ArchiTECHt Daily: Google's capex creeps back up

I vowed I wouldn't write about AI in this space again today, but I feel compelled to point out there
ArchiTECHt Daily: Google's capex creeps back up
By ARCHITECHT • Issue #9
I vowed I wouldn’t write about AI in this space again today, but I feel compelled to point out there was a lot of cool AI news and research on Wednesday. Be sure to check it out below. Also, this (only somewhat unrelated) take on the unique challenges of innovating in consumer hardware is worth reading.
Instead, I’ll highlight Google’s data center spending in 2016, which, at $10.2 billion, puts it back on the uptick after the rare down year in 2015 (when it spend only $9.9 billion after dropping $10.9 billion in 2014). Presumably, the increased investment is tied, at least in part, to Google’s burning desire to grow its upstart cloud business that only began in earnest in 2013. All estimates put Google’s cloud market share far, far below that of Amazon Web Services, and also significantly below Microsoft Azure. 
Growing its piece of the cloud pie will require Google to not just add capacity and expand globally, but also to add the right capacity. While Google has optimized its overall infrastructure to run its core search-plus business as efficiently as possible, serving cloud customers’ wide-ranging workloads probably requires some different gear. On Wednesday, for example, Google announced the availability of SQL Server instances on Compute Engine—not novel workloads by any stretch, but maybe relatively novel to Google.
Essentially, large platform providers like Google, Facebook and Amazon sometimes need to build for the workloads they want or expect, rather than just the workloads they already have. For example, when I read on Wednesday that Mark Zuckerberg views video as a “megatrend,” I was reminded of my recent trip to the company’s Prinveville, Oregon, data center, where the company is investing in 100-gigabit networking gear—and looking toward 400-gigabit—in part to handle all the video traffic it expects to serve.
If Google wants to compete with AWS in the cloud, it’s going to need the infrastructure to handle all those enterprise workloads it wants to run.

A Georgia Tech system called DeepNav can navigate cities using Google Street View images.
A Georgia Tech system called DeepNav can navigate cities using Google Street View images.
Around the web: Artificial intelligence
In this paper, the company describes its CommAI framework for building algorithms that can handle, and learn, a wide range of capabilities. Researchers have been making progress in this direction, but the authors suggest we’re still quite a long way from achieving generally intelligent machines.  •  Share
This could be a good application of Watson’s NLP skill set, especially if IBM and H&R Block nail the user experience. For real tax questions, though, nothing beats a human who can translate tax code into English.
The Financial Times examines the hype around consumer AI, particularly Amazon Alexa and seemingly ubiquitous chatbots, and whether it can live up to expectations.  •  Share
 A review of a several popular deep learning and machine learning frameworks, including Microsoft CNTK and Amazon-favorite MXNet. TensorFlow comes out on top.
For only $1,000 you can build yourself a computer for doing all sorts of fun deep learning experiments. Apparently, though, you should study up on GPUs first and not skimp on that component.
To be clear, this list is neither all AI companies nor, from what I can tell, all European. But … the blog post has some good info and the spreadsheet certainly could help ID some interesting new companies.  •  Share
Question: Is there any disease deep learning can’t learn to diagnose with enough images? Answer: Probably not. The potential for AI-informed health care in remote areas is really promising.
The headline links to a paper from Carnegie Mellon on the state of the art for controlling robots. This link points to a paper from Georgia Tech explaining a system for navigating cities using Street View images. This one is about a method for generating synthetic sensor data  to protect privacy.  •  Share
Marketing material from Microsoft, but interesting use cases including Carnival and Mars.
Professor Winfried Hensinger and Dr. Bjorn Lekitsch from the University of Sussex
Professor Winfried Hensinger and Dr. Bjorn Lekitsch from the University of Sussex
Around the web: Quantum computing
Remember when computers were as big as rooms? This one is as big as a football field and would cost $126 million. It also could be revolutionary.
 Oak Ridge National Laboratory researchers have set a new record in quantum communication, which involves sending lots of information via qubits. In this case, over fiberoptic cable.
Want to sponsor the ArchiTECHt newsletter or podcast? Email
Around the web: Cloud and infrastructure
RISELab, the successor to the very successful AMPLab project at UC-Berkeley (it gave the world Spark, Mesos and Tachyon), officially launched late last month. Here’s more info on what it’s building.
Seriously. Researchers—and, no doubt, large networking and cloud providers—are hard at work figuring out the ideal models for moving computation to the edge.  •  Share
I wouldn’t bet the farm on its forthcoming machine learning GPU. Nvidia owns that space.  •  Share
Trying to force all CNCF projects to adopt the ASL could be a risky bet IMHO. I suspect smart companies will get a lot more creative about how they do open source.  •  Share
It’s not the most forward-looking solution, but there’s also still a big market for this type of thing. 
Did you enjoy this issue?
The most interesting news, analysis, blog posts and research in cloud computing, artificial intelligence and software engineering. Delivered daily to your inbox. Curated by Derrick Harris. Check out the Architecht site at
Carefully curated by ARCHITECHT with Revue. If you were forwarded this newsletter and you like it, you can subscribe here. If you don't want these updates anymore, please unsubscribe here.