$510 Billion Invested in Startups: Record Hides a Bubble

In an unprecedented surge, global investments in startups reached $510 billion in the first half of 2026. While this marks an all-time high, the underlying story is far more complex. The bulk of this capital is overwhelmingly concentrated in AI, leaving other sectors to wither. Are we witnessing a bubble akin to the dot-com bust of 2000 or the crash of 2022? In my experience, all signs point to yes. Despite the headlines celebrating this era of venture capital optimism, this ecosystem is less a technological revolution and more a bubble in disguise.

Photo: Precondo CA on Unsplash

The prevailing narrative suggests that generative AI has kicked off a new cycle of massive investment. After years of drought, funds are flush with cash, and we're supposedly witnessing a transformation akin to the advent of the internet. But beware: when you break down those astronomical figures, patterns emerge that should alarm any entrepreneur not developing an LLM or selling GPU infrastructure.

The Anatomy of a Misleading Record

Of those $510 billion, a staggering 68%—or $347 billion—was funneled into startups with "AI" in their pitch or related to infrastructure. This concentration is the most severe ever seen in any tech category over a similar period. During the dot-com boom of 2000, pure internet companies only captured 54% of the capital, and during the fintech boom of 2021, financial startups snagged 42%. We're now 14 percentage points above the previous historical high.

Geographically, the numbers are equally revealing. The U.S. attracted $198 billion (39%), followed by China with $112 billion (22%), and Europe with $89 billion (17%). However, in the U.S., 73% of the capital went to Bay Area startups, mainly less than 120 companies. Anthropic raised $7.2 billion in its Series D, Cohere closed with $4.8 billion, and Paris-based Mistral AI secured $3.1 billion just 18 months after founding. Mega-rounds—those above $500 million—now account for 31% of total capital, when historically they never surpassed 18%.

Meanwhile, the rest of the ecosystem is bleeding out. Seed rounds outside the AI sector dropped 47% in volume compared to the first half of 2025. Series A in sectors like enterprise SaaS, non-AI edtech, or healthtech without automated diagnostics fell 39%. Early Stage funds are completely reorienting their theses towards AI or shutting down. Index Ventures, one of Europe's most respected, allocated 80% of its new $2.9 billion fund exclusively to companies with a "core machine learning component." Lightspeed Venture Partners went further: they closed two of their four regional funds, consolidating capital into a single vehicle focused on "robust AI companies."

The Problem of Unrealistic Valuations

Photo: K C on Unsplash

Current valuations defy all logic. Anthropic is valued at $42 billion with projected annual revenues of $1.8 billion by year-end, meaning a 23x multiple on revenue. Cohere, valued at $19 billion, generates approximately $340 million in ARR, with a multiple of 56x. For comparison, Salesforce at its 2021 peak traded at 14x. Stripe, infamous for being overvalued, never exceeded 18x in private rounds.

Mistral AI is an extreme case. Founded in May 2024 by former DeepMind and Meta researchers, it raised $3.1 billion in February 2026, with a post-money valuation of $12.4 billion. Its revenue then: $78 million annualized, a multiple of 159x. Not even during the height of Pets.com or Webvan did we see ratios so disconnected from operational reality. VCs justify this by claiming they are valuing the potential market TAM of AI, not current metrics. It’s like valuing Amazon in 1998 on the assumption that all retail would eventually be online.

This approach has direct consequences for any startup seeking capital in 2026 or 2027. Investment terms have become brutal: multiple liquidation preferences (2x, 2.5x, even 3x in some cases), full ratchet anti-dilution, and pay-to-play provisions forcing founders to participate in subsequent rounds or face massive dilution. We’ve seen Series B term sheets with 2.2x liquidation preference and full ratchet protection for companies with real traction. Honestly, three years ago, these terms were exclusive to financially distressed situations. Now they’re standard in any deal above $100 million.

The Secondary Market Is Already Bleeding

While some celebrate in the primary markets, the secondary market tells a different story. Here, employees and early investors sell their shares to specialized funds. In May 2026, the average discount in secondary transactions of AI startups reached 38% compared to the last primary valuation. This means if Anthropic is officially worth $42 billion, secondary buyers are paying as if it were worth $26 billion. This disparity is clear: those with inside information—employees with vested stock options, early angels—are cashing out at any price.

Forge Global, the largest secondary marketplace, reported a 340% increase in the volume of AI startup equity transactions in the first half of 2026 compared to the same period in 2025. However, 71% of those trades were executed with discounts exceeding 30%. EquityZen, another major player, noted that 84% of sell orders on its platform came from employees of just 23 companies, all AI unicorns. When those with direct access to internal numbers are selling aggressively, something doesn’t add up with the public narrative of exponential growth.

CartaX, Carta’s secondary platform, temporarily suspended transactions in April for eight AI unicorns after detecting "unusual selling pressure patterns." While they didn’t confirm names, multiple sources point to at least two being Anthropic and Cohere. The pattern is always the same: senior employees selling between 60% and 80% of their vested positions, something statistically unusual unless they have information that the current valuation is unsustainable.

The Architecture Behind the Money: Who's Funding the $510 Billion

The capital composition has radically changed. Traditional VC funds like Sequoia, Andreessen Horowitz, Accel, and Benchmark represent only 34% of the total invested. The rest comes from non-traditional sources that historically didn’t participate in venture at this scale: sovereign funds (18%), BigTech corporate venture arms (23%), ultra-wealthy family offices (14%), and pension funds entering directly without going through funds of funds (11%).

What surprises me most is how Saudi Arabia's Public Investment Fund has directly invested $8.4 billion in eight AI startups, the largest single-country sovereign fund allocation to venture in history. Abu Dhabi Investment Authority (ADIA) invested $6.1 billion and Singapore's GIC added $4.7 billion. These players, with no tech experience, do not conduct deep operational due diligence and are driven mainly by institutional FOMO. When PIF justifies a $1.2 billion investment in a startup generating $40 million ARR by saying they want "exposure to AI TAM before it closes," we’re dealing with pure speculative capital, not grounded investment.

Microsoft, Google, Amazon, and Meta have collectively invested $41 billion through their corporate venture arms. But these "investments" are complex hybrids: part equity, part cloud credits, part commercial agreements disguised as investment. Microsoft invested $10 billion in OpenAI, but most are Azure credits that OpenAI must spend on Microsoft infrastructure. Google put $3.8 billion into Anthropic, with $2.4 billion in GCP credits. These aren't pure venture bets; they are strategic moves to secure enterprise customers in their cloud and prevent migration to competitors. Consequently, the real cash capital is significantly lower.

The 2022 Pattern Is Repeating Exactly

Entrepreneur friends, remember the first half of 2022? Then, global investment hit $347 billion, a record at the time. Multiples were sky-high: fintech valued at 32x revenue, crypto at 87x, direct-to-consumer at 19x. There was talk of a "new paradigm" and that "fundamentals had changed." But six months later, the market collapsed. Klarna went from $45.6 billion to $6.7 billion, Stripe dropped from $95 billion to $50 billion, Instacart canceled its IPO and cut its internal valuation from $39 billion to $10 billion. Layoffs at VC-backed startups reached 165,000 people in just the second half of 2022.

Today, the patterns are identical: record valuations, extreme concentration in one category, non-traditional funds dominating volume, aggressive terms, and a secondary market with massive discounts. The only variable that changed is the sector: in 2022 it was fintech and crypto; in 2026 it’s generative AI. But the bubble mechanics are identical.

The critical difference is interest rates. In 2022, the Fed raised them from 0.25% to 4.5% in nine months, eliminating cheap capital. In 2026, rates are at 3.75%, and the Fed has indicated they could drop to 3.25% by year-end. That means dry powder—committed but uninvested capital—remains massive: $2.1 trillion according to Preqin, the highest number ever recorded. As long as that capital seeks returns and rates stay reasonable, the music will keep playing. But when it stops—and it always does—the adjustment will be more violent because multiples are more inflated.

What to Do if You're a Founder in 2026

If you’re seeking capital in this environment, you need clear strategic decisions. First: if you don’t have a defensible AI component in your core value proposition, prepare for an extremely difficult Series A. Funds that used to do 25 deals a year across diverse sectors now do 8, and seven are AI. You have two options: pivot your narrative to include machine learning or turn to micro funds that still believe in diversified theses—they exist, but they are few and write smaller checks.

Second: if you’re in AI and raise a large round with an inflated valuation, assume it’s the last easy round. Design your burn rate for a minimum 48-month runway, not the 18-24 that was standard. When the market turns, down rounds and inside rounds will be the norm. If you don’t have enough cash to reach profitability or at least default alive without raising more, you’re dead when the cycle changes.

Third: negotiate investment terms assuming your next round will be lower. Avoid full ratchet anti-dilution at all costs—it’s pure poison for your cap table if you raise down. If VCs insist, negotiate narrow-based weighted average at most. Resist multiple liquidation preferences. If you accept 2x, you’re saying your investors recover double their money before you see a dime in an exit. In a downturn scenario, that means a sale of your company at 3x the current valuation could leave you with nothing.

Fourth: diversify your investor base towards corporate strategics that could be buyers in an M&A. If Google, Microsoft, or Salesforce are in your cap table, you have a natural buyer when the market turns ugly and you need to sell. Purely financial VCs will want a 10x exit; a strategic might be happy with 3x if your technology fits into their roadmap.

The Question No One Is Asking

Here’s the uncomfortable topic no one on Sand Hill Road wants to discuss: how many of these billion-dollar AI startups really have a defensible advantage five years out? OpenAI has a brand advantage and consumer distribution. Anthropic has Constitutional AI and a security narrative that resonates with enterprise. But Cohere? Mistral? The other 73 startups doing LLMs or fine-tuning or embeddings? When open source models like Llama 4 and Falcon reach capability parity with GPT-4—something Meta promised for Q4 2026—what justifies valuations in the tens of billions?

The honest answer is: nothing, except the expectation that someone else—Google, Microsoft, Amazon, or a retail investor IPO—will pay more in the future. That’s the greater fool theory applied to venture capital. And it works until the fools run out.

Are you fundraising in 2026, or waiting for the bubble to burst and raise in the ashes of 2027? Does your startup have enough runway to survive the adjustment that statistically comes every 4-5 years?