovative but ahead of their time. Calxeda gambled on a market for 32-bit blade server processors which never really materialized, and once ARM moved to a 64-bit architecture with v8, it found itself left behind by the larger firms.
A classic case of bad timing, and the fate of many chip start-ups, but Calxeda will certainly have left its legacy. As it said in its own farewell statement: “The concept of a fabric of ARM-based servers challenging the industry giants was not on anyone’s radar screen when we started this journey. Carrying the load of industry pioneer has exceeded our ability to continue to operate as we had envisioned.”
Companies like Cavium and Qualcomm picked up the baton once the 64-bit architecture was available, but big server buyers remain divided on the merits of ARM-based SoCs. However, another milestone has been the emergence of 10-nanometer cores, which are helping ARM designs to compete with Intel Xeon on performance, not just relying on power efficiency to score points (a 10nm Xeon is not expected until next year).
It looks as though 2018 will be the decisive year, because demand from telcos for virtualized infrastructure will achieve significant scale in that year, after a slow 2017; and because there will be several new and powerful entrants. As well as 10nm Xeon and Centriq, Cavium is expected to ship its 14nm Thunder X2 with 54 cores running at 2.6 GHz, making it a real workhorse for applications like Cloud-RAN. AMD is developing processors based on its new Zen architecture, and will release a 48-core design targeted at data centers in 2018. And the first non-IBM OpenPower server processors are expected to emerge from China then, opening up the choice of suppliers and the ecosystem around that architecture.
Qualcomm shareholders want another $10 from Broadcom:
Reports in Bloomberg News indicate that Qualcomm shareholders want an additional $10 in order to look more favourably on Broadcom’s $103bn takeover bid, bringing the per-share price to around $80.
Meanwhile, Broadcom has hinted that it would make big changes to Qualcomm’s patent licensing business, presumably to try to avoid coming under the same attack from customers and regulators as Qualcomm has been suffering in recent years.
Patent licensing accounted for about $5.1bn in pre-tax profits for Qualcomm in its fiscal year 2017, compared with $2.7bn in pre-tax profits for its chip business. If Broadcom makes changes that would reduce the licensing business’s profits or cashflow, it would be likely to look for cost cuts elsewhere, and many observers fear the key target would be the San Diego firm’s legendary R&D budget and engineering teams. That, of course, would generate fewer patents and would start to make Qualcomm look far more like an average chip company in its financial structure.
Broadcom would be particularly eager to end the dispute with Apple, which it also counts as a customer. A spokesperson told Reuters that regulators around the world would “welcome this deal as a solution to the double-dipping issue”, while Broadcom CEO Hock Tan said in an interview earlier this month that he believes Broadcom “can be very constructive in resolving these issues and resetting relationships. We would not have embarked on this offer if we were not confident that our common key customers would not embrace it.”
Possible options for the licensing business include selling it or spinning it off; moving to a flat per-device fee model rather than the controversial percentage of the device price; or charging more for chips, including the patents in the price.
Bernstein analyst Stacy Rasgon thinks the solution would be to do away with separate licensing fees and increase chip prices – even though many licensees are not chip clients. Rasgon told Reuters: “It’s a completely different model of how to run Qualcomm. [Broadcom’s CEO Tan] is essentially saying, ‘You guys are wrong — the licensing revenue is never coming back’.”