Artificial Intelligence (AI) as a Service and “Core AI”

May 8. 2017. 7 mins read

We find ourselves writing about artificial intelligence (AI) a lot these days because AI is now starting to resemble the dot-com craze that some of us are old enough to remember. As retail investors, we know there aren’t many ways to invest in AI at the moment. For venture capital firms though, it’s a veritable mine field gold mine of anywhere from 1,500-2,000 AI startups actively trying to create Mark Cuban’s AI trillionaire.

While it’s nearly impossible to keep track of what everyone is getting up to, what we can do is follow the money. Where are the world’s best venture capital firms placing their bets? Which companies do the experts say we should watch? When it comes to expertise and just good old horse sense, look no further than the bright minds at CB Insights who came out with a list of the top-100 AI startups to keep an eye on as seen below:

While we’ve covered many startups across all of the above categories, what we’re interested in talking about today is “AI-as-a-service” (AIaaS) or the startups above that fall under the category of “Core AI”. Before we go any further, here’s a definition of the terminology we’re using:

  • An AI platform you developed internally = Core AI
  • Application Programming Interface = API
  • Core AI + API = AI-as-a-Service (AIaaS)

When a technology like artificial intelligence offers a distinct advantage in business, you will expect a whole bandwagon of CEOs and entrepreneurs making a beeline to get access or at least get to know more about it. A novel technology never comes cheap and this means business enterprises will naturally find ways to get the benefits of AI without the cost of an army of data scientists or AI experts. The solution will most likely be to use AIaaS.

Artificial Intelligence (AI) as a Service

In looking at the above list, we see 15 startups offering “core AI” or AIaaS. Here they are:

Startup Funding (millions) Description
Sentient $135.78 Inspired by biological evolution, “Evolutionary Intelligence” creates trillions of “genes” that are unique solutions to a problem.
Ayasdi $106.35 Ayasdi is targeting two niche applications for their “topological big data analysis” technology; healthcare and financial services.
Voyager Labs $100 Creates tools that are out there scouring unstructured data and deriving meaning from it.
Digital Reasoning $73.96 Analyzes relationships in human communications with applications in fraud and human trafficking
Vicarious $72 Stealth mode. Backers include Elon Musk, Peter Thiel, Dustin Moskovitz, Mark Zuckerberg, Vinod Khosla, and Jeff Bezos,
Affectiva $33.72 Affective computing technology can analyze your photographs and identify any of the 10,000 possible facial expressions
H2O.ai $33.6 World’s leading deep learning platform used by 80,000 data scientists in 9,000 organizations. Used by 20% of Fortune 500 companies.
Cognitive Scale $25 Platform uses IBM Watson and targets healthcare, commerce, and financial services. Build your own cognitive app within 10 hours.
Petuum $15 Created an AI operating system that is much faster than others out like Google’s TensorFlow. Expected in summer of 2017.
Bonsai $13.62 Created own programming language that encourages “AI for everyone”. Platform automates the management of algorithms and machine learning libraries.
Scaled Inference $8 Operating in stealth mode. Ex-Googlers solving problems like pattern recognition and anomaly detection
Skymind $3.32 Tencent lead investor. Deep learning library for Java that wants to be the “Red Hat of deep learning”.
Algorithmia $2.4 An algorithm marketplace. Crowdsourced AI algorithms that developers share for a nominal fee and also requests for algorithms with “bounties”.
Loop AI Undisclosed 100% unsupervised AI algorithms turn unstructured data to structured data and can do so in any language or even mixed languages.
Numenta Undisclosed Silicon valley startup. Framework inspired by neuroscience called “Hierarchical Temporal Memory” or HTM.

Regular readers may recall that we highlighted “A List of 15 Free AI Software Programs to Download” which not only helps explain why there are 1,500-2,000 AI startups now but also shows that you don’t need to pay for expensive AI software frameworks. What you do have to pay for is access to the nice clean delicious datasets which you feed your AI algorithms. When using an AIaaS provider, you don’t have to worry about all that. Just plug into their API and you can easily integrate the power of AI into your applications. Here are the articles we’ve written about the above companies to-date:

With nine companies already covered, let’s take a look at the remaining six startups.

Digital Reasoning

Click for company websiteFounded in 2000, Tennessee startup Digital Reasoning has taken in $73.96 million from 6 rounds of funding that included participation from NASDAQ, Goldman Sachs and Credit Suisse. Their flagship product Synthesys® (a Java-based software platform) uses Natural Language Processing (NLP) and Machine Learning to analyze entities and relationships in context within Big Data to understand human communication. They are presently scouring online escort ads with Thorn to identify human traffickers. Here’s something they did a few years ago around “word analogy accuracy” which is a big thing for NLP:

Neural networks generate a vector of numbers for each word in a vocabulary and do “word math”. For instance “king” minus “man” plus “woman” would yield a result of “queen”. There is an industry standard dataset of around 20,000 word analogies. Previously, Google was able to get 76.2% of the word analogies “correct” in their system. Stanford’s best score was a 75.0% accuracy. Digital Reasoning’s model achieves a score of 85.8% accuracy, which is a near 40% reduction in error over both Google and Stanford.

In 2016, Digital Reasoning acquired a Nashville company in 2016 called Shareable Ink that transforms point-of-care documentation into structured data for analysis, positioning Digital Reasoning strategically in healthcare. Last month they brought on a CEO who has a track record of taking a startup public (Axent Technologies which was bought by Semantic for almost $1 billion). 


Click for company websiteFounded in 2016, Pittsburgh startup Petuum is a Carnegie Mellon spinout that has taken in $15 million from a single round of funding that includes Chinese Internet giant Tencent as an investor. They provide organizations with a platform to create fast and scalable AI/ML solutions using drag-and-drop while consuming minimal computing resources. Petuum claims their technology can now solve certain AI problems about 200 times faster than Apache Spark, and about 20 times faster than Google TensorFlow. Here’s the performance of their PetuumOS which will be released this summer:

250 million Matrix Factorization model parameters, 1.3GB input data, running on 32 CPU machines in a datacenter.
250 million Matrix Factorization model parameters, 1.3GB input data, running on 32 CPU machines in a datacenter.

If you don’t speak nerd, the long orange line is bad and the small blue line is good. Here’s another example where the blue line (Petuum) is much higher than the orange line (Google) which means Petuum is faster:

ResNet-152 model (ImageNet 2015 challenge winner) with 60.2 million parameters, running on 32 GPU machines in a datacenter.

They’re also looking at healthcare applications. For example, when detecting tuberculosis from chest x-rays, PetuumMed achieves 91.2% +/- 5.0% sensitivity (recall) versus 82.0% for physicians, and achieves 78.0% +/- 7.0% specificity versus 77.0% for physicians. While there is a shortage of radiologists presently, majoring in radiology is probably not the best idea.

Bonsai AI Inc.

Click for company websiteFounded 2014, Bonsai is a Berkeley startup that has taken in $13.62 million in 4 rounds of funding the latest of which was a $7.6 million round that just closed in May 2017. Investors included Microsoft, Samsung, ABB Technology Ventures, and Siemens AG. This startup offers a platform automating the management of algorithms and machine learning libraries with an “AI for everyone” concept. They’ve created their own programming language called Inkling. There are quite a few interesting videos out there showing actual Bonsai code do things like controlling stability in Lunar Lander or teaching a system how to play Breakout in 40 lines of code

Scaled Inference

Founded in 2014, this Palo Alto startup has taken in $8 million from 2 rounds of funding. The Company operates in stealth mode and offers artificial intelligence as a cloud service. The founders who used to work for Google envision the startup to be a neutral platform for different kinds of problems for AI to solve, problems like pattern recognition and anomaly detection. Investors include Khosla Ventures and Tencent, again. They’re operating in stealth mode so that’s about all we can tell you about them. Oh, and their CEO is wicked good at tennis and their lead scientist once aspired to be a professional hip-hop dancer. 


Click for company websiteFounded in 2013, Seattle-based, ALGORITHMIA has taken in $2.4 million in a single round of funding. The business model is truly unique, and is centered around an “algorithm marketplace” where programmers create algorithms and share them in an opensource fashion. The difference is that the programmer can charge per API call and there is also a charge per-second for computing time. The whole thing is very affordable, and there are algorithms for social sentiment analysis, face detection, smart thumbnail, nudity detection and thousands more. They also have a concept called “bounties” where you can pose a problem and people can solve it for a “bounty”. For example, if you can create an online real-time duck detector then you can walk away with a cool $150. They have over 30,000 developers using the platform that have created +3,000 microservices

Loop AI Labs Cognitive Computing

Click for company websiteFounded in 2012, San Francisco startup Loop AI Labs Cognitive Computing is the first company to offer a commercial unsupervised general-purpose cognitive platform that learns and reasons in any language. It lets any enterprise quickly roll out intelligent Cognitive Robotic Process Automation (RPA) workflows and utilize working cognitive Q Robots that learn from human peers. The system excels at turning unstructured data into structured data. Here’s a cool picture to look at:

They can do cool stuff like understand sentences that contain both English and Tagalog (you Filipinos out there know what we’re talking about here) and they are already using their solution to create value for Fortune 2000 clients in Asia, Europe, and America. We’re not sure if it can speak Engrish yet.


If you have an API, a website, and a label that says “powered by AI”, then technically you would be an AIaaS. We can’t possibly include every AIaaS out there in this article, so please bear with us before you send that email asking why we didn’t include your startup in this article. Do drop us a note though, and we’ll be sure to check out your AIaaS and possibly highlight it in a future article.


Leave a Reply

Your email address will not be published.