DEV Community

TechSparkLive
TechSparkLive

Posted on

The AI Industry's Trillion-Dollar Infrastructure Problem

OpenAI wants ChatGPT to replace your browser. They're adding apps, chasing cheap subscriptions in Asia, and spending a trillion dollars they don't have on data centers. Google thinks the answer is just letting anyone build whatever they want.

1. Apps Inside ChatGPT

You can now use Spotify, Figma, and Expedia without leaving ChatGPT. Just ask it to do something and the app shows up in the conversation. Need an apartment? ChatGPT pulls up a map and you can ask questions about listings without opening Zillow.

OpenAI already launched checkout last month. Now they've got Uber, Instacart, and DoorDash integrated. If this works, they take a percentage of everything you buy. Plus all the data about your shopping habits.

They also launched AgentKit, which lets developers build AI agents in minutes instead of weeks. An engineer built two working agents on stage in eight minutes.

The problem: nobody's proven people want to shop through a chatbot. Product searches, sure. But actually buying stuff? That's different.

openai

2. A Trillion Dollars in Deals That Don't Add Up

OpenAI needs massive data centers to run AI models. Data centers cost billions. OpenAI doesn't have billions. So they're making deals where they pay with equity and promises instead of cash.

Here's how it works: Nvidia "invested" $100 billion, but they're not giving OpenAI money. They're giving GPUs, the expensive chips needed to run AI. In return, Nvidia gets equity in OpenAI. Nvidia is trading their products for ownership in the company.

AMD is giving OpenAI up to 10% of AMD itself in exchange for OpenAI using their chips. They are literally handing over part of their own company to get OpenAI as a customer. Then there's Stargate ($500 billion) and Oracle ($300 billion). All massive commitments to build data centers. These aren't funded yet. They're promises about future construction.

When Nvidia's CEO was asked on CNBC if OpenAI can actually afford all this, he was blunt: "They don't have the money yet." He also mentioned that each gigawatt of data center capacity costs $50-60 billion. OpenAI is committing to multiple gigawatts without the revenue to pay for it.

Sam Altman says more OpenAI deals are coming. The entire industry is betting on AI models that don't exist yet, using money they haven't raised, for demand nobody's proven. It's a massive bet that everything works out before the bills come due.

3. ChatGPT Go Goes to 16 More Countries

chatgpt_go

The $5 subscription tier just launched in Afghanistan, Bangladesh, Bhutan, Brunei, Cambodia, Laos, Malaysia, Maldives, Myanmar, Nepal, Pakistan, Philippines, Sri Lanka, Thailand, East Timor, and Vietnam.

You get higher message limits and better memory than the free version. Some countries can pay in local currency, others use USD. OpenAI says their Southeast Asia users grew 4x and India subscribers doubled since August.

Google is doing the same thing giving cheap AI subscriptions in developing countries. Their version includes 200GB of storage and costs about the same.

OpenAI has 800 million weekly users now, up from 700 million two months ago. The company is worth $500 billion but spent $7.8 billion more than it made in the first half of 2025. The plan seems to be: get as many users as possible in Asia while building a commerce platform for rich countries.

Both strategies depend on infrastructure that costs a trillion dollars and users actually wanting to use it.

4. Google's Open Approach vs OpenAI's Walled Garden

Two days after OpenAI announced their app platform, Google launched extensions for Gemini CLI. It's basically ChatGPT for programmers working in their terminal.

The difference is in how extensions work. Extensions are add-ons that give the AI new abilities, like generating images or connecting to databases.

OpenAI's approach: Want to build an app for ChatGPT? You need approval from OpenAI. They pick which apps get featured, take a cut of transactions, and maintain a curated list of partners like Spotify and Uber. It's like the iPhone App Store. Polished but controlled.

Google's approach: Want to build an extension for Gemini CLI? Just publish it on GitHub. No approval needed, no partnership required, no fees. Anyone can build anything. It's like the open web. Messier but unlimited.

Gemini CLI has about a million users, mostly developers. Google even uses it to build their own products. The bet is that developers will choose tools where they don't need permission or have to share revenue.

Developers typically prefer open systems. That's why open protocols tend to win even when closed systems look better initially.

The strategy split: OpenAI wants regular people shopping through ChatGPT and takes a percentage of sales. Google wants developers building tools with Gemini CLI and charges nothing. Different customers, different business models, different philosophies about who controls the ecosystem.


The AI industry is splitting into different strategies: controlled platforms versus open protocols, premium commerce versus cheap subscriptions, building infrastructure versus using what already exists.
The question is whether any of these bets pay off before the money runs out.

Top comments (0)