The Ai Bubble Will Burst Harder Than Crypto

Behind the screens. A publication on Medium about technology, culture, and humans.

Credit to Author

I was convinced we’re living in the biggest tech bubble in history when Mira Murati’s AI startup Thinking Machines Lab raised $2 billion at a $12 billion valuation without having announced a single product.

When a six-month-old company with zero revenue can command a valuation higher than most Fortune 500 companies based purely on the founder’s resume, you know the market has lost all connection to reality.

But Murati’s deal was just the tip of the iceberg. When generative AI companies worldwide raised $56 billion from VCs in 2024 across 885 deals, a 92% increase from the previous year, something tells me that we’re witnessing the mother of all tech bubbles, and the crash will make crypto’s collapse look like a gentle correction.

The numbers are something I have never seen before. Close to a third of all global venture funding went to companies in AI-related fields in 2024, making AI the leading sector for funding. What they don’t want you to know is that 90% of these “AI companies” are just expensive wrappers around the same five foundational models.

Murati’s $12 Billion Reality Check

Mira Murati’s $2 billion seed round is the perfect symbol of our bubble economy. The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, making it by far the largest seed round in the Crunchbase dataset.

What exactly did investors get for $12 billion? A former OpenAI CTO who has yet to reveal what the company is working on and promises to announce its first product “in the next couple months”. The company’s work remains unclear, yet it’s a round of unprecedented hugeness.

This also reminds me of Ilya Sutskever’s Safe Superintelligence (SSI) makes her deal look almost reasonable. The former OpenAI chief scientist raised $1 billion at a $5 billion valuation just three months after founding the company. Six months later, that valuation skyrocketed to $32 billion after raising another $2 billion, despite having no product, no revenue, approximately 20 employees, and a website that’s literally just a single page of text. As one report noted: “The high valuation seems to rest entirely on founder Ilya Sutskever’s reputation.” Well, of course.

Compare this to the dot-com bubble: even Pets.com had a website and inventory before going public. Thinking Machines Lab raised more money in six months than most successful companies generate in revenue over their entire lifetimes without shipping anything.

Investors Burn $100 Billion

The scale of this bubble dwarfs anything we’ve seen before. AI startups raised $110 billion last year, 62% more than the year before, while AI startups raised $104.3 billion in the U.S. in the first half of this year, nearly matching all of 2024. The valuation madness is reaching fever pitch.

Consider these recent examples:

OpenAI raised $40 billion at a $300 billion valuation
xAI went from $24 billion to potentially $45 billion in a year
Sierra jumped from $1 billion to $4.5 billion in eight months — a 4.5x increase
Perplexity nearly tripled its valuation from $3 billion to $8 billion in months
Anthropic raised $3.5 billion at a $61.5 billion valuation

The multiples are completely detached from reality. LLM Vendors top the chart with an average revenue multiple of 44.1x, while seed-stage AI companies show an average multiple of 20.8x despite generating just $2M in revenue on average. Some categories are trading at 70x revenue multiples.

Compare this to crypto’s peak: even during the 2021 crypto frenzy, VCs poured $30 billion into the market compared to AI’s current $100+ billion annually. The AI bubble is more than three times larger than crypto at its most euphoric moment.

Goldman Sachs expects Big Tech to spend over $1 trillion on chips and data centers to power AI over the next five years. Software engineer Stephen Diehl puts this in perspective:

“A few tech giants will spend over three hundred billion dollars this year, an amount equal to Portugal’s entire economy, on this one pursuit.”

Yet OpenAI, the current market leader, expects to lose $5 billion this year, and its annual losses are expected to swell to $11 billion by 2026.

The returns are already dismal. Reports indicate that Amazon is getting back maybe twenty cents for every dollar it pours into AI development. Meanwhile, the cost to train the next frontier model is ballooning past a billion dollars.

Enterprise adoption is failing spectacularly. One employee at a major B2B SaaS company going “all in on AI” reports:

“Our AI products aren’t being adopted at the rate they need to be, and the cost of the AI ‘agents’ are considerable… we charge by each AI agent action, so every email summary, every answer to a prompt costs money… it’s extremely expensive to adopt.”

The same employee noted that their AI “can’t consistently provide answers without random hallucinations” and described it as “a dead on arrival product that no one is asking for.”

The training costs alone are crushing companies. The same insider revealed:

“Every single employee of the company has to do these multi-hour trainings each quarter and we are required to pass certification tests every quarter which is such an incredible time sink and expense.”

Startups Wrap Others’ Intelligence

The dirty secret of Silicon Valley’s AI boom is that most startups aren’t building AI; they’re building interfaces to other companies’ AI.

Jasper.ai is the poster child for wrapper company problems. The marketing AI tool initially achieved rapid growth by wrapping GPT-3 in a user-friendly interface, reaching ~$42M in ARR in its first year.

But when OpenAI released ChatGPT with better capabilities, Jasper got “steamrolled” — to use Sam Altman’s own words. Companies that built basic wrappers on GPT-3 were crushed when ChatGPT 3.5 was released because they weren’t adding value.

The wrapper economy is surprisingly lucrative, for now. Examples include:

PhotoAI is generating $77K monthly recurring revenue (MRR)
Chatbase at $70K MRR
InteriorAI at $53K MRR
PDFai at around $30K MRR
Copy.ai and dozens of similar content generation tools

But the fundamental business model is broken. ChatGPT wrapper startups build on OpenAI’s technology with custom features and interfaces, but face catastrophic risk when the underlying provider changes terms or adds competing features. The ease of building ChatGPT wrappers has led to a saturated market where many startups offer similar functionality.

The market saturation is already brutal. No fewer than four companies developing AI coding assistants — Augment, Magic, Codeium, and Poolside closed rounds exceeding $100 million last year. All four use the same underlying language models from OpenAI or Anthropic, differentiated mainly by their user interfaces.

The funding numbers for identical businesses are insane. In 2024, there were 49 startups that raised funding rounds worth $100 million or more, with seven companies raising rounds that were $1 billion or larger. To even crack the list of largest AI funding rounds, a company had to raise more than a billion dollars in a single shot.

Five Companies Hoard Innovation

Real AI innovation is concentrated among a handful of players: OpenAI, Anthropic, Google, Meta(Maybe?), and Microsoft. OpenAI, Anthropic, Google, and Meta are racing to develop the best generative AI, while everyone else is essentially reselling their technology.

Even these leaders are struggling. OpenAI was on the cusp of a milestone. The company finished an initial round of training in September 2024 for a massive new artificial intelligence model that it hoped would significantly surpass prior versions of the technology behind ChatGPT and move closer to its goal of powerful AI that outperforms humans. But the model, known internally as Orion, did not hit the company’s desired performance.

The technical progress is hitting diminishing returns. As Diehl notes:

“The jump from GPT-2 to GPT-3 was a revolution; the jump from GPT-4 to its successors feels like a minor revision, yet it costs orders of magnitude more to achieve.”

Companies are spending exponentially more in energy, data, and hardware for improvements you can barely notice.

The technology is hitting fundamental limits. In a Bloomberg report, OpenAI, Google, and Anthropic have encountered critical issues, including a lack of high-quality training data and sufficient resources to develop advanced AI models. The market, however, is pricing these companies as if revolutionary breakthroughs are guaranteed every quarter.

Even more concerning, AI development is running out of fuel. University of Chicago researchers note that AI systems are literally running out of data to scrape, and humans aren’t producing content fast enough for the next big breakthrough. As one researcher put it:

“It’s literally running out of data to scrape and humans aren’t producing enough for the next big breakthrough.”

Companies Fail, Investors Pour

The failures are already starting, and they’re accelerating. In 2024, 966 startups shut down, compared to 769 in 2023, a 25.6% increase. AngelList found that 2024 saw 364 startup wind-downs, compared to 233 in 2023, a 56.2% jump.

Even high-profile AI startups with famous founders are collapsing:

Artifact, the AI-powered news app from Instagram’s co-founders Kevin Systrom and Mike Krieger, shut down in January 2024 after just one year. Despite being named “everyday essential app of the year” by Google Play Store, Systrom admitted:

“We have concluded that the market opportunity isn’t big enough to warrant continued investment.”

The founders tried everything from pivoting from a news reader to a social network to a discovery platform, but couldn’t find product-market fit.

Builder.ai, a $1.5 billion AI company backed by Microsoft, recently shuttered operations after being exposed as a fraud. The company promised to make software creation “as easy as ordering pizza” with its AI assistant Natasha, but it turns out their “neural network” was actually 700 engineers in India. The company raised more than $450 million before entering bankruptcy proceedings.

Ghost Autonomy, an AI startup that raised $238.8 million and filed 49 patents for autonomous driving, shut down in April 2024. Despite securing funding from OpenAI, the company failed to establish a path to profitability and couldn’t gain industry acceptance.

Over 90% of AI startups fail within the first five years, and many never make it past their first product launch. As one industry insider noted about a failed AI startup called NeuralGrid:

“ The product barely worked beyond demo mode. Burn rate: $680K/month. By mid-2024, they had 3 enterprise clients, two of which churned due to accuracy issues.”

Even worse, AI hardware companies are burning through hundreds of millions with zero results:

Humane AI Pin raised $230 million to create a $699 wearable AI assistant that would replace smartphones. The device was called “The Worst Product I’ve Ever Reviewed” by influential tech YouTuber Marques Brownlee. It suffered from constant overheating, couldn’t complete basic tasks like setting a timer, required a separate $24/month subscription, and had a charging case that was recalled due to fire hazards. The company predicted 100,000 sales in year one but achieved just 10,000. HP eventually acquired Humane’s assets for $116 million, far less than the $230 million raised, and discontinued the product entirely.

Rabbit R1 promised to be the “$200 iPhone of AI” that could handle complex tasks through voice commands. After 100,000+ pre-orders based on flashy demos, the reality was devastating. The device failed to deliver on most promises, had terrible performance, and only 5,000 of the 100,000 registered users actually used it five months after launch. CEO Jesse Lyu admitted the device was “rushed into the market” and couldn’t compete with Apple or Google.

Friend AI pendant shows all the signs of the next spectacular failure. Created by 22-year-old Avi Schiffmann, the startup spent $1.8 million of its $2.5 million in funding just to buy the domain “friend.com” — that’s 72% of their entire budget on a website address. The $99 “always-listening” AI necklace promises to combat loneliness by texting users throughout the day, but has already faced multiple red flags: shipments delayed from Q1 to Q3 2025, a pivot away from their original $600 “Tab” product after taking $100,000 in pre-orders, and a viral launch video that was widely mocked as “horror movie,” “depressing,” and “disgusting.” The founder even compared talking to his AI pendant to “talking to God”, a level of delusion that’s becoming typical in AI hardware startups.

Overpromise revolutionary capabilities, raise massive funding, ship broken products, and watch adoption collapse. As a tech industry observer has noted, “The Rabbit R1 and the Humane AI Pin were tech fever dreams that missed the point. They failed to solve problems that people truly care about.”

IPOs Vanish, Valuations Soar

While funding reaches record highs, exits tell a dramatically different story. Most exits have been bolt-on acquisitions, with CoreWeave’s IPO at the end of the first quarter the rare exception.

The IPO market remains effectively closed to AI companies. Nine such venture-backed companies went public above a billion dollars in 2024 — with ServiceTitan being the most recent — compared to 10, including four SPACs, in 2023. That’s barely any increase despite the massive funding surge.

Even successful AI companies are struggling to find exits. Artifact, despite having a core group of users who loved the product and winning awards from Google, couldn’t attract buyers at a meaningful valuation. Yahoo eventually acquired the technology in a fire sale, but the company itself was essentially worthless.

PitchBook’s senior research analyst Dimitri Zabelin put it bluntly:

“The dominant exit trend right now is frequent but lower-value acquisitions and fewer IPOs with significantly higher value”.

VCs Create Market Distortion

The numbers reveal dangerous market concentration. 33 percent of VC portfolios are committed to AI, while almost two-thirds of all U.S. venture funding went to AI in the first half of 2025, up from 49% last year. This level of sector concentration hasn’t been seen since the dot-com bubble.

The venture capital strategy has become “aggressive funding regardless of profitability.” In 2024, the investment strategy was heavily characterized by capitalizing on hype around AI technology, leading to substantial valuations and rapid deal cycles. VCs were eager to back groundbreaking technologies regardless of immediate profitability, leading to inflated valuations.

Even more alarming: funding for generative AI startups soared to $22.4 billion in 2023, and investors expanded into infrastructure and application-layer AI with a forecasted $175 billion in AI spending by 2024. When two-thirds of all venture capital flows into a single sector built largely on repackaging the same underlying technology, the parallels to 1999 become impossible to ignore.

How Bubbles Explode

The Internet Bubble (1995–2001).

When.com became.gone. The dot-com bubble followed a predictable pattern. Investors threw money at any company with “.com” in its name, believing the internet would change everything overnight. Companies like Pets.com raised $82.5 million to sell dog food online, while Webvan burned through $1.2 billion trying to deliver groceries.

The crash was swift and brutal. When investors finally asked, “Where are the profits?” most companies had no answer. The NASDAQ fell 78% from its peak. Hundreds of internet companies vanished overnight.

Only companies with real business models, like Amazon and Google, survived and eventually thrived.

Then the Crypto Bubble (2017–2022).

Crypto followed a similar playbook. During the 2021 peak, people mortgaged houses to buy digital currencies they couldn’t even pronounce. Every startup became a “blockchain company.” NFTs of cartoon monkeys sold for hundreds of thousands of dollars.

Reality hit hard. Bitcoin crashed from $69,000 to $15,500 — a 77% drop. Ethereum fell 84% from its peak. Entire crypto empires like FTX collapsed in days. The total crypto market lost over $2 trillion in value. Millions of retail investors lost their life savings chasing promises of getting rich quickly.

AI’s Structural Problems

The AI bubble has structural problems that make it more dangerous than both previous bubbles:

1. Dependency Risk: If OpenAI changes its pricing model or terms of service, startups may find their business models unworkable. Unlike crypto projects that at least own their technology stack, AI wrappers are entirely dependent on external APIs.

2. Lack of Differentiation: The ease of building ChatGPT wrappers has led to a saturated market. Many startups offer similar functionality — text generation, content creation, or customer service automation. Without significant differentiation, these companies compete on price, leading to margin compression.

3. No Competitive Moats: GPT wrapper companies don’t have a competitive “moat” or advantage that stops competitors from stealing their market share. This reminds me of Perplexity, for instance.

The Reckoning Is Coming

Market observers are already sounding alarms. Silicon Valley has hyped AI as the next internet or iPhone, and has invested like it cannot afford to miss out on the next big tech revolution.

But Ed Zitron, AI critic, calls this part of the “rot economy,” where VCs overhype a series of digital technologies, like the blockchain, then cryptocurrencies, then NFTs, and then the metaverse, promising the limitless growth of the early internet companies.

The Wellington Partners 2025 outlook warns that the explosive growth of AI draws parallels with the dotcom bubble, and one key lesson from that period is the fallacy of the first-mover advantage. Not all of the early AI winners will be able to convert today’s initial hype and success into long-term dominance.

The Coming Crash

When the AI bubble bursts, the damage will be swift and brutal.

The crash scenario becomes clearer when you consider deals like Murati’s. If a six-month-old company with no product can raise $2 billion, what happens when reality hits and investors realize they’ve been funding just PowerPoint presentations and empty promises?

The immediate casualties will be the hundreds of wrapper companies with no real technology that will collapse overnight when funding dries up or foundational model providers change their terms. Companies like Thinking Machines Lab by Murati will see their valuations evaporate when the market demands actual products and profits.

Then we have a valuation implosion. Companies valued at billions based on revenue multiples of API reselling will see valuations collapse to near zero. Big Tech companies are the biggest spenders on AI. While VCs dropped approximately $200 billion into AI between 2021 and 2024, Big Tech is on pace to surpass that amount this year alone. When returns fail to materialize, the entire tech sector faces correction.

The bubble is already bigger than the dot-com. Torsten Sløk, chief economist at Apollo Global Management, warns AI stocks are even more over-valued than dot-com stocks were in 1999. In a research note, he illustrated how shares of Nvidia, Microsoft, and eight other companies are creating an even bigger bubble than we saw during the dot-com era. And that could cause serious market damage if and when it pops.

The stock market’s bull run is deeply dependent on the growth of the Big Tech companies, fueling the AI bubble. Of course. In 2023, 71 percent of the total gains in the S&P 500 were attributable to the “Magnificent Seven”, Apple, Nvidia, Tesla, Alphabet, Meta, Amazon, and Microsoft, all of which are among the biggest spenders on AI.

We Survive

Like the dot-com crash, some companies will survive and thrive. The handful of companies building real AI infrastructure, true foundational model creators, specialized chip manufacturers, and companies with genuine AI-driven business model innovations will emerge stronger.

But the carnage among the pretenders will be unprecedented. When investors realize that paying $50 million for a company that’s essentially a fancy ChatGPT interface makes no economic sense, the correction will be swift and merciless.

The AI bubble has already exceeded crypto’s peak by more than 300%. Unlike crypto, which at least had the excuse of creating new financial primitives, most AI startups are building glorified chatbots using someone else’s intelligence.

Last published 4 hours ago

Behind the screens. A publication on Medium about technology, culture, and humans.

1M+ reads. Ex-Forbes. 2x YC designer. Top Writer in Technology, Innovation, and Design. 📩 For collaborations: ddotangelo@icloud.com

Responses (6)

Talbot Stevens

What are your thoughts?

Like always, thanks everyone for the clappers!

You also need to correct your gender biases, it weakens your argument substantially. Murati is far from a nobody or "resume", she was literally the CTO of OpenAI for a long time. Respect where respect is due, bro.

Great article, you explained it crystal clear. In theory, economic bubbles are hard (if not impossible) to predict, but this case is so obvious that sometimes is hilarious listening to grown up men and women talking about what AI will acomplish in…

Recommended from Medium

[

See more recommendations

](https://medium.com/?source=post_page---read_next_recirc—3e61e81ba826---------------------------------------)