AI is no longer a sector. It’s the market.
When 4 in 10 venture dollars go into one theme, we’re no longer talking about a hot trend; we’re talking about a structural shift. New data from Carta, reported by TechCrunch, confirms what anyone fundraising or deploying capital has felt for the past 18 months: AI is eating the venture industry from the top down. Money is concentrating into a tiny set of AI champions, valuations are exploding and, on paper, returns look fantastic.
The real question is not whether this is a bubble. It’s whether this extreme concentration is an efficient reallocation of capital to a genuine platform shift—or the beginning of a systemic mispricing that will haunt LPs, founders and even GPU-rich incumbents.
This piece unpacks who’s winning, who’s getting squeezed out, and what this means for European and global tech.
The news in brief
According to TechCrunch, citing new data from equity platform Carta, AI startups accounted for 41% of the $128 billion raised by companies on Carta last year—a record share for any single category.
Despite this, the money is not broadly distributed. Around 10% of startups soaked up roughly half of all funding, with huge rounds going to companies like OpenAI, Anthropic and Elon Musk’s xAI. These firms raised multi‑billion‑dollar late‑stage rounds at towering valuations, with OpenAI’s latest financing reportedly bringing it close to a $1 trillion private valuation.
Carta’s data also shows that venture funds launched in 2023 and 2024—after the public debut of ChatGPT—are posting the highest internal rates of return compared with funds raised between 2017 and 2020, whose performance has trended down. Much of that apparent outperformance comes from fast markups: a seed‑stage AI investment that quickly raises a richer Series A looks like a big win on paper.
The result, as TechCrunch notes, is a K‑shaped venture market: capital and returns are concentrating at the top.
Why this matters: a winner-takes-most experiment in real time
The venture industry is running a massive, real‑time experiment in winner‑takes‑most economics.
Who benefits right now?
- Frontier AI labs and their early backers. OpenAI, Anthropic and xAI have turned access to compute and talent into unprecedented late‑stage rounds. Their early investors are sitting on enormous paper gains and can raise new funds with eye‑catching IRR figures.
- Infrastructure providers. Those billions don’t buy ping‑pong tables; they buy GPUs and cloud capacity. A large share of this capital is flowing almost directly to the likes of Nvidia and the big cloud platforms.
Who loses—at least for now?
- Non‑AI startups. When 41% of venture dollars on a major platform go into AI, everything from climate tech to digital health gets crowded out. Even strong non‑AI pitches are being told to “add an AI angle” or wait.
- Smaller funds and emerging managers. Mega‑rounds are often led by a small club of multibillion‑dollar funds with access to the hottest AI deals. Everyone else is left to fight over the leftovers or stretch beyond their usual stage to get into anything AI‑flavoured.
The immediate implication is a bifurcated market with rising systemic risk. Capital is clustering in a handful of companies whose business models are still evolving, whose unit economics depend on volatile compute prices, and whose regulatory risk is growing. If even one of these AI giants stumbles post‑IPO, it will reverberate through the entire VC stack—especially through those 2023–24 funds that currently look brilliant on paper.
The bigger picture: lessons from past manias—and what’s different
This isn’t the first time venture has gone all‑in on a dominant narrative. We’ve seen:
- Dot‑com infrastructure (late 1990s): Capital flooded into telecoms and hosting, on the assumption that traffic and valuations would grow forever.
- Social/mobile (2010s): Massive bets on network‑effect platforms where a few companies—Facebook, Google, Apple—captured most of the value.
- SaaS and fintech (2018–2021): Huge late‑stage rounds at revenue multiples that only made sense in a zero‑rate world.
AI combines elements of all three. Like dot‑com infrastructure, it is capital‑intensive and hardware‑dependent. Like social networks, it leans toward scale and data flywheels that favour a small number of giants. And like 2021 SaaS, it’s currently buoyed by FOMO‑driven markups rather than realized exits.
What’s different this time:
Compute is the new land grab. In previous cycles, capital built user bases or sales teams. In AI, much of the money is prepaid to infrastructure providers. That may mean less durable differentiation for the startups themselves: owning more GPUs today doesn’t guarantee a moat if models commoditize tomorrow.
Open‑source is stronger. From 2023 onward, open‑source models have rapidly improved, often matching proprietary systems on many tasks. If that trend continues, the value of closed‑model scale at all costs could be overstated—especially for downstream application startups.
Regulation is marching faster. AI safety, copyright, data protection and competition authorities are all circling. In contrast to the early days of the web or social media, governments are unlikely to let a handful of private labs shape the entire stack without oversight.
The AI funding surge fits into a broader trend: venture capital is becoming more concentrated, more thematic and more correlated with macro narratives. That amplifies upside in good times—but also deepens drawdowns when sentiment turns.
The European angle: opportunity, constraint and FOMO
For Europe, this concentration of AI capital is both a threat and a once‑in‑a‑generation opening.
On the threat side, the biggest AI megarounds are overwhelmingly US‑centric, with some notable exceptions. European founders face a harsher reality:
- GPU access is scarcer and more expensive if you’re not plugged into the US cloud‑VC complex.
- Fund sizes are smaller, which makes writing multibillion‑dollar checks effectively impossible for most European firms.
- The EU AI Act, GDPR and upcoming digital regulations raise compliance costs and uncertainty, particularly for frontier‑model work.
Yet Europe also has advantages that don’t show up in Carta datasets:
- Deep strength in industrial, scientific and B2B software, where AI can be embedded into manufacturing, healthcare, energy and logistics.
- A cultural and regulatory focus on trust, safety and privacy, which—if used strategically—can become a differentiator for “boringly reliable” AI in critical sectors.
- A growing cadre of AI‑native startups and labs building on top of open‑source models, rather than competing head‑on with trillion‑dollar US labs.
European LPs and policymakers should be less obsessed with “missing” the next OpenAI and more focused on owning the applied AI layer: the thousands of domain‑specific companies that will actually deploy these models into factories, clinics, power grids and public services.
If AI is a new general‑purpose technology, the real economic impact in Europe will come from diffusion, not from owning one or two global foundation models.
Looking ahead: what to watch in the next 24 months
Several fault lines will determine whether today’s AI venture boom matures into a durable paradigm—or ends as another cautionary case study in capital misallocation.
1. IPO outcomes for the AI giants.
TechCrunch notes that the major labs are teasing IPOs. If those offerings happen and trade well over several quarters, they will validate current paper markups and unlock more capital for AI. If they stumble—or if public markets heavily discount private valuations—the IRR of those celebrated 2023–24 funds could compress sharply.
2. Unit economics under real customers.
Many AI startups still derive a large chunk of revenue from other startups or from experimental budgets. As enterprise CIOs push for hard ROI and regulators tighten the rules, we’ll learn which AI products actually justify their compute bills—and which were subsidized experiments.
3. Compute and model commoditization.
If cloud providers and open‑source communities keep narrowing the gap with frontier labs, the edge may shift from owning the biggest model to owning the best distribution and workflows. In that world, today’s mega‑rounds look less like defensible moats and more like expensive prepayments for a temporary advantage.
4. Regulatory inflection points.
New AI rules in the EU, US and elsewhere could nudge value away from a few hyperscalers toward more diversified, sector‑specific players—or they could entrench incumbents who can afford compliance.
For founders and investors, the strategy should be clear: avoid chasing whatever just raised a $10‑billion‑plus round; instead, build or back companies that would still make sense if GPU prices halved and model quality doubled for everyone.
The bottom line
AI is not just attracting venture capital; it’s reshaping how the asset class works, concentrating power and risk in a handful of labs and their backers. The early IRR numbers look great, but they’re built on fragile markups, not exits. If you’re in the top decile of AI deals, this is a golden moment. For everyone else, it’s time to be contrarian: focus on durable, sector‑specific AI value rather than mega‑round spectacle.
The real test is still ahead: will this wave produce enduring infrastructure—or just very expensive training runs?
