Nvidia’s blockbuster quarterly results make it plain that the race to build generative AI products is well and truly afoot. The GPU giant crushed earnings expectations in the second quarter and forecast a monster future. Investors, already content to value Nvidia north of $1 trillion, added tens of billions more to its market cap after the report.
The Exchange explores startups, markets and money.
Read it every morning on TechCrunch+ or get The Exchange newsletter every Saturday.
The hardware story is simple enough to write: Many tech companies are buying hardware to train their own AI models, and major cloud providers are also bulking up, both for their own purposes and to offer a service on their public platforms. Nvidia, meanwhile, is minting cash while taking orders.
But what about the software side of the equation? How are software companies faring in the generative AI era? There’s some hope that AI-related revenues can boost growth, but the real question is just how and when tech companies should charge more for AI-powered software tools, in addition to their current products.
Microsoft has taken big strides in monetizing AI. Not only can you pay for generative AI services on its public cloud platform, Azure, but you can also pony up for GitHub Copilot, which can generate code for you for $10 to $19 per month, per user. And the company is rolling out a $30 per user, per month add-on to its Office suite as well.
We’ve touched on how companies may charge for AI products. In May, we noted that some tech companies were planning to offer paid add-ons, which has become the Microsoft model to a degree. In contrast, some tech companies appeared content to bake new AI-powered tooling into their existing software for no extra fee. In June, we reported that a number of tech shops were waxing poetic about the power of proprietary customer data as a way to make their own AI projects more valuable.
Recent conversations with Amplitude and Appian, both public software companies, gave us much needed clarity on this crucial question of AI pricing. Amplitude CEO Spenser Skates, in an interview with TechCrunch’s Equity podcast, differentiated when to charge and when not to along the axis of new functionality versus accelerated functionality. And Appian CEO Matt Calkins had an interesting take on how companies can earn more from their existing software products with AI but not have to even raise prices. Let’s talk turkey.
New or improved?
When asked why Amplitude is not charging its customers for new AI features, Skates said (emphasis ours):
So I think you’ve got to be clear on what it is we’re doing with AI, versus what Microsoft and Salesforce and ServiceNow and all these other companies are doing with AI. Microsoft and [company] are using generative AI to help automate existing workflows: Let me write that email for you, let me fill out that ticket for you, let me respond to this question for you automatically. And so it makes a bunch of sense [to charge] because you’re introducing net new functionality that you haven’t had before. And you’re automating and saving a bunch of money by doing so.
What we’re doing with AI is a little bit different. We’re focusing it on automating existing workflows that you already use Amplitude [for] and making them a lot easier. [Our recent AI tools are] a perfect example [of this], where you [can] type in your query and it will generate a chart for you. People were already still doing that before AI; now it’s going to be a lot easier to do. We’re doing taxonomy management. We’re doing it with suggested chart names and suggested formulas. So there’s a bunch of places where we’re using AI to make existing workflows easier, not create net new ones. So it doesn’t make as much sense to charge for it. Instead, what you’re focused on is developing a tool that makes it easier to get on board and up and running on your product.
Skates is differentiating implementations of AI that automate existing workflows by executing an entire body of work from AI that accelerates existing workflows by helping the user do more, faster.
Both of those approaches are valuable, but, in Skates’ view, the former case better justifies an added charge than the latter. He double-clicked on that distinction later on in our conversation, saying that the major tech companies that put a discrete price on generative AI features are:
Charging for the ability to automatically fill out text for people [for example]. That makes a ton of sense. What we’re doing is speeding up a bunch of workflows. And when you’re doing that, it’s a distribution play, not a monetization play.
That’s fair enough, but my first impression of Skates’ comments is that he is being too generous. The tech sector has lately been reevaluating the strength of software businesses. Suddenly, it was no longer fashionable to burn cash to grow fast, and cash flow grew a halo of primacy. But given how low software pricing is today, it seems a lot of companies may not be collecting enough of the value they are offering as revenue.
My view here is more than ancillary: I write for a living, and AI-powered software companies charge others so bots can do that instead of people, right?
Let’s bring Appian into the discussion.
Appian is an interesting company. It does so many things in a single platform that it has remained pertinent across several tech cycles. When I was interested in the no-code/low-code world, Appian had a play there. When I was learning about robotic process automation and process mining, there I found Appian. And now that we are in this generative AI moment, well, Appian is here again.
For those unfamiliar with this company: Its service is predicated around data, automation, and the construction of web and mobile apps, and it rolled out a number of new AI tools and services this week. That combination makes it a great company to chat with at this moment in time.
According to its CEO, Matt Calkins, Appian sees a way to derive revenue from its new AI products and services without raising prices (emphasis ours):
I think that when we add value, we have leverage. And that’s just true. I think [AI] is valuable. So it does amount to leverage. It doesn’t mean we use it.
Remember, there’s two main ways that leverage can translate into revenue. One of them is, you put a price on it. You got something, you make the product better and then you put a price on it. The other way is, you make the product better and [then] you have more negotiating power; you eliminate competitors, you sell faster. Right? It can help you get a better deal.
If we were a factory and we sold candy bars, and every candy bar was the same but we decided to put hazelnuts in our candy bar, the only way to monetize that — if it was more valuable or expensive — would be to raise the price, because they all sell for the same [price] anyway. But in our case, almost every deal is a negotiation. So we don’t have to change the price in order to change the level of transaction. The official price and the price we eventually sign on are sometimes different, right? So we can translate leverage directly into revenue without the price level, as printed, ever changing.
When we noted above that some folks would charge for new AI tools and others would not, we missed a key piece of nuance: Software that’s been improved by AI can lead to greater revenues even sans price changes. I would hazard that Calkins’ prediction will also be true for Amplitude.
In other words: If new AI features can reduce the discounts you give to new customers, then it really is driving incremental revenue.
For startups, I view all of the above as good news. Why? Because there are a host of evolving models for generating more revenue for new tooling. From pricing new add-ons outright and raising prices for bundled offerings to simply getting an edge in negotiations, there’s a lot you can do. That means scrappy startups that are building and launching quickly this year should be able to reinforce their businesses.
It remains to be seen if the generative AI moment will really transform as much of the software world as some expect it to. But it’s clear that even if it doesn’t, AI could still prove a shot in the arm for anyone slinging hosted bits.