BOND–Artificial Intelligence#3 | 매거진에 참여하세요

AI Isn’t Free – Inside the Business Models and Budgets Behind the Boom

Based on insights from BOND's 2025 AI Trend Report

By now, you’ve seen the headlines: ChatGPT. Claude. Gemini. Perplexity. New AI models seem to appear every week.

But have you ever wondered:

Who’s paying for all of this? How do these systems stay online—and why are they free (or cheap) to us?

Let’s make one thing clear: AI isn’t just smart technology—it’s expensive infrastructure,

fueled by staggering investments, powerful GPUs, and highly strategic business models.

AI Is a Billion-Dollar Engine

Behind every smooth chatbot interaction is a machine powered by tens of thousands of GPUs,

running in multi-billion-dollar data centers, funded by some of the sharpest investors on the planet.

In 2024 alone, just three companies—Amazon, Google, and Microsoft—invested over $150 billion USD (~200 trillion KRW) into AI infrastructure.

At the center of this investment? NVIDIA.

Why? Because they sell the H100—arguably the most in-demand chip of the AI era.

One H100 costs around $25,000 to $30,000, and training large models like GPT-4 requires thousands of them.

No surprise NVIDIA is considered the biggest winner in the AI race.

Everyone’s Building Their Own Chips Now

To reduce dependency (and cost), Big Tech is racing to build proprietary AI chips:

• - Google has TPUs

• - Amazon has Trainium

• - Apple has the Neural Engine

• - Tesla has Dojo

This vertical integration—from hardware to software to services—is now a survival strategy in AI.

The logic is simple: those who control the stack, control the future.

But Where’s the Money Coming From?

Despite all this investment, profitability is still elusive for most AI companies. Let’s look at the key revenue streams:

1. Subscription Plans: Charging for Access

• - OpenAI’s ChatGPT Plus: $20/month

• - Anthropic’s Claude Pro: Similar pricing

These subscriptions add up—OpenAI alone generates an estimated $2 billion in annual revenue. But for models this expensive to run, that’s not enough.

2. API Sales: The Real Business

The most lucrative source of revenue comes from businesses. If a startup wants to embed GPT into its app,

it must pay to access OpenAI’s API. This B2B revenue model accounts for the bulk of most AI companies’ income.

• High-volume usage

• High unit pricing

• Recurring contracts

3. Custom Models & Agents: Tailored AI as a Product

More companies now want AI models trained on their own data,

built for their exact needs. That’s where custom LLMs and agent systems come in.

• - Salesforce launched Agentforce

• - xAI introduced Auto-agent

• - Consulting firms like PwC and Bain now offer enterprise-grade AI deployment as a service

4. Data Vendors: Powering the AI Behind the Scenes

Training powerful AI models requires high-quality, labeled data. Enter a new class of businesses:

• - Scale AI

• - Snorkel AI

• - CoreWeave

These companies either label, refine, or pipeline data for LLMs—turning data into a revenue stream.

The Harsh Truth: Most AI Startups Still Lose Money

Yes, even the biggest names are mostly operating at a loss.

Why? Because operating costs scale with user growth. More users = more compute = more cash burned.

Who Will Survive?

According to BOND and other market analysts, the winners in AI will be companies that:

• - Own or control their chip supply

• - Have platforms to bundle AI with

• - Own their users and data

Think of:

• - Microsoft: Windows, Office, Azure

• - Amazon: AWS, Prime, Alexa

• - Google: Search, Android, YouTube

These giants don’t just build AI—they can distribute, monetize, and reinforce it across ecosystems.

TL;DR – AI Runs on Money

AI is no longer just a breakthrough in computer science. It’s a capital-intensive, full-stack industry.

It’s not enough to build a great model.
You need to power it, scale it, and sell it.

In the age of AI, cash isn’t just king—it’s the GPU budget.