Microsoft announced on Tuesday that it has acquired Cycle Computing
, a company that helps users run high-performance computing workloads in the cloud. I’ll skip over the newsy parts of this announcement and instead, for more details, point you to:
However, there are a few things about this acquisition that I think are worth calling out explicitly, some of which are glossed over in the coverage:
1. Cycle Computing began as an AWS-centric company
(you can see some of my old Gigaom coverage of it here
), and currently supports Azure and Google Cloud, as well. Microsoft has stated current customers can keep running on those platforms, but that future product work will be Azure-focused. That’s the obvious move here, although it would
be interesting to see a cloud provider buy a multi-cloud company, keep it that way, and turn a profit on usage of someone else’s platform.
2. HPC is a big market, somewhere between $20 billion and $36 billion in 2016
, depending on the analyst firm providing the estimate. According to Intersect360
, spending on cloud HPC resources was $783 million last year—although, not surprisingly, that’s also the fastest-growing segment of HPC spending according to the analyst firm. I’ll assume that’s correlated with the 19.2 percent drop in “departmental” HPC server sales in 2016—to $3.1 billion from $3.8 billion—according to Hyperion Research
Essentially, Cycle gives Microsoft a better opportunity to capture lots of workloads from teams that have high-performance workloads but can’t justify the expense of spending hundreds of thousands of dollars on HPC systems.
3. I think that deep learning workloads will become an increasingly big part of the revenue base for Cycle Computing inside Microsoft
—and perhaps are a big reason why Microsoft (which is now “AI-first”
) bought the company. Microsoft suggests this in its blog post, stating: “We’ve already seen explosive growth on Azure in the areas of artificial intelligence, the Internet of Things and deep learning. As customers continue to look for faster, more efficient ways to run their workloads, Cycle Computing’s depth and expertise around massively scalable applications make them a great fit to join our Microsoft team.”
Running these workloads at scale requires lots of computing and fast networks, which Azure has, but Cycle simplifies the on-boarding and management process. It already works with GPU resources, and I assume future work will include supporting Azure FGPA resources
, as well. We’re not talking about hobbyist AI experiments here, but the types of large models that large enterprises and research facilities will probably need to start training if AI catches on like we all expect.
It looks like Andrew Ng is starting an AI investment fund
Speaking of AI and deep learning, it looks like one of Andrew Ng’s next ventures (I wrote yesterday
about his new deeplearning.ai education effort) could be a venture capital fund focused on AI. PE Hub highlighted
an SEC filing by
Ng for a $150 million fund, aptly named “AI Fund.”
Some of the details here are subject to change, but this makes sense. Ng has some pretty firm ideas about the future of AI, and has played a role in its rise both in the U.S. and China, so he’s in as good a position as anybody to identify AI startups that could actually make a difference and grow into large companies. It will be interesting to get the full picture on this once more details are available.
For about the 800th time, I’ll point to my February podcast with Ng
(back when he was with Baidu) for more insight into how he thinks about this space.