Photo credit: delfi de la Rua

What Google Cloud Vision API means for Deep Learning Startups

Google announced today the Cloud Vision API, which makes available to the general public their Deep Learning algorithms for image recognition, OCR, facial detection, etc.

Example of Google Cloud Vision API result (source: Google)

There are already many Deep Learning startups offering these services, in my opinion Google could easily capture most of the market.

There are 3 reasons behind the Deep Learning renaissance, each of them a potential competitive advantage, let’s see them:

  • Algorithms. New discoveries are published on a daily basis on arxiv. Long gone are the days in which Google published their GFS/Map-Reduce paper years after they implemented it, leaving the rest of the world to build Hadoop once they were already moving to another technology. I don’t see a sustainable competitive advantage on some algorithms that others will not be able to replicate in few months.
  • Computing power. Yes, the cloud has democratized the access to computing power. But Deep Learning requires GPUs, and AWS is the only one providing GPUs, and they are old and do not perform very well. When Google open-sourced Tensorflow, it was soon mocked for being 10x slower than other open source libraries already available. One of the explanations by some analysts was that Google computing power is so massive that they can use their engineering resources to build new things instead of focusing on performance.
  • Data. Few companies can compete with Google on available data, and for sure they are not startups. Maybe an open common effort could be done to build large datasets, but it will not bring a competitive advantage to any single company.

How could Deep Learning startups then compete with Google?

  • Price. They could price their API lower than Google. Suddenly one of the hot emerging technologies has become a commodity.
  • Look for an acquihire. Instead of building a sustainable business, they could try to be acquired for their talent. I doubt that building a public API is the best strategy for it, it would be wiser to work on hard unsolved problems like DeepMind did.
  • Focus on a Niche. Google API will shine on general problems, and by its nature will not probably go after specific markets like industrial applications and healthcare. For the same reason, a public SaaS API business model may be not effective for this markets, it might require embedded or on premise systems and a more standard B2B sales approach.

Google has often failed at monetizing their business services, startups may hope that it will happen again. I would not sleep well on this thought, the competitive advantage is dramatically on Google side and they will keep improving the algorithms for their internal services, regardless of API revenues.

Today Google entered the Vision market, the same could soon be true also for Natural Language Processing (NLP), the other field which is seeing lots of improvements with Deep Learning. For example, most of the available services are in English, probably limited by the fact that there are much more available datasets and corpuses in this language. Google is in the best position to provide NLP services in other languages too.

While it could be a bad day for many Deep Learning startuppers, once they get over it and refocus I think there will be a lot of very smart people working on new hard problems, making the world a better place in the process.

These are my 2 cents, what is you opinion on this topic?