Here is another article on Caldxeda bowing out
---------------------
Applied Micro, Cavium to Benefit as Calxeda Bows out of Intel Challenge
By Tiernan Ray
Calxeda, a privately held chip maker that has been for a couple of years now one of several hopefuls in the battle to supplant Intel (INTC) in the server market, is apparently going out of business, according to a piece from AllThingsD‘s Arik Hesseldahl, who last night wrote that one source tells him “they just ran out of money” after having bagged $90 million in venture funding.
That news has left the Street today to contemplate the fate of other companies trying to use the ARM Holdings (ARMH) chip designs to build microprocessors that will power an emerging class of “micro-servers.”
Shares of chip maker Applied Micro Circuits (AMCC), which is one of the hopefuls in that battle, are down 20 cents, or 1.6%, at $12.40, and Cavium (CAVM), another one, is down 62 cents, or 1.7%, at $35.37. In addition to those two, many other ARM-based chip designers are considered in the running, including Nvidia (NVDA) and Qualcomm (QCOM).
Shares of Advanced Micro Devices (AMD), already a major player in servers, and long an Intel rival, are up 4 cents, or almost 1%, at $3.69. Intel shares are up 13 cents, or half a percent, at $25.27.
FBR & CO.’s Christopher Rolland today pounds the table for shares of Applied Micro, reiterating an Outperform rating and a $16 price target, arguing the company’s own 64-bit ARM-based sever chip, “X-Gene,” is underappreciated by investors.
To Rolland’s thinking, Calxeda just had the wrong approach, and its demise is no reflection on the general ARM-based server market, which still is viable:
We believe this shuttering will have limited near term impact for the ARM server opportunity. We offer the following thoughts; firstly, Calxeda was reportedly looking for another round of funding (perhaps for well over a month now). Secondly, the company‘s current EnergyCore processors were 32-bit with two generations before they released a 64-bit processor (and $5 to $10 million a mask set for each generation). Finally, the company ‘s ARM based processor strategy was based on the FAWN (Fast Array of Wimpy Nodes) approach, an outdated but popular theory about five years ago for low-power microserver processors. Indeed, hyperscale applications require 1) 64-bit and 2) a minimum threshold of processing performance unlikely to be reached by Calxeda’s approach.
X-Gene, he thinks, will surprise some people with how good it is:
Since ARM TechCon over a month ago, we have held numerous follow-up conversations with software and hardware engineers who have had hands-on experience with X-Gene. Some of these engineers work for open source application developers that have ported their apps to run on X-Gene, while others were associated with server OEMs. Universally, these engineers have reached the same conclusion: when specifications are finally released, X-Gene’s performance should “surprise to the upside.” That said, there is currently no time table for X-Gene servers to be made publicly available for benchmarking.
Evercore Partners‘s Patrick Wang also is upbeat on Applied Micro today, writing that “Intel’s DCG [Data Center] business remains structurally challenged, and expect competition in LP server to intensify,” and that both Applied and ARM are “key beneficiaries” of Intel’s challenge.
Like Rolland, Wang thinks Calxeda’s departure is not a defeat for the overall ARM server movement, and he expects someone will buy Calxeda’s assets:
Mgmt confirmed plans to shut down as they literally ran out of money, likely unable to close its D round of financing. The company was founded under the Smooth Stone name in 2008 and raised a total $103mn of capital [...] We believe the LP server space will eventually be dominated by 3 or 4 major players by 2016. Other than Intel, other suitors include Samsung, Qualcomm, Broadcom, and Applied Micro [...] Along with Marvell, we believe Calxeda was actually ahead of the curve and perhaps too early to market as it didn’t exist. Further, they based their SOC roadmap on standard 32b ARM cores which is limiting their opportunities as the market runs on 64b code. They were planning to intersect the 64b market with the Lago SOC in 2H14 or perhaps later [...] Despite these challenges, we would be surprised if the Fleet Fabric IP isn’t quickly acquired by an opportunistic OEM partner, Web 2.0 customer, or competitor. Conceptually based on the industry standard Ethernet switch, the fabric roadmap was Calxeda’s strength, in our view [...] The fabric provides the basis for how the server chips talk to each other. Notably, AMD acquired SeaMicro ($334mn) just for their Freedom Fabric [...] While bears will use this as a talking point, there is absolutely no change to our view of the LP server opportunity. We expect the market to intensify next year led with 64b chips from Applied Micro followed by others (i.e., AMD, Cavium, Samsung, etc).
And William Blair’s Anil Doradla reiterates an Outperform rating on Cavium stock today, and he, too, thinks the Calxeda was a poor approach, the opportunity is still real for other chip makers, and that Cavium has the goods:
While some might believe Calxeda’s failure is a reflection of poor end-market demand for ARM-based servers, we strongly believe that it is a reflection of company-specific challenges rather than weak end-market demand. We believe greater emphasis on 32-bit ARM servers, lack of experience with large customers (both enterprise and service providers), and focus on lower core counts were some of the key reasons for Calxeda’s failure. We believe companies that have a track record with large enterprise customers and have expertise in higher core count processors are best positioned to succeed. The bottom line is that Cavium has both these attributes [...] We believe multicore processor vendors focused on providing higher core counts (higher than four cores) are best positioned to succeed in the market. We believe the ability to differentiate on price, size, and power from incumbent X86 architectures (Intel) is more pronounced at higher core counts. While there may be a 10% to 15% advantage in four core (or fewer) processors, we believe there is about 50% advantage (power or cost) at higher core counts (greater than eight cores). Within the multicore processor industry, Cavium and Broadcom (BRCM), through NetLogic, are vendors focused on the higher core count processors (other than Intel).
AI
