When I first read it this morning, I almost laugh! Not because its a friking amount of money for a company that has the monopoly of the world fastest Interconnects, but simply because Microsoft and Intel!!!... failed to outbid NVIDIA.

Click on the picture to head to NVIDIA source.

Perspectives

From my own knowledge, I don't even understand why Microsoft is trying to go for it... in my view it was just "pay for the buck" strategy in case no one had the guts to go for it seriously, or on the other side of the coin, to create pressure to other parties in hope to drive market share.

But from Intel, the history is very different. OmniPath is really not a competitor when compared with Infiniband, and most likely Intel was trying to save the day with biding for Mellanox. Of course that would mean a huge new market share for Intel, since CPU wise, they have basically the entire market, but when it calls to build HPC (High Performance Computing) clusters, a third of the slice usually goes to Mellanox. Because OmniPath can't really do it's job well done.

I mean, I loved the idea of OmniPath and the tech was really an amazing breakthrough, but due "the usual" market pressure, it didn't had the resources to come in time to make Intel call for it as a profitable product. They might in fact be "currently" in a very tricky situation, especially when AMD is catching up quite nice lately.

This also means, that other HPC vendors, will now need to consider a bigger partner... which might in fact benefit HPC customers (like myself). The idea of bundling super big deals is always beneficial to both vendor and customer.

Maybe we will see also something new emerging from this new strategy from NVIDIA. I think that it empowers NVIDIA on developing the new Exascale trend, which looks like targeting memory and interconnects to be the center of importance of the Exascle Supercomputers (and not the CPUs anymore). In addition, we might see Switches becoming more powerful and/or maybe being able to execute computational workloads (they can already do some offloading, but very little and selective MPI stuff).

Eager to see how this goes next year!

Disclaimer: All I wrote here is based on my own opinions. No information is NDA and I am not using any specific words to target or hurt any parties. All of the above are pure thoughts of my own, interpretations, or amusements and do not necessarily constitute the reality.