Remember when the internet was going to change everything, and any company with a ".com" in its name could raise a billion dollars on a napkin sketch?
If you're feeling a sense of déjà vu right now, you're not alone. We are currently living through the most aggressive capital deployment cycle in modern history. Venture capitalists are practically throwing money out of moving vehicles at anything with "LLM," "GenAI," or "Neural" in the pitch deck. But here's the uncomfortable truth: while the technology is real, the valuation bubble is arguably just as inflated as it was in 1999.
The question isn't if the correction comes, but who is holding the bag when it does. I've spent the last year watching the AI landscape shift from "cool experiment" to "boardroom mandate," and the parallels to the dot-com crash are becoming impossible to ignore.
What You'll Gain: By the end of this piece, you won't just understand the historical context of the AI boom; you'll have a framework to distinguish between the foundational infrastructure that will survive the winter and the thin-wrapper startups that will vanish into the ether.
The Infrastructure Layer: The "Picks and Shovels" Play
During the Gold Rush, the guys selling shovels got rich, not necessarily the guys digging for gold.
In the dot-com era, this was Cisco. They made the routers that powered the internet. Their stock skyrocketed because everyone needed their hardware. But here is the contrarian take: Being the infrastructure provider doesn't make you immune to the crash.
Cisco's stock famously dropped 80% from its peak and took two decades to recover its highs. Why? Because they overbuilt capacity based on demand that was projected, not actual.
Today, the "shovels" are GPUs, cloud compute, and data centers. Nvidia, Microsoft, and AWS are the modern-day Ciscos.
The Bull Case: They are generating actual, massive revenue right now. This isn't speculative future money; it's real cash flow.
The Bear Case: Their valuations assume that everyone will need 100x more compute forever. If the AI application layer fails to generate profit, the infrastructure spending will freeze overnight.
The Survivors: Companies with hard assets and essential utilities (Cloud providers, chip manufacturers). Even if valuations drop, they won't disappear because the world needs their compute power.
The Vanishers: Companies providing niche hardware that solves a problem that turns out to be temporary.
The Application Layer: The Thin Wrapper Trap
This is where it gets messy. This is the Pets.com territory.
Right now, there are thousands of startups that are essentially a fancy user interface wrapped around OpenAI's API. They have no proprietary data, no unique distribution channel, and no moat. They are what industry insiders call "Thin Wrappers."
If your entire business model is "We use GPT-4 to write emails better than the next guy," you are in grave danger. When the platform providers (OpenAI, Google, Microsoft) decide to build that feature into their core product (which they will), your company evaporates.
The Survivors: Companies using AI to solve a specific, painful workflow in a highly regulated industry (Health, Legal, Finance) where trust and compliance are barriers to entry.
The Vanishers: Generic content generators, basic art tools, and "ChatGPT for X" clones.
The "Moat" Theory: Why Data Beats Algorithms
In the late 90s, the technology stack was the differentiator. But as the tech became commoditized, the survivors were the ones who built network effects (eBay, Amazon) or aggregated unique data.
The same rule applies to AI. The model itself (GPT-4, Claude, Llama) is becoming a commodity. It's the fuel, not the engine.
Bad AI Strategy: "We have the smartest algorithm." (Nobody cares. It'll be open-source in six months).
Good AI Strategy: "We have 10 years of proprietary medical records that this AI can analyze in seconds, and nobody else has access to this data."
The companies that will vanish are the ones relying on the "magic" of the AI itself. The survivors are using AI to unlock value in data or workflows they already own.
The Revenue Reality Check
Here is the most sobering parallel to the dot-com crash: Revenue is vanity, profit is sanity.
In 1999, companies went public with zero revenue. Today, we have AI companies raising Series B rounds with zero revenue, just a "vision." The market is currently tolerating this because of FOMO (Fear Of Missing Out).
However, when the tide goes out (and it always does), the market will stop asking "What can you build?" and start asking "What did you earn?"
If an AI startup cannot show a clear path to 10x ROI for their customers within the next 18 months, they will find it impossible to raise another round. The bridge financing will dry up, and they will fold.
The "Service-as-Software" Shift
Here is the unique insight that many are missing: The biggest disruption isn't software replacing software; it's AI replacing labor.
The winners in the next phase won't be SaaS (Software as a Service) companies selling a tool for $20/month. They will be "Service-as-Software" companies that sell a result for $2,000/month, replacing a human consultant or agency.
Which vanish: Tools that help humans do their job 10% faster.
Which survive: Systems that do the job entirely, with a human supervising the output.
If your AI company sells a tool that makes a graphic designer slightly more efficient, you're at risk. If your AI company replaces the graphic designer's output for a flat fee, you're onto something massive.
Actionable Takeaways: Your AI Survival Guide
Whether you're an investor, a founder, or an employee at an AI startup, here is how you navigate the coming contraction:
Audit the Moat: Ask yourself, "If OpenAI/Google released a feature tomorrow that does exactly what we do, would we die?" If the answer is yes, you are a feature, not a company. Pivot to proprietary data or deep workflow integration immediately.
Focus on Gross Margin: AI is expensive to run. If your gross margins are 30% because you're paying for GPU time, you're in trouble. The survivors will figure out how to run smaller, specialized models or pass the compute cost to the enterprise customer.
Ignore the Hype Cycle: Stop building based on what is trending on Twitter/X. Start building based on what is painful for your customers. The best AI companies right now are boring. They are automating supply chains and insurance claims, not generating funny poems.
The Verdict
The AI bubble will pop. It's not a matter of "if," but "when." The air will come out of the valuation balloon, and the "tourists" (the VCs who jumped in just to chase the trend) will flee.
But here is the good news: Unlike the dot-com crash, the AI technology is actually useful. When the crash comes, we won't lose the technology; we will just lose the excess valuations.
The survivors will be the companies that use AI to solve boring, expensive problems in unsexy industries. The ones that vanish will be the ones trying to sell you a chatbot for $29.99 a month.
What's your take? Do you think the AI bubble is about to burst, or are we still in the early innings? And more importantly, what is the biggest "moat" you've seen an AI company build so far? Let me know in the comments.
Top comments (0)