When we consider earlier technology waves, there is often one company (or possibly two) that emerges as a leader and becomes synonymous with the technology. Mainframes became synonymous with IBM, virtualization spawned VMware, open source gave us Red Hat, web search was reinvigorated by Google, and social networks gave us Facebook and Twitter. We would argue that this has yet to happen with AI and machine learning (ML).
The 451 Take
AI – and its variants, notably machine learning – is an enabling technology. It enables computers to do things that weren’t possible in earlier programming paradigms. With modern cloud computing, networking and storage, those things can be done at a scale unimaginable when the term ‘artificial intelligence’ was first coined in the mid-1950s. Of course, it doesn’t really matter whether AI spawns large software companies, but it demonstrates how intertwined it is with every other aspect of the technology industry – and even more so with companies that wouldn’t have previously perceived themselves as technology companies or as heavily dependent on technology.
Potential enterprise AI software giants
Let’s look at some of the contenders to be that massive AI company. AWS, Google Cloud Platform (GCP) and Microsoft Azure are large divisions of huge companies that provide AI services, applications and infrastructure. We doubt AI is anywhere close to being the largest contributor to revenue of any of those divisions, let alone the companies themselves, however central AI may be to their overall strategies.
Of those, possibly Google has the biggest dependency – and we don’t mean that in a pejorative sense – on machine learning because of its consumer businesses. Not only is ML at the heart of its ranking algorithms, but its massive advertising business couldn’t really function without ML in areas like smart-display campaigns and its broad match keyword tool, which uses ML to select keywords for the advertiser. Does that make it a pure-play AI company in the way that VMware was a pure-play virtualization company? Hardly.
An argument could be made for SAS Institute (which generated revenue of $3.1bn in 2019), which has been using ML for more than 40 years and offers tools used by data scientists to build and train models, but it has been a bit reluctant to brand itself as an ‘AI company’ (unlike thousands of other much smaller and less-qualified companies).
Salesforce was perhaps the first large software company to grasp the importance of AI for its business, which led (through a combination of organic and inorganic means) to the launch of Einstein in 2016, which the company has now deployed across much of its application portfolio. In some cases, customers pay more for Einstein features; in others, they come as standard. Again, Einstein alone is not going to drive a huge amount of revenue.
The applications businesses of companies such as Microsoft, Oracle and SAP all make use of AI to a greater or lesser degree, but they don’t derive that much revenue from it directly, when looked at in the context of their overall revenue.
Social media companies feed off the data exhaust we all leave, using that data to train and tweak their AI models to serve us what they believe to be more relevant and personalized content or offers. Facebook is obviously hugely dependent on AI, but it was launched in 2004, in the middle of another AI winter, and didn’t even start to use algorithms on its news feed until 2011. Of course, it makes use of the technology and has a renowned research division called Facebook Artificial Intelligence Research, but it’s more of a beneficiary than an original instigator. Facebook’s enterprise business – our focus here – is a tiny proportion of the overall company. In Q4 2020, advertising revenue was $27.2bn and ‘other’ revenue was $885m, which was driven mainly by sales of Quest 2 VR headsets, but there’s some Workplace by Facebook revenue in there too, just not very much.
Chip companies can be AI giants too
The myriad chip companies that have sprung up over the past five years or so to accelerate AI workloads represent one category that could also claim the crown of large AI companies. Founded in 1993, NVIDIA has become synonymous with its GPUs powering deep leaning and machine learning training, but it has many more arrows in its quiver than just AI – gaming makes up the largest portion of its revenue, generating $2.3bn of the $4.3bn it generated in Q3 2020.
Startups like Cerebras Systems and Graphcore are shipping AI chips aimed at accelerating the training or inference stages of the ML process (in Graphcore’s case, both), and a lot of other startups are in the development or sampling stages with similar plans to take on the incumbents (AMD, Intel and NVIDIA). One of those could make it to our large AI company mantel, if they’re not acquired by one of those big-three chip companies before they even get to go public, which is a likely outcome that we touched on as one of our AI trends of the year.
For genuinely sizable AI-specific companies, we need to look to China, where SenseTime, a specialist in computer vision, has raised at least $1.6bn in five rounds (and potentially more if Chinese media reports in January are confirmed) and Megvii has raised $1.4bn in five rounds. Both were heading for IPOs in Hong Kong in 2019 or 2020 before the political climate changed, and Megvii is now preparing to go public on the Shanghai STAR Market, according to S&P Global Market Intelligence, but has yet to formally file. SenseTime’s plans are rumored to involve an IPO at some point in 2021. Neither company’s revenues have been disclosed yet, but SenseTime’s is thought to be north of $500m, while Megvii’s is somewhere around $300m.
Future software giants
Although not large, there are specialist AI companies out there that are large in terms of valuations, if not yet revenue. Take C3.ai for example. It went public in early December, grabbing the ‘AI’ ticker symbol in the process, and saw its price open at $100 after having been priced at $42. Its market capitalization is currently about $13bn and its revenue is set to be less than $200m in 2020. DataRobot raised a $320m round in November – its 11th – at a $2.8bn post-money valuation, according to S&P Global Market Intelligence. Its revenue was about $140m, according to 451 Research’s private company database. Databricks – a different kind of company than C3 and DataRobot since it’s a data analytics platform provider – recently raised a $1bn series G funding round at a valuation of $28bn, but again, it’s hard to claim that it’s a predominantly AI company. The IPO market is currently buoyant for these kinds of software companies, so one of these firms may go on to be an independent AI giant, or end up within one of the large aforementioned software firms.