In partnership with

Ever since OpenAI scored a consumer hit product with ChatGPT, they’ve been exploring creative ways to shake off the organization’s non-profit legal status. The latest news is telling now that the transition to a for-profit has hit a roadblock and will not happen.

Looking at their business today, the profit part is a non-issue. Because OpenAI has no viable path to turning a profit in the coming years. Despite having 400 million weekly users for ChatGPT, it’s a machine that is only capable of producing ever increasing losses.

I may like to post rants about Microsoft and AI occasionally (well, regularly) but I’m no match to Ed Zitron and his passion. His latest article, Reality Check, is a brutally honest take on the state of AI as a business. More specifically, how it isn’t a business for anyone if you look at the real numbers.

The tech media at large isn’t so keen on covering the cold hard numbers of OpenAI et al because the story isn’t very compelling. There isn’t something cool and exciting to look forward to. After 2+ years of mainstream GenAI availability, the market leader hasn’t built a single profitable product yet. Even though they are raising billions from investors like SoftBank by showing projected figures like this:

I won’t repeat the numbers and analysis that Ed covers in excruciating detail in his newsletter. I’ll pick this one quote from the article to celebrate “2025, the year of AI agents”:

Since we’re getting just more & more proof that no OpenAI product is on the road to profitability, the AI evangelists are increasingly resorting to saying that profits don’t matter. “Who cares if the investors lose all their money when we’re heading for AGI?”

OpenAI is a company like no other. The first of its kind and thus difficult for us to benchmark against anyone else. So, let’s turn to the existing players instead to see what’s the state of profits in AI.

Microsoft’s agentic future & artificial profits

In the Business Applications Launch Event for 2025 Release Wave 1, MS shared the below momentum slide for Copilot Studio. The numbers of course are big as usual - hundreds of thousands/millions. But is it really a big deal?

Let’s do the math here. Microsoft is today promoting the pay-as-you-go pricing model of Copilot Studio as the preferred sales motion. The list price of one message is $0.01. While enterprise clients may get discounts, there’s also the chance of prepaid message capacity being unused, so things may even out. With this price point, Copilot Studio usage generates $2.5M revenue per month, and $30M per year.

Of the other metric shared, 160k organizations using Copilot, this translates to around 1.5K messages per org per month. Or 52 messages per day. Now, we have to remember that one action in a Copilot Studio agent often consumes more than one message. Even though Microsoft recently dropped the pricing of Copilot Studio in March, after only 2 months of having first announced it in January 2025, it still far less clear than one might imagine without examining the model:

If those 52 messages were only about regular GenAI usage without any business process logic, that would mean 26 responses from Copilot Studio agents per day. If they were to include things like agent actions (meaning, AI does something more than just chatting back at you) or AI tools, we’re quickly at a point where the average Copilot Studio customer organization does a couple of agent runs per day.

This is shockingly low. It is plain and obvious that most customers are merely experimenting with trying to build agents. Hardly anyone is running it in production yet. Which wouldn’t be that bad if this was a new 2025 product. But Copilot Studio has been out since November 2023. The service itself, Power Virtual Agents, which was a traditional non-AI chatbot has been around since December 2019.

So, five years after launching a product, it’s making $30M annual revenue. Let’s assume that it’s on a growth trajectory like the services OpenAI draws on its investor materials and Copilot Studio would do 3X of that in 2025. Roughly $100M of AI agent revenue. How does that register in software business where Microsoft has been in for 50 years now?

It doesn’t! It is still less than a rounding error. In FY24 Microsoft’s revenue was $245B. As in Billion. A $100M Copilot Studio licensing revenue from custom agents would be 0,04% of the entire revenue pie. MSFT is a massive business and the figures they deal with are beyond even the wild predictions of OpenAI in 2029. For reference, here is the latest quarterly income statement visualized:

What did Microsoft say about Copilot in the latest Q3 FY25 investor call then? Let’s look at the transcript:

“More than 230,000 organizations – including 90% of the Fortune 500 – have already used Copilot Studio.” Okay, so 30% of the users have not become monthly active users after their trials, when compared to the 160k figure on the momentum slide.

“This quarter alone, customers created over 1 million custom agents across SharePoint and Copilot Studio, up 130% quarter-over-quarter.” Right, because the term “agent” also refers to .agent files users can create almost accidentally by just clicking on the first item in the “New” menu of a SharePoint site:

“Hundreds of thousands of customers across geographies and industries now use Copilot, up 3X year-over-year.” Okay, so this was said in Microsoft 365 Copilot and the future of work context. Now, as much as Satya would surely like these to be subscribers of the $30/user/month M365 Copilot, I suspect it includes the users of the free ”Copilot for all” version - meaning Microsoft 365 Copilot Chat.

This leads us to an important aspect to address: even those Microsoft AI services that bring no direct licensing revenue aren’t being adopted by customers. At least when you compared them to the meaningful benchmark.

The void behind OpenAI

There were no figures mentioned about the consumer side Copilot offering in the quarterly earnings call. Luckily, there’s independent reporters out there who have managed to gain insights into how Microsoft feels about the state of things. This piece on the Newcomer from ex-WSJ Tom Dotan revealed an interesting moment from an internal MS meeting:

400 million vs. 20 million is a lot worse situation for Microsoft than Chrome vs. Edge, for example. That comparison is fitting in the sense that these are free tools to use. Only a small fraction of ChatGPT or Copilot usage is for paying subscribers. And yet when it’s free, people still choose not to use the AI product from Microsoft.

Hmm, I wonder if that would have anything to do with the forced inclusion of Copilot in every product, combined with the sneaky price hike of M365 personal/family subscription…?

In fairness, they don’t choose Google Gemini either. At least if you look at Google’s own internal disclosures shared in the courts:

When we look at the broad mass market adoption of AI tools, there really isn’t anyone else than OpenAI. Who, as we remember, is making more losses the higher its ChatGPT user volume grows. Microsoft and Google haven’t found a way to win over the hearts and minds of end users. How about vendors that aren’t selling services directly to users then? What does it look like for Amazon?

The capex spending for the big hyperscalers is still growing - despite the recent announcements about putting some of the planned datacenter projects on hold. Here’s a Yahoo Finance graphic of annual spend:

That’s the money spent on investments to enable AI services. How about the expected revenue back from those investments? You have to really zoom in to see any of it. In the case of Amazon, the analyst expectations for 2025 are $5B:

It’s again a big number on its own - yet completely out of proportion if you think about the money being poured into building AI infrastructure. In 2024, AWS annual revenue exceeded $100B for the first time. In 2025, they will spend more than $100B on investments to grow the AI market. And every dollar spent will give $0.05 back. Wow.

The real AI profit model

The bottom line is, there is no positive bottom-line impact from AI services anywhere to be seen. It’s still about the future. Once we reach “something”. To me, it’s starting to sound more like a religion than a business. If all we are waiting for is the arrival of AGI or ASI before the economic impact turns positive from all this, then why should everyone be rushing into deploying things at this moment yet?

So far, in the AI gold rush, the one verified positive revenue impact has been in the consulting business. Across the top-tier companies like Accenture, BCG etc. the revenue growth is not just a future projection but already something tangible. It almost seems like it’s impossible to go wrong when providing consulting services to customers who want to understand what to do with GenAI:

It’s a win-win situation for consulting companies. There’s huge billable demand from clients, regardless of whether they’re ready to implement GenAI or just exploring it. At the same time, tools like ChatGPT offer massive internal margin gains from automating non-client-facing work.

I’m sure that this doesn’t in any way affect the advice given by management consulting firms or IT service providers on where customers should be investing…