ARM's Race: Who Will Buy Applied Micro's Compute Unit?
Macom, having hardly drawn breath since announcing it is buying Applied Micro Circuits, has already drawn up a list of potential buyers for Applied Micro's Compute business, which Macom intends to sell. (See Macom Buys Applied Micro for $770M.)
Macom isn't sharing the list (the spoilsports), but who might be on it?
There are any number of companies that might be interested in Applied Micro Circuits Corp. (Nasdaq: AMCC)'s X-Gene ARM-based processors designed for data center applications. Intel looks impregnable in the market, but there are vendors that believe ARM-based products could thrive in niche applications, and once ensconced there, they could then start nibbling away at Intel's lead from those toeholds.
Hewlett Packard Enterprise used X-Gene devices in its HPE StoreVirtual 3200 virtualized infrastructure solution introduced in 2015, but that's one of the few notable design wins thus far.
Applied Micro has high hopes for its forthcoming third generation of X-Gene products, however. The company says X-Gene 3 will provide six times the performance of X-Gene 2. Two weeks ago, the company said it sent its designs for the X-Gene 3 processor to Taiwan Semiconductor Manufacturing Co. (TSMC) (NYSE: TSM) to manufacture. If all goes very well, the X-Gene 3 might ramp into production as early as the end of 2017. Early 2018 might be more likely.
Intel won't be standing pat with the current versions of its Xeon processors, which the X-Gene 3s will compete against. Still, even a sliver of the Xeon business could be a profitable adjunct for someone selling other products into data centers.
The group of companies that provide optical interconnect, adapters and NICs into data centers includes several companies that might be in a position to buy, including Finisar Corp. (Nasdaq: FNSR), Inphi Corp. , and Lumentum Holdings Inc. , but of that group, Mellanox Technologies Ltd. (Nasdaq: MLNX) might be the most obvious candidate for Macom's list.
Mellanox believes there is an architectural issue developing in data centers, that CPUs -- no matter whose -- are becoming bottlenecks that simply cannot be cleared with more CPU resources. Mellanox is advocating distributing intelligence throughout the data center, including in interconnect. To that end, it bought EZchip last year, and introduced its initial smart interconnect products earlier this year. (See Mellanox Eases Into Network Processor Market and Mellanox Completes EZchip Acquisition.)
But even if you can't solve the CPU bottleneck problem with another processor, having a CPU to optimize to work with intelligent interconnect might be useful.
Heavy Reading Analyst-at-Large Simon Stanley notes that Mellanox already has a good position in data centers with adapter cards and switches, and has just introduced an ARM-based processor for adapters.
"The opportunity to attack the data center processor market could be very attractive for them and they have been working with Applied Micro to counter Intel," Stanley said in an email exchange.
He also believes X-Gene would be a good fit for Marvell, "but I am also not sure they would want the risk at the moment."
Which might be a bit of British understatement. Marvell Technology Group Ltd. (Nasdaq: MRVL) is in the process of selling non-core operations, and new CEO Matt Murphy told analysts late last week that he is shutting down any research into any product line lacking a clear path to making fat margins. The X-Gene 3 is going to have to compete with any number of other companies who want to take on Intel in the data center market, some with ARM-based processors, others with different processing technologies; IBM is promising its Power9 in 2017, and Rackspace and Google appear interested. That doesn't sound like a clear path to fat margins. (See For Marvell, TCB Might = M&A)
Advanced Micro Devices Inc. (NYSE: AMD), which, in the semiconductor world, is sort of Liam Hemsworth to Intel's Chris Hemsworth, might also be interested. AMD already traffics in ARM based products, but might be interested picking up a custom ARMv8 core from Applied Micro.
"The AMD Opteron A1100 has standard ARM Cortex-A57 cores so is rather slow compared with the Applied Micro X-Gene devices. AMD has backed off on ARM due to lack of design resources so they would benefit," Stanley said, "but may not want to take the risk at this time."
Broadcom Corp. (Nasdaq: BRCM), meanwhile, started developing an ARM-based processor, called Vulcan, but that program failed to bear fruit on schedule, and the company abandoned the program. Arguing in favor of a Broadcom acquisition, Stanley said, buying Applied Micro's Compute operation might be an easy way to restart that project with an established product family with a new version on the way. After all, he noted, Broadcom is missing a new high performance processor beyond the XLP II family. On the downside, Stanley explained, Broadcom might consider the X-Gene 3 too data-center-focused.
"Qualcomm is a possibility if the Applied Micro X-Gene 3 is significantly better [in terms of] performance than the Qualcomm ARMv8 24-core multicore processor. This was expected to start shipping in 2016 but Qualcomm has gone quiet on this project. Qualcomm is already acquiring NXP so the assumption is that Qualcomm is planning to combine the NXP and Qualcomm ARM v8 developments. TI is still focused on DSP but has no high-performance cores to complement the DSPs."
— Brian Santo, Senior Editor, Components, T&M, Light Reading