In a single day, OpenAI laid out the two pillars of its next empire: first, it signed a sweeping deal with AMD to secure no less than six gigawatts of GPU compute, an agreement that could give it up to a 10% stake in AMD if certain milestones are met. Then, on stage at DevDay, it unveiled a new layer of “mini-apps” that live inside ChatGPT, turning the chatbot into something much bigger: not a product, but a platform.
Together, these moves define OpenAI’s ambition with perfect clarity: control the power and control the interface.
Power, literally
The AMD deal is more than a supply contract: it’s a signal. Six gigawatts of GPU compute by 2026, the first one-gigawatt plant in construction, and stock warrants worth up to 160 million shares at a cent apiece if performance goals are hit.
That’s not procurement: it’s vertical integration through financial engineering. By embedding itself in AMD’s roadmap for the next-generation MI450 chips, OpenAI is locking in compute capacity at a planetary scale. It’s also buying influence: the right to co-design, the ability to shape pricing, and a hedge against Nvidia’s dominance.
Compute has become the new oil, and OpenAI just secured drilling rights.
From app to ecosystem
Then came DevDay. On stage, Sam Altman introduced “mini-apps” from Spotify, Canva, Expedia, Zillow, and others, micro-interfaces that live inside ChatGPT. The goal: let users interact with third-party services without ever leaving the chat, OpenAI’s bid to make ChatGPT your conversational operating system.
Think of it as the app store without the store. No icons, no screens, just conversation. You ask ChatGPT to plan a trip, it calls Expedia; you ask about housing, it queries Zillow; you design a logo, and Canva appears, seamlessly. The interface disappears. The agent decides.
This is not a super-app in the Asian sense. It’s something deeper: an orchestration layer that sits above every other digital service, turning natural language into the default control surface for your digital life.
If it works, ChatGPT stops being a chatbot and becomes the front end of the internet.
We’ve been here before
Anyone who has watched the history of Silicon Valley knows how this story goes. Platforms begin as enablers and end as gatekeepers. In the 1980s, Microsoft used Windows to control distribution. In the 2000s, Google turned search into an auction for attention. In the 2010s, Apple and Meta built app stores and ad ecosystems that extracted rents from everything that passed through them.
Now, the interface itself, the conversation, becomes the platform. And the pattern is repeating.
When ChatGPT “suggests” which app to use, who decides which ones appear? Zillow proudly claims to be the exclusive real-estate partner inside ChatGPT today. But what happens when competitors arrive, and we all know they will? Will placement depend on merit, or on bidding? Will we see a market where companies pay for their slot in the agent’s recommendations, as SEO for AI conversations?
History suggests we will. The difference is that, this time, there’s no search results page to scrutinize. The decision happens invisibly, in the flow of a chat.
The illusion of agency
For users, the promise is pretty seductive and sounds apparently very well. You no longer need to juggle tabs or apps, the agent does it all, it even starts the conversations. But the price of convenience is asymmetry. When you ask ChatGPT to “find the best flight,” you’re not searching, you’re delegating. And we all know that delegation without transparency leads to dependence.
Who audits the logic behind your agent’s choices? What data informs them? What economic incentives bias them? The more the interface simplifies, the more opaque the underlying process becomes.
We’ve spent two decades complaining about algorithmic black boxes in search and social media. Now we’re about to build one around every digital decision we make.
Compute as a barrier, distribution as capture
The AMD alliance and the mini-apps announcement are two halves of the same strategy. Compute is the barrier to entry, distribution is the mechanism of capture.
By securing vast energy and chip capacity, OpenAI ensures that no competitor can easily match its scale. By embedding itself as the interface to other apps, it ensures that even if competitors exist, they’ll have to go through its ecosystem to reach users. It’s the classic Silicon Valley playbook, executed with breathtaking speed and a layer of AI pixie dust.
Altman learned from the best. He watched Apple, Google, and Facebook turn control of interfaces into control of economies. Now he’s applying the lesson to the age of agents: own the conversation, and you own the user.
The energy question
The AMD deal also underscores an uncomfortable truth: large-scale AI is energy-intensive by design. Six gigawatts is roughly the output of six nuclear reactors. Training and running advanced models already consume staggering amounts of power. What happens when the world’s most popular interface is also one of its biggest electricity buyers?
OpenAI is not just building software: it’s building infrastructure with a carbon footprint and geopolitical consequences. When a private company starts locking up gigawatts of generation capacity, regulators should treat it not as a startup, but as a utility.
The governance gap
Every platform shift creates governance lags: rules arrive years after dominance is established. That’s how we ended up with app-store monopolies, ad-tech cartels, and search markets worth trillions, but accountable to no one.
ChatGPT’s platformization is happening faster than any previous transition. And regulators, distracted by content moderation and copyright disputes, seem completely unprepared.
The risks are not theoretical. Once an agent acts on your behalf (booking travel, recommending purchases, even making hiring decisions) it will be impossible to disentangle convenience from manipulation. The more we outsource judgment to machines, the easier it becomes for those who own the machines to shape our behavior.
What happens next
The momentum is undeniable. OpenAI is buying computing, embedding partners, and positioning ChatGPT as the front end of everything. The financial press reads it as a triumph of execution. The tech industry reads it as the dawn of agentic computing. Both may be right.
But beneath the excitement, there’s a warning written in the footnotes of tech history. Every time a platform promises frictionless integration, it ends up centralizing power. Every time we think “this one will be different,” it isn’t.
I’m not one more European obsessed with regulating everything, I’m just old enough to remember several previous experiences akin to this one. The world doesn’t need another operating system that mediates access to everything: it needs transparency, interoperability, and competition. If we don’t insist on them now, we may find ourselves living inside the most powerful black box ever built: one that doesn’t just answer our questions, but quietly decides which ones we’re allowed to ask. Be warned.