I've been tracking VC funding in my sector and the numbers are getting absurd. American companies plan to spend $400 billion on AI infrastructure this year alone. That's more than the entire GDP of most countries, for technology that still hallucinates basic facts.
My current company just raised a Series A by slapping "AI" in front of our existing product name. Literally the same team building the same damn features, but now our ARR projections are 10x higher because we added a shitty chatbot interface that half the time gives wrong answers.
The math makes no goddamn sense when you actually break it down. OpenAI reportedly burns through millions daily on compute costs just for ChatGPT. They'd need every single American to pay like $2.50 per day to break even at current usage rates. That's Netflix money for a chatbot that still thinks 9.11 is bigger than 9.9.
The Unit Economics Don't Work
We lose money on every customer because compute costs are insane, but investors don't care as long as we have 'AI' in our pitch deck.
Our investors don't care. They keep saying the costs will come down, but I've seen the roadmap. GPU prices are going up, not down. Power costs are increasing. The only thing dropping is our runway.
Anthropic is burning billions annually and has maybe 18 months of cash left at current burn rates. Their revenue is about $200 million annually. The math doesn't work unless they grow 20x in the next year.
Everyone's betting on massive scale bringing costs down, but that's exactly what crypto miners thought. Instead, AWS raised GPU instance prices by 30% when demand spiked.
What I'm Seeing in the Job Market
The job market feels exactly like 1999. Companies are hiring "AI engineers" for $300K+ salaries to fine-tune open source models. Most of these roles are just data engineering with extra steps and better marketing.
I know people who got promoted from "Senior Software Engineer" to "Principal AI Engineer" without changing what they actually do. Their LinkedIn profiles now say they're "building AGI" when they're really just implementing basic RAG patterns.
The Hacker News hiring threads are full of companies offering equity packages worth millions on paper. But the valuations are based on comps to OpenAI and Anthropic, which I think are themselves massively overvalued.
I'm getting recruited for AI positions that require "5+ years of LLM experience" when GPT-3 was released 4 years ago and wasn't even usable for most applications until 2023.
The Infrastructure Spending Is Insane
My company just signed a $2 million annual contract for NVIDIA A100 instances to train a model that might get 1000 users if we're lucky. The ROI calculations assume we'll magically scale to millions of users, but we've never successfully scaled any product past 50K users without everything catching fire.
I watched Oracle's stock price double because they announced partnerships with AI companies. Larry Ellison briefly became the richest person just from promising to host other people's GPU clusters. That should tell you something about market rationality.
I've been tracking the data center construction boom - it's approaching 2000 megawatts of new capacity. That's enough to power two million homes, all for training models that lose money on every inference.
I toured a new facility in Virginia last month. They're building 40MW data centers specifically for AI workloads. The power infrastructure alone costs $100 million before you put a single server in the building.
Personal Experience with the Bubble
My options from my previous AI startup are worth exactly jack shit. The company raised $50 million in 2023, burned through it in 18 months, and shut down last quarter. We had a "revolutionary" AI assistant that was really just GPT-4 with a custom prompt and a nice UI. I wasted six months building a complex RAG pipeline to make our chatbot slightly less stupid, only to realize we could have achieved the same results with three fucking lines of prompt engineering.
The founders immediately started a new AI company with the same pitch. Same investors, same business model, same inevitable outcome. But hey, they raised another $30 million because "this time it's different."
I'm back to building normal web applications, but now I have to call them "AI-powered" to get funding. We added a Hugging Face model that suggests text completions, and suddenly we're a "machine learning company" worth $100 million. The entire "AI" feature is literally 20 lines of Python calling the OpenAI API. I've built more complex autocomplete features for search boxes.
The scary part is how many people actually believe this horseshit. Investors keep saying AI will replace our entire engineering team. These people are completely fucking delusional - they can't even use Git properly but they think GPT-4 is going to write production-ready code.
When This Ends
I lived through the dot-com crash. This feels the same but bigger. Back then, we wasted money on Super Bowl ads and office perks. Now we're wasting money on compute clusters that cost more than skyscrapers.
The difference is physical infrastructure. When Pets.com failed, they just shut down a website. When AI companies fail, we'll have thousands of unused H100 clusters sitting in data centers burning electricity for no reason.
But I think the infrastructure will survive. Amazon, Microsoft, and Google will buy distressed AI assets for pennies on the dollar, just like they did with dot-com wreckage. The technology will eventually work, but I bet 90% of current AI companies won't live to see it.
My plan is to keep building normal software that happens to use AI APIs when they make sense. When the bubble pops, boring CRUD applications will still need to exist.