Andreessen Horowitz VC Chris Dixon recently wrote about The Decline of the Mobile Web and argued that this "favors the status quo over new innovations." This view was supported by Union Square Ventures VC Fred Wilson in The Mobile Downturn, where he also noted: "The downturn is always followed by something radical and new that starts on the fringe and becomes mainstream." Let me get back to just that.
I strongly disagree with Chris's claim that this favors the status quo over new innovations.
The Return of the Fat Client, This Time Mobile
The Web gave birth to the first widespread thin client — the Browser (I'm sorry VT100) — and Cloud services became a natural evolution of the Web, used from both browsers and mobile apps. Many tasks we used the first widespread thick client — the PC — to do in the past have been replaced by Cloud Services (for instance, this post was originally written using WordPress's cloud service).
On mobiles, tablets, and wearables, cloud services (REST APIs) have been used much as on the desktop, except inside native apps — dramatically reducing time spent on the mobile web as Chris points out (from 20% of time spent to 14%, a 30% reduction in just one year). HTML5 has been mentioned as the technology that will bring back the mobile web, yet the trend toward "natification" of mobile apps doesn't support that.
The Rise of Bring Your Own Supercomputer Mobile Device (BYOSMD)
As Marc Andreessen — inventor of the web browser — pointed out in this tweet, his mobile phone is faster than supercomputers of the past, and this is just the beginning of mobile supercomputing.
1. Mobile CPUs and GPUs
Marc's phone has a very fast 64-bit Apple A7 Cyclone CPU (see also ExtremeTech's A7 analysis), and the corresponding tablet — iPad Air — has a GPU that Bartosz Ciechanowski (@BCiechanowski) demonstrated in Exploring GPGPU on iOS can be up to 64× faster than the A7 CPU using OpenCL. As Bartosz put it: "That's right, the GPU was over 64 times faster. Sixty four. Times. Faster." (code on GitHub).
And the competition isn't standing still:
- Qualcomm announced the 64-bit Snapdragon 808 and 810 processors with Adreno 400-series GPUs
- Intel announced the 22nm 64-bit Atom CPU
- Samsung is working on 64-bit mobile CPUs
- Nvidia unveiled the Tegra K1 with 192 cores
- Underdog Adapteva created Parallella with the 16- or 64-core Epiphany Multicore Accelerator
Dear Chris Dixon: The effect of mobile super-CPUs on innovation is largely unexploited territory (perhaps with the exception of Infinity Blade 3).
2. Mobile On-Device Storage
Mobile on-device storage is currently neglected compared to cloud storage, yet it may be the most exciting enabler for innovation. SanDisk's MicroSD offers 1,000× more capacity in 2014 than in 2005; Lexar's SD cards reach 0.5 TB, Kingston's USB stick reaches 1 TB, and upcoming RRAM from Crossbar aims to store 1 TB on a 2 cm2 chip. Accessing local storage on a mobile device has roughly 1,000× lower latency than hitting a 4G network. To put 1 TB in perspective: that is more storage capacity than Google and AltaVista combined had in the late nineties.
And if one wants to extrapolate even further, researchers have shown how to store at least 700 terabytes in 1 gram of DNA.
Dear Chris Dixon: The effect of mobile storage on innovation is largely unexploited territory (though I must admit Memkite is working on it).
Conclusion
What Will This Mean for the Cloud?
Given the tremendous increase in processing power and low-latency storage capacity of mobile devices, a great deal of processing is likely to happen on-device instead of in the cloud (improving latency, privacy, and offline availability). At the same time, the volume of data prepared and synced from the cloud to mobile devices will grow substantially, increasing the need for cloud infrastructure — even as cloud services that require a REST call for every tiny user interaction fade.
What Will This Mean for Mobile?
It will be a catalyst for innovation. Having a mobile supercomputer opens exciting opportunities in search and machine learning — especially if connected to the brain in 10–20 years via EEG-based technologies or implants.
It feels right to round off with Pete Warden's (CTO of Jetpac) demonstration of How to add a brain to your smart phone, using Deep Learning on a mobile device to turn a phone into a sonar-like detector for — drumroll — a cat (Deep Belief SDK Demo on Vimeo).
Dear Chris Dixon: Mobile Eats the Cloud.
Related: Technical Feasibility of Building Hitchhiker's Guide to the Galaxy, i.e. Offline Web Search — Part I
Amund Tveit (@atveit) — Co-founder of Memkite