OpenAI Will Burn Through $115 Billion by 2029 and Still Might Not Turn a Profit

OpenAI just admitted they're going to burn through $115 billion by 2029, which is $80 billion more than they thought they'd need just a few months ago. That's not a rounding error - that's "oh shit, we have no fucking clue what we're doing" territory.

They're projecting something like $12 billion in revenue for 2025, ramping up to over $100 billion by 2029 - if you believe their math. Sounds impressive until you realize they'll still be hemorrhaging money even at peak revenue. Each ChatGPT query costs actual money to run, and they're finding out the hard way that AI economics are completely fucked.

Why AI Companies Burn Cash Like Teenagers With Dad's Credit Card

Here's why OpenAI's spending is so insane: AI isn't like normal software where you write code once and serve millions of users. Every time someone asks ChatGPT a question, it costs actual money to run thousands of GPUs that cost hundreds of thousands each. Plus massive data centers with cooling systems that could power small cities. And the electricity bills? Don't even ask.

Their recent deals show just how desperate they are for compute power. The $300 billion Oracle partnership and $10 billion Broadcom chip order are just the tip of the iceberg. Then there's the $500 billion Stargate Project with SoftBank and others - basically throwing money at the problem and hoping it works.

Everyone's Throwing Money at AI Like It's 1999 Again

OpenAI isn't alone in this spending madness. Meta, Amazon, Google, and Microsoft are collectively burning through roughly $320 billion in 2025 on AI stuff. That's more money than most countries' GDP, betting that AI will somehow become profitable before they all go broke.

Here's the scary part: MIT found that 95% of companies trying AI aren't seeing jack shit for returns. Forbes analysis of the MIT report shows that even the 5% of successful AI implementations struggle with scalability. The failure reasons identified by MIT point to fundamental issues with how companies approach AI deployment, while McKinsey research similarly found dismal ROI across enterprise AI initiatives. Even worse, AI adoption is actually declining among big companies - you know, the ones with actual money to spend on this stuff. Enterprise AI pilots are failing to deliver meaningful ROI, despite massive investments. So we have companies spending hundreds of billions on technology that most customers don't want to pay for. This reminds me of Webvan burning $1.2 billion on grocery delivery that nobody wanted, or Quibi blowing through $1.75 billion because they assumed people wanted to watch shows vertically on phones.

Here's the uncomfortable truth nobody wants to say out loud: AI companies have fundamentally broken economics. Every time someone asks ChatGPT a question, OpenAI loses money. Not metaphorically—literally burns cash on GPU compute, electricity, and data center costs. This is the opposite of every successful software business ever built.

Traditional SaaS is beautiful: write code once, serve millions. Each additional user costs almost nothing. AI is the inverse: each query costs real money, and the costs scale linearly with usage. It's like running a restaurant where every meal costs more to make than you can charge for it.

The numbers are insane when you break them down: 700 million weekly ChatGPT users, but only 5 million pay the $20/month. That's a 0.7% conversion rate trying to subsidize 99.3% freeloaders. Even if OpenAI converts 10x more users to paid plans, they're still spending billions serving free users while hoping enterprise customers will somehow make up the difference.

AI Coding Tools Are Making Developers Slower, Not Faster

Here's where it gets worse: the one area where AI was supposed to print money—developer productivity—is turning into a shitshow. Studies show developers using AI tools are actually slower when you measure results, not just lines of code produced. A METR study found AI coding tools made experienced developers 19% slower despite developers thinking they were faster. Research by Bank Info Security confirmed AI tools slow down experienced developers by significant margins. Sure, GitHub Copilot spits out 10x more code, but most of it is buggy garbage that takes longer to debug than writing it properly from scratch.

I've seen this firsthand: junior developers "vibe coding" with ChatGPT, generating thousands of lines they can't understand, debug, or maintain. The code looks impressive in demos but falls apart in production. Companies are starting to realize that AI-generated code often introduces security vulnerabilities that cost more to fix than the time saved writing it. Studies show 48% of AI-generated code contains security vulnerabilities, while GitHub Copilot specifically had 38.8% of programs with security flaws. Research found 36% of Copilot code suggestions had vulnerabilities ranging from SQL injection to hard-coded secrets. One study documented 29.5% of Python and 24.2% of JavaScript code from Copilot containing security weaknesses.

The enterprise market that's supposed to save OpenAI's economics? They're getting burned by AI code quality issues and pulling back. When your "productivity tool" increases technical debt and creates security holes, CFOs start asking hard questions about that $20/month per developer spend.

We've Seen This Movie Before

This feels exactly like 1999 all over again. Massive spending on technology with no clear path to profits, companies burning through billions while promising they'll figure out the business model later, and investors throwing money at anything with "AI" in the name. Analysis comparing the AI bubble to the dot-com era shows striking similarities in valuation patterns and investor behavior. Recent surveys show nearly half of money managers believe we're in an AI bubble, while market analysts warn the AI bubble may be worse than 1999.

I remember when pets.com was spending millions on Super Bowl ads while losing money on every dog food sale. OpenAI spending $115 billion by 2029 while losing money on every ChatGPT query is the same fucking playbook. The New Yorker's analysis of the AI boom draws direct parallels to the unprofitable internet companies of the late 1990s. The only difference is GPUs instead of fiber optic cable.

The dot-com crash wiped out 78% of the NASDAQ and took 15 years to recover. Companies that survived had to actually, you know, make money. Turns out "we'll make it up in volume" doesn't work when your unit economics are broken.

Look, the internet eventually became essential infrastructure, but most of the companies from that era went bankrupt first. The dot-com crash wasn't the end of technology - it was actually the beginning of more sustainable, mature digital innovation. Maybe AI follows the same pattern - massive crash, then gradual adoption of the stuff that actually works. Wall Street predictions suggest the AI bubble will burst in 2025 from multiple converging factors. But if you're betting your career on OpenAI being the Amazon of AI instead of the pets.com, you better hope they figure out how to stop bleeding cash before 2029.