Xilinx is Killing Altera!

At a recent outing with FPGA friends from days gone by, the long running Xilinx vs Altera debate has come to an end. The bottom line is that Xilinx has used the FUD (fear, uncertainty, and doubt) of the Intel acquisition quite effectively against Altera and is racking up 20nm and 16nm design wins at an alarming rate. It will be a while before they all hit HVM (high volume manufacturing) but it is coming, absolutely.

The question I have now is: Will Xilinx be acquired next?

The Intel acquisition of Altera became official in December of 2015 at a cost of $16.7B which was more than a 50% premium based on the stock price when the deal was announced. Altera is the second largest maker of programmable chips (FPGAs), Xilinx (XLNX) being the first.

Altera and Xilinx are fierce competitors in a two horse race for a $5B+ market with an expected CAGR of 8.4% from 2016 to 2020. It is widely recognized in the FPGA world that the vendor who is first to a new process node wins the commanding market share. This was proven when Altera overwhelmingly beat Xilinx to the 40nm process node. Altera was partnered with TSMC (TSM) and Xilinx was partnered with UMC (UMC), UMC stumbled at 40nm and took Xilinx down with them. Prior to 40nm Xilinx was the node leader.

After the 40nm debacle at UMC, Xilinx moved to TSMC (joining Altera) making it a level process node playing field. Xilinx then beat Altera to the 28nm process node by a matter of months and again beat Altera to the 20nm node by more than a year. Xilinx of course was awarded majority market share as a result.

Altera then switched their manufacturing to the new Intel Custom Foundry division for 14nm to better compete with Xilinx. This was a huge PR event as Altera and TSMC were intimate partners since the beginning of the fabless semiconductor transformation in the 1980s. This also brought a direct competition spotlight on Intel versus TSMC for the first time.

Unfortunately Altera stumbled at 14nm and is hoping for high volume manufacturing (HVM) before the end of 2016 while Xilinx started 16nm HVM in 2015. Xilinx also has access to the low cost TSMC 16FFC process to which Intel has no answer so look for additional market share to be won by Xilinx moving forward.

The next process node is 10nm which Intel is currently delaying while TSMC will start 10nm HVM in the first half of 2017 with a quick step to 7nm HVM one year later. Given that quick step, Xilinx has chosen to skip 10nm and move directly to 7nm in 2018. As a result of the challenges Altera experienced with the Intel process bureaucracy at 14nm, I do not expect to see Altera in HVM at Intel 10nm until mid to late 2018, giving Xilinx another solid process advantage.

Will Xilinx be Acquired?

After Intel acquired Altera the rumors started about Xilinx being next. Really it was more about people, like myself, thinking out loud but I would not rule out due diligence and executive level discussions. My take is that Xilinx felt that the Intel/Altera acquisition would falter and favor Xilinx in the long run (which it already has). Xilinx is now in a much stronger position for acquisition discussions with companies that can leverage the dominant position they have in the high margin merchant FPGA market and the desire to compete with Intel in the data center business.

The first name that comes up in the media is Qualcomm but my bet is on Avago/Broadcom CEO Hock Tan. Hock has proven that he can make big deals happen (LSI Logic for $6.6B and Broadcom for a record $37B) and the data center business is in his sights, absolutely.