Competetive Landscape & new startup Groq

http://www.cnbc.com/2017/04/20/ex-g...e-ai-unit-to-form-groq-with-palihapitiya.html

Looks like some of the TPU devs broke off with Google to kick off a new AI chip startup.
Not confirmed whether they'll be working on a DL-accelerator ala TPU, or neuromorphic ala BRN, Knowm etc. Likely the former as that is their background.

Existing TPU design shown to train inference 15x faster than GPU so any further improvement upon that could be  very exciting for DL devs if and when the GPU requirement is fully eliminated. The holy grail for them would be a TPU equivalent that can both train and execute a DNN model on-die as a simple embedded computer can handle the TensorFlow framework execution and Inputs/Outputs (vision, audio, text etc).
https://arxiv.org/pdf/1704.04760.pdf

Commercial competitive landscape (let me know if any missed)
Neuromorphic ASICs / FPGAs

• Brainchip
• Intel (Numenta)
• IBM (TrueNorth)
• Knowm

DL Accelerator ASICs / FPGAs

• NVIDIA (Drive)
• Google (TPU)
• Groq
• ARM (DynamIQ)
• Qualcomm (Zeroth)
• Movidius
• Teradeep

It's anyones game as commercial agreements tend towards fit to use case and requirements, rather than whichever has the most sophisticated tech. Excited to see what business BRN can carve out in the next few years in both a niche and general sense.