đAI is not about productivity gains.
At least not primarily. It's about selling.
TLDR Summary
AI means that machines perform tasks commonly performed by humans. At least thatâs what most people associate with it. Thatâs why everyone is focusing on the productivity gains that todayâs AI tools promise. Thatâs why many white collar workers fear unemployment.
IT-driven productivity gains will likely be meaningful in the coming years and generative AI will play a central role in that. But itâs not what the AI revolution is really about. Itâs not the prize that the big spenders are competing for. Itâs not about B2B.
Instead, itâs about B2C. Itâs about putting online marketing on steroids. The internet is the greatest product distribution machine ever created. The core of that machine is the quality of information and the trust in the messenger. AI tools set a new standard on both fronts. The winner will likely have the first $10 trillion Dollar asset ever created.
This battle is shaping up to have two finalists. OpenAI and Google. OpenAI has a huge lead. But Google has been catching up lately as the only true contender in the West. Itâs strength comes from its financial health, vertical integration and broad ecosystem of legacy services that can be empowered by AI.
However, I doubt the victory will come easy for them. OpenAIâs financial health can only be opined on in conjunction with NVIDIA which generates as much cash flow as Google does. They can and will finance OpenAI if they have to. And Altman is brutally pursuing the Zuckerberg playbook.
The internet is a giant product distribution machine.
From an economic perspective, the best way to think about the internet is to view it as a giant product distribution machine. Its primary purpose is to connect those who want to sell stuff to those who might want to buy that stuff. You can see this by simply looking at the most valuable companies the internet has created. Amazon is a marketplace which connects sellers with buyers. Google and Meta sell ads, also with the intent to connect sellers with buyers for a commission.
Why is helping sellers to find buyers such a high value service that it enables these companies to command trillion Dollar market caps? Because we live in a world of abundance. Whatever problem we could possibly be facing in our lives, chances are quite high that there is a commercial solution for it out there. Intense competition has made entrepreneurs extremely creative in coming up with new product ideas and how to manufacture them cheaply. Their primary problem is not how to improve their product portfolio. Itâs how to find customers that will actually buy their products.
As a result, developing and manufacturing is not the most valuable skill (anymore). Selling is. Distribution is the ultimate moat. Chances are quite high that you are seeing that in your direct environment as well. Who are the highest paid people at your company? Probably not the scientists in the lab or the automation engineers on the factory floor.
Itâs likely the people in sales. Those who own the customer relationships. Construction workers building homes typically make less than the realtors selling them. CEOs are primarily sales people. Partners in consulting firms or law firms are primarily sales people. Up or out is downstream from whether you can sell or not. Even the heads of hospitals and university professors are primarily sales people.
The internet is the most powerful sales machine ever created and it regularly goes through iterations that make it even more powerful. At first, websites were static and standardized. They showed the same content for every visitor. Companies bought ad space to show their products to everyone visiting the site. They had a very limited ability to target specific customers. Some websites had huge traffic. But in principle, such an ad wasnât any more valuable than a billboard on a much frequented highway.
Then search engines started employing techniques for more personalized ads. User data were stored to allow for a better targeting. Ads could then be placed for consumer groups that fulfilled certain criteria in their past internet activity. This was the first big differentiator between online distribution and traditional distribution in the early 2000s.
Then social media emerged. It allowed for even more personalized product distribution. Influencers attracted very specific groups of consumers to whom they can act as very high value brand ambassadors. They gather followers who have very similar interests and who place a lot of trust in their recommendations.
AI is the next step in this process of ever higher value online product distribution. That is the core of everything happening right now.
The true AI revolution is not about productivity.
To understand why, we first must properly think through what AI actually is. Britannica defines AI as âthe ability of a computer to perform tasks commonly associated with intelligent beingsâ. This includes for example forming conclusions based on data, recognizing patterns in data, solving problems, understanding language and human interaction, making decisions and generally creating digital content.
This definition entices people to view AI primarily as a productivity enhancement tool. It certainly is. But a calculator is, too. A phone also. Any machine is. Productivity enhancements have been happening since the beginning of civilization.
Itâs important to distinguish todayâs AI generation from this general idea of AI. When we talk about the proliferation of AI these days we typically refer to statistical models that digest large amounts of data and transform it using statistical methods into a very small and tailormade set of data for a certain task triggered by a prompt. This generates new data, but it doesnât generate new knowledge.
Itâs a powerful tool, though. One which has a ton of potential to automate business processes. But that automation is an evolutionary process, not a revolutionary one. We have been using information technology for business process optimization for decades. Most low hanging fruits have been harvested. Remaining problems need highly sophisticated solutions. And if AI models are hammers, not all of these remaining problems are nails.
Letâs think through the main functions of a typical corporation and where generative AI can be used. An obvious first application is in the interaction with customers. A company has a ton of internal data about its products. An AI model is a suitable method to provide this data to customers in a structured and useful manner. Call centers and marketing/sales departments can become more powerful.
Administrative tasks are also suitable targets for AI. You can use it for financial functions to create reports for example. It can help management to improve its sales, product and sourcing strategy. You will still need business analysts though to review these reports and form decisions. Generative AI may even increase the amount of content circulating in an organization in an extent that more people are needed to deal with it. I worked in consulting for a decade. New tools typically made us raise the bar of our work, not work less.
Many other business processes canât use AI tools at scale. Non-digital business processes for example. AI canât much about them because it isnât capable of embodiment, yet.
ChatGPT was launched in November 2022. Itâs not that business executives woke up the next day with the revolutionary idea to automate business processes. They are thinking about it today just like they have done ten or twenty years ago. Some efforts work, others donât. Itâs a trial and error process. I wasnât very surprised when the MIT titled in August that 95% of generative AI pilots at companies are failing.
It will take time and I am sure Big Tech executives are aware of that. If it was only about squeezing out more productivity, they wouldnât have entered into the current potentially suicidal capex race. They are fighting for the next moat in product distribution. The real lever is with the customers, not with the vendors.
Itâs about distribution.
Let me illustrate that through my personal AI use. I experiment with several chatbots, but I work primarily with ChatGPT. Itâs my research assistant that helps me to understand the topics that I am studying. All of my articles are 100% written by me. But the process of writing them involves asking questions to the AI. Often hundreds of them per article if itâs a challenging topic. In fact, itâs probably more than a hundred questions per day on average. Not just for my content creation, but for many other subjects that I am interested in outside of Fallacy Alarm. Developing my children, making progress on my athletic ambitions, learning about history, getting up to speed to controversies in politics and economics to name a few.
The chatbot directs me through the internet. It influences which content I consume. It directs me to podcasts of people or topics that I am curious about. It directs me to data sources to form and support my opinions. And ultimately, it drives my purchase decisions.
Some of my purchase decisions are on autopilot. I wonât ask ChatGPT what to buy in the supermarket. And when I need new t-shirts or socks, I just go on Amazon to get it over with. But when I think about my conscious purchases, i.e. those I donât do routinely where I have a very specific problem to solve, approximately 100% of them originate from a discussion with the chatbot. Sometimes indirectly, for example when I am learning about something and âaccidentallyâ figure that I might want to buy product A or B. Sometimes quite directly, for example when I start the conversation by explicitly inquiring about the benefits of product A or B.
Now, think about how extremely powerful it is to have such a close connection to billions of users like me. Some of them more gullible than me, others less, I am probably somewhere in the middle of the bell curve. Metaâs original moat was that Zuckerberg managed to get people to sign up with their actual real names and much other personal data. He did so by appealing to their vanity. Remember, Facebook started as a social medium for US elite universities. People were proud to be listed there with their real names. This allowed for extremely high value ad targeting.
AI platforms have a much more valuable moat. They learn about the deepest intricacies of peopleâs minds. Google Search is famous for receiving high value questions. But AI takes this to another level. Itâs so good with responses that it triggers an endless amount of follow-up questions which reveals a ton about a userâs preferences, much of which can be exploited commercially.
It doesnât only learn a lot about the user which allows for tailormade product offerings. These interactions also create enormous trust. Probably much higher than the trust people have in the influencers they follow. You donât trust Sam Altman. But you probably trust his chatbot when 9 out of 10 responses prove to be helpful. The bot gives back much more than an online creator could ever do. It will always respond and never judge. So, if the bot recommends a product at the end of a long conversation, the user will very likely buy.
And the monetization potential doesnât even stop there. AI platforms can create user profiles that do not only store their preferences, but also their skills. The product distribution goes then the other way. Imagine you are a company looking for new recruits. What stops an AI platform from selling users as candidates to you?
Measuring the userâs interaction with the chatbot likely allows for great insight into their background, knowledge and intelligence. The signal in that data is likely much stronger than in IQ tests or school grades. Especially since the emergence of AI completely invalidates many traditional means of measuring student progress (homework for example).
The winning AI platform will not necessarily be the one that excels at automating business processes. It will be the one that owns the majority of user interactions. The battle for that crown is shaping up to have just two finalists: OpenAI and Google. There are no other serious contenders in the West. Perplexity and Anthropic primarily target enterprise customers. There are and will continue to be nothing more than footnotes.
Letâs look at OpenAI and Google in more detail.
OpenAI is currently miles ahead in distribution.
Three years after the launch of ChatGPT and hundreds of billions of Dollars of industry spending later, ChatGPT still owns the chatbot space. It owns 61% of the market and 75% if you include Microsoft Copilot.
Itâs important to understand that this market share estimate doesnât measure revenue or user accounts or usage volume because this information is not publicly available. Instead, it measures search-traffic share, i.e. which chatbot people click when they search for AI tools.
This limits the value of the overview somewhat. But I consider it still useful. Other sources support this market share estimate, especially when you focus on regular usage, i.e. daily or weekly active users. It also largely aligns with the takeaways from my AI economics article a few months ago. OpenAI is by far the largest AI spender.
Based on this measurement approach, the market share is remarkably stable, pretty much unchanged for 15 months. It dipped a bit in November 2025 sequentially. But itâs still higher than in November 2024.
Google is far behind, but catching up.
Over the last few months, Google as emerged as the new AI darling. Itâs stock is up 80% in 2H25 so far. NVIDIA stock as a proxy for OpenAI is up just 20%.
Several factors have made this outperformance possible. Google has made technological progress on Gemini which was found to be outperforming ChatGPT on several standardized benchmarks. The company started to push it aggressively into its main products (Search, Android, Chrome etc.). Their 3Q25 earnings were strong with accelerating revenue and earnings growth. And there is a new narrative emerging that portrays Google as the first real challenger to NVIDIAâs GPU dominance.
I wonât attempt to opine on ChatGPT vs. Gemini or GPU vs. TPU from a technological perspective. Others are much more capable to evaluate that. However, I will highlight that Google still has a lot to prove from a market share perspective. Itâs highly speculative whether the latest market share increase will continue. At what pace it will continue. And how costly it will prove to be for Google to challenge OpenAI.
Geminiâs market share visibly jumped in November. Thatâs the first real evidence to support the $1.6tn market cap increase over the last five months.
ChatGPT still has a clear first mover advantage and people use it with intent. We donât know how much of Gemini use is simply the byproduct of Google pushing it to users of their established services who didnât want or need it.
Looking at single entities vs. looking at entire ecosystems.
One of the strongest and most popular arguments to favor Google over OpenAI is to point to Googleâs strong operating cash flow and OpenAIâs insane cash burn. Per 3Q25, Googleâs LTM free cash flow stands at $74bn. OpenAIâs free cash flow is unknown. There are estimates floating around online that their cumulative losses between today and 2029 will be north of $100bn. HSBC even estimates they will need to raise $200bn until 2030. The AI arms race will likely be about the last man standing. The winner will be who can spend the most. The odds seem to favor Google from that perspective.
I donât think itâs that easy though. I believe itâs important to look at AI from an enterprise value perspective, not from a equity value perspective. Itâs about valuing assets, not corporations. OpenAI canât be analysed financially without looking at its stakeholders. Most importantly: NVIDIA.
NVIDIA generated $60bn in FCF for the first nine months of their FY26. This means their annual FCF generation is easily on par with Google. OpenAI is NVIDIAâs most important customer. They will fund them if they have to. Even if they wonât get paid in cash, but in shares.
Being paid in shares might actually end up better for NVIDIA in the long-term. The global AI chatbot space will likely become the most valuable asset ever created. If it was put into a single legal entity structure, it would likely become the first 10 trillion Dollar stock. ChatGPT is in the pole position to capture much of that value. I highly doubt OpenAI will ever run out of money.
Remember Facebook.
It was founded in 2004 and went public in 2012 in a highly controversial IPO. It dropped pretty much immediately and lost over half of its value over the next months. Its $100bn valuation was ridiculed by many investors and financial commentators. The company had very little revenue at that time, just $4bn in 2011. Many doubted the company could turn on the money switch to justify the valuation.
Zuckerberg deliberately delayed monetization at that time in an attempt to figure out the true potential of the company first. He believed that focusing on monetization too soon would ultimately compromise the companyâs ability to thrive. Instead, he believed that making money was the ultimate byproduct of providing good services. The following quote from the IPO S-1 Founder Letter perfectly encapsulates that philosophy:
âWe donât build services to make money; we make money to build services.â
It seems that the AI revolution will play out very similarly to the mobile computing revolution that Meta ended up dominating. Most companies seem to be closely following the Zuckerberg playbook. Altman is probably doing that with the highest level of conviction. It will be very interesting to see how a more conservative opponent like Google with deal with that.
Sincerely,
Rene
Superb reframe that distribution,not productivity, is the real prize. Your point about AI revealing user prefernces through follow-up questions is understated. That loop of trust building plus data capture creates a moat way stronger than anything social media achieved. The comparison to Zuckerberg's delayed monetization strategy is spot on. Most investors miss taht OpenAI isn't just burning cash but positioning for control of the most valuable asset in histoyr.
Distribution indeed, excellent piece.