Now Nvidia only has to buy Xilinx to dominate servers

Why not? They certainly can find the funds.

Amazon to buy Xilinx….. 2017 prediction

IBM is definitely positioning itself to be a continued producer of CPU IP, and the R&D towards that end, but IBM has gone more towards the ARM holdings licensed IP business model in an effort to give its Power/Power9 micro-architecture a larger economy of scale and a larger Linux based software ecosystem availability to assure that power9(Power8s as well) will be used by many third party licensees. This and IBM going fabless will allow IBM at some future time to maybe start using some alternative competitively bid out power9 suppliers as the power/Power9 ecosystem builds out into a larger market of Power9 makers with IBM the holder of the basic power9/older power IP.

The only reason for IBM to want Xilinx is for its FPGA IP to maybe have that insurance against both AMD’s and Nvidia’s professional GPU accelerator dominance as Intel has done with its acquisition of Altera. I would rather see AMD get some of its own FPGA IP or acquire some via acquisition. Also there is nothing stopping both AMD and Nvidia from getting some Power9 licensing going and branching out into that market with their GPU IP leveraged and used with the Power9 IP. Nvidia lacks an x86 license but Nvidia could very well produce some form of power9/Nvidia HPC/server class Interposer based SOC IP of its own to compete with AMD. AMD will be taking its Zen micro-architecture and making some very powerful HPC/Server and workstation APUs on an interposer foundation with much larger Vega GPU resources than have not to date been found on any consumer grade APUs of AMD’s current/past lines of APUs.

AMD will be perfecting the APU on an interposer for both the consumer and professional markets this time around with its Ryzen/Vega consumers variants and some form of Zen/larger Vega designs for the workstation/HPC/Server and exascale markets. AMD has patent filings for placing some FPGA in HBM2 memory compute on the HBM2 stacks to work form the HBM2’s DRAM memory dies for is government exascale R&D grant design proposal.

I do not see IBM ever getting its hands on Nvidia, too much ego to be overcome on the part of Nvidia’s CEO and most likely shareholders.

What’s the value of acquisition, over partnering, or building their own? NVidia’s specialty is SIMD, for example, but IBM is no stranger to designing and fabricating their own SIMD silicon. If IBM bought NVidia, how much time and money would it take to incorporate the useful ideas from NVidia into their own processes? Is this cheaper or faster than doing it in-house, and over what time scales? On what do you base this?

IBM doesn’t really make consumer PC hardware any more, so would you have them branch out into PC graphics cards to maintain NVidia’s current lineup, or would you have them abandon that market and hand it over to AMD and Intel, or would you have them run NVidia independently (in which case, again, what exactly do they gain from an acquisition over a partnership)?

I see 2000+ words here on why company X and company Y have great and complementary technology, but nothing that indicates why this would be a good idea for X or Y, or even how X and Y were selected. I can name hundreds of companies with great and complementary technology — just pry the top off your PC and start reading off names of components — but that doesn’t mean any of them would be good corporate acquisitions.