From today on, let’s all set aside 200 euros a month to use artificial intelligence. That’s more or less the proposal from both OpenAI, with its Pro subscription, and Google, with its Ultra subscription.
These so-called core subscriptions are what the AI Big Tech companies are offering to give you the best models and the best services, always at hand. The vision behind them - looking also at how they’re organizing partnerships and updating their interfaces - is that artificial intelligence becomes the entry point to technology and its features.
Another platform shift
We’ve seen various platform shifts over time, starting with operating systems, and each of them came with a fierce battle to win a monopoly. There was the operating system and the battle between Microsoft and Apple; the browser wars. With the web we saw battles over search engines, over social media, over office suites. Then came the mobile wars, Apple and Android. And today we’re facing a war over artificial intelligence.
We’re used to subscriptions of 10, 20, 30 euros a month. These high-end core subscriptions, in my view, are tied to big changes underway, and behind those prices there’s much more. Let me offer a more detailed analysis.
The interface: talking to a bot
These AI interfaces, in terms of platform shift, want to wrap up features and offer access to information that is flat, driven by intent. You talk to the bot, and the bot does things, organizes files, connects to various services. In this sense there’s a network of agents that can in turn fulfill these requests.
I’ll come back to agents, because in my opinion the role of us ordinary mortals who offer technology services lies in building these agents, just as we built websites and mobile apps in the past.
And the role of the human changes. Technology use is moving toward framing the human not so much as an executor, someone who knows what’s underneath and all the steps to take, but as a strategist, an innovator, a creator - someone with a high-level vision who mainly monitors the behavior and results of these agents, coordinates them, and supports them when authentication, payment or delicate authorization steps are needed. In this sense it’s the human who supports the machine, not the other way around.
The models are not the point
This platform shift concerns the interface, and yes, the AI models too. But on that front we need to wake up, because the focus hasn’t been on models for a long time. When GPT came out I was already going around saying “Guys, let’s build applications with AI, there are agents, there’s stuff to do”. And everyone was like “oh, they trained this model that scores 0.2% better on this and that”.
All of artificial intelligence is moving to the so-called application layer. That application layer is called agents: software that does things, and in doing them uses language models, computer vision models, multimodal models and whatever else. They’re fascinating hybrid architectures, and they’re the new apps or the new websites, but in the AI context. They’re still tied to long-tail dynamics, like YouTube videos, mobile apps and websites.
The infrastructure
These platform shifts also have to be seen in terms of infrastructure. Looking at the changes in interfaces, the move to the application layer, and the infrastructure investments together, you start to understand what this core subscription is really aiming at.
Notice, folks, that data centers on the order of gigawatts are currently under construction around the world. I took one as an example: the one OpenAI is building in Texas, more than 1 GW of compute, to have GPUs grinding away to train and run artificial intelligence.
You can picture a 1 GW data center as a battery capable of powering a city of half a million homes. To understand the orders of magnitude: a 1 GW data center draws as much energy as half a million homes.
In these data centers, which are enormous, there are hundreds of thousands of AI graphics cards - dedicated hardware for parallel computation - and each one costs tens of thousands of dollars. If you line up those zeros - hundreds of thousands, if not millions, of GPUs, each worth tens of thousands of dollars - you realize we’re talking about something immense.
I’d like this to land for those of you who don’t work in tech but watch these videos to stay informed: this is like building the railways, like building the electrical grid. It’s gigantic. In the United States there’s also the Stargate project, a 500 billion dollar investment to build these data centers, but there’s a lot happening in China, in the United Arab Emirates, all over the world.
Where does that leave us?
How do these changes affect us? Given how slow Europe, and Italy too, have been to react to these big changes, for various reasons - and I say this with great regret - in my view we’ll be third parties participating in that infrastructure, in particular the American infrastructure, and we’ll participate as builders of agents.
I think it’ll be hard enough for us to manage building that kind of thing, and even harder to carve out platforms of our own. So I’d start thinking in these terms: go build agents. These agents will be part of third-party American platforms, certainly not ours, circuits that sit mostly on that side of the world.
From subscriptions to pay-for-outcome
One last prompt, beyond the much-discussed core subscription. There’s another big topic I wanted to bring to your attention - tell me what you think: the change in business model for software as a service.
We’re used to subscriptions: you pay monthly and they let you use digital services. When the talk turns to agents, some people start talking about pay for outcome: instead of paying over time to use these tools, you pay when they actually solve a problem, deliver something, give you what you asked for. It’s an interesting concept, still poorly defined, but I see it discussed more and more.
Tell me what you think, and if you’re exploring these things, how it’s going. I’d have more to say about all this, especially for small businesses, but I’ll stop giving it all away on TikTok. Call me, pay me.