How To Think About Value in AI

Ever since big data came on the scene, the tech industry has been on an incline of hype and investment in data and AI. From big data to deep learning, and now to generative AI. The rate of innovation has rapidly accelerated, and today it's near impossible to keep up with the capabilities of newly released models. To make matters worse, everyone has tacked "AI" onto their product marketing this year.

So, how do you figure out where the value is? Whether you're trying to pick a winner among today's competitors, or whether you're investigating AI use cases in your own industry, things aren't totally obvious.

Despite the endless cycle of releases, the value is not in the model itself: they are becoming commoditized by virtue of their broad availability, either through open source or cheap access. Instead, lasting value resides in ownership or strong influence over one or more of: silicon (processing and storage), data, or workflow integration.

Let's break this down further.

More data beats clever algorithms, but better data beats more data —Peter Norvig

• Data: a model is ultimately bound in its quality by the data available to train it. Little wonder then that existing aggregators of data such as Google and Microsoft have an immediate advantage. However, the true advantage comes from having "better" data, whether that's an enormous breadth, or deep proprietary data. In an enterprise setting, you'll only reap a competitive advantage from AI if you are able to bring unique data to the table. This makes things even more attractive for Google and Microsoft, with vast amounts of data being stored in Workspace and Office 365, and don't count Oracle out of this particular race.

During a gold rush, sell shovels

• Silicon: this is a no-brainer. AI consumes massive amounts of processing, and as Nvidia will tell you, it's great to be in the chip business. The advantage applies equally to the cloud providers, who are the practical means of accessing AI compute for most of the world. But, beware! The silicon owners need to drive usage ever upward, so they may not always be the most trustworthy sources if you're looking to decode what's hype and what's not.

You can check-out any time you like, but you can never leave

• Workflow: it's early days with AI, but if you can lodge your solution into an enterprise workflow, you have a point of lasting value. (Consider Oracle's ERP business: everyone grumbles about paying Oracle, but it's almost impossible to change.) For generative AI, this confers an advantage to companies whose tools are already in your business: Microsoft, GitHub, Google, Adobe, for example*. And as compelling as tools like ChatGPT's Canvas and Claude Artifacts may be, it's highly unlikely to see them as challenging to Google Docs or Microsoft Word.

When presented with new AI tools and use cases, use this framework to help evaluate whether it's part of the hype, or provides lasting value. It applies equally whether you're looking to analyze the competitive landscape, or wondering where AI can provide value for your own organization.

And finally, if you want a more detailed and contrarian take on the AI hype, I highly recommend Ed Zitron's newsletter.

--

* One of the more difficult aspects today of inserting AI into workflow is pricing. Right now, the value delivered to the user rarely justifies the extra it costs for the AI features—GitHub Copilot is the prominent exception here. This means vendors are eating the bill themselves, and training users to expect AI for low or no cost. Over time, inference will become cheaper and likely solve this problem.

--

This article is part of a series in which I'm sharing hard-won insights from my tech career. Subscribe to Wilder Thoughts to read more, and if you like this post, please share it to your network!