The Concentration Trap: How AI Mega-Rounds Turned 'Diversified' VC Into a Single Bet

Global venture capital hit a record $510 billion in the first half of 2026. OpenAI and Anthropic alone absorbed roughly 43% of it, about $217 billion between two companies, according to Crunchbase's H1 2026 venture data. If that record number made you feel good about the diversification in your VC fund allocation, I have bad news. You did not sign up to bet your capital on two foundation-model labs. But if you are an LP in a generalist VC fund right now, that is functionally what happened to you.

I want to be direct about what I am arguing here. This is a thesis piece and I am not going to hide behind hedges. Venture capital in 2026 is no longer a diversified asset class. It is an AI infrastructure concentration bet wearing a diversified-fund costume. The math is not subtle. Once you see it, you cannot unsee it.

The data: five companies, three-quarters of the money

Start with the headline mega-rounds from this week. Kling AI, the video-generation unit of Kuaishou backed by Tencent, is closing in on a near-$3 billion raise at an $18 billion valuation, with General Atlantic among the investors, according to reporting that broke July 2 and 3, 2026. Together AI, the AI infrastructure and inference company, closed an $800 million Series C at an $8.3 billion valuation, with backers including Aramco Ventures. These are not seed rounds. These are single checks larger than most entire venture funds raise in a decade.

Now zoom out to the full first half of the year. According to the PitchBook-NVCA Venture Monitor for Q1 2026, the top five deals of the quarter, OpenAI, Anthropic, xAI, Waymo, and Databricks, captured approximately 73% of all venture deal value in the United States. Strip those five deals out of the data and total venture deal value falls by more than 73%. Read that again. Three out of every four dollars deployed by American venture capital in the first quarter of 2026 went to five companies, and every one of them is an AI or AI-infrastructure business.

The trend accelerated into the second quarter. AI's share of global VC dollars rose to roughly 70% for Q2, up from about 50% a year earlier, after touching 80% in Q1, per Crunchbase's tracking. This is not just a Silicon Valley talking point anymore. At the TechCrunch StrictlyVC panel in Athens in May 2026, three senior venture investors said the concentration out loud, on stage, without hedging. Niko Bonatsos of Verdict Capital, Andreas Stavropoulos of Threshold Ventures, and Ben Blume of Atomico put a number on it, citing roughly 75% of all 2026 venture capital going to just five AI companies. They called it unprecedented, and flagged something you should sit with. The panel cited widespread ARR (annual recurring revenue) inflation among AI startups, meaning the revenue numbers propping up these valuations may not be as real as the pitch decks suggest.

Here is the table that should be on the wall of every LP committee room:

Data Point	Figure	Source
Global VC funding, H1 2026	~$510 billion (record)	Crunchbase News
OpenAI + Anthropic share of H1 2026 VC dollars	~43% (~$217B)	Crunchbase News
AI share of global VC dollars, Q2 2026	~70% (up from ~50% YoY)	Crunchbase
AI share of global VC dollars, Q1 2026	~80%	Crunchbase
Top 5 deals' share of Q1 2026 US VC deal value	~73%	PitchBook-NVCA Venture Monitor
Kling AI raise (announced July 2-3, 2026)	~$3B at $18B valuation	Press reporting
Together AI Series C	$800M at $8.3B valuation	Company/press reporting
LP capital to seasoned firms, Q1 2026	91% (up from 74% in 2025)	PitchBook data
Emerging manager fundraising, YoY change	Down 35%, to $12B (lowest since 2020)	PitchBook data

The last two rows matter as much as the mega-round headlines, and they get less attention. Money is not just concentrating by company. It is concentrating by manager. Institutional LPs put 91% of new Q1 2026 commitments into seasoned, brand-name VC firms, up from 74% just a year earlier, per NVCA-affiliated PitchBook data. Emerging managers, the newer funds most likely to write early checks into the next generation of non-AI, non-mega-cap companies, saw fundraising drop 35% year over year to just $12 billion, the lowest level since 2020. When you starve emerging managers, you starve the pipeline of everything that is not already famous.

Jeff's thesis: you think you own a fund, you actually own a basket of five bets

Here is my argument, stated plainly. If you are a limited partner writing checks into a generalist venture fund in 2026, the kind of fund that markets itself as diversified across sectors, stages, and geographies, you need to ask your general partner one question. How much of this fund's actual deployed capital, directly or through co-investment and follow-on exposure to the broader market, is riding on the fortunes of OpenAI, Anthropic, xAI, or a small handful of AI infrastructure names like Kling AI and Together AI?

Diversification is supposed to mean your outcome does not hinge on any single bet. But when 70% to 80% of an entire asset class's dollars flow into one theme, and three-quarters of deal value sits in five names, the market itself has stopped being diversified. You cannot out-diversify a market that has concentrated this hard. Every fund benchmarking itself against that market, every fund-of-funds spreading LP dollars across "20 great managers," and every secondary buyer pricing off comparable transactions is implicitly pricing off the same five to seven companies. The diversification is a mirage sitting on top of genuine concentration.

This is not a claim that AI is overhyped as a technology, and I am not making a call on whether OpenAI or Anthropic succeed. My argument is narrower and, I think, harder to dismiss. The venture capital asset class, as currently constructed, has become a leveraged proxy for a handful of AI infrastructure outcomes, and most LPs have not repriced their risk models to reflect that. You can read more of AIN's ongoing coverage of these dynamics in AIN's venture capital coverage.

The historical parallel: telecom, 1996-2001, and the LPs who got wiped out

I have seen this movie before, and so have you if you were paying attention between 1996 and 2001. The telecom infrastructure buildout of the late 1990s ran on a logic close to what you see in AI infrastructure today. A small number of "sure thing" categories, fiber optic networks, CLECs (competitive local exchange carriers), and undersea cable systems, absorbed a wildly disproportionate share of both venture and public capital, justified by demand projections that assumed exponential internet traffic growth forever.

The projections were not entirely wrong. Internet traffic did grow enormously. What broke the model was the gap between the pace of capital deployment and the pace of revenue realization. Companies like Global Crossing and WorldCom, along with a long list of CLECs, raised and spent tens of billions of dollars laying fiber ahead of demand, financed heavily by vendor debt and venture dollars that assumed the growth rate of that moment would hold for a decade. When growth normalized and financing costs came due, the overcapacity became a liability instead of an asset. Global Crossing alone had raised more than $20 billion before its 2002 bankruptcy. WorldCom's collapse remains one of the largest bankruptcies in U.S. history.

The LPs who got hurt worst were not the ones who had a telecom position. They were the ones who thought they had a diversified technology or growth portfolio and discovered, only after the correction, how much of their "diversified" exposure ran through the same infrastructure buildout theme. University endowments and pension funds that had allocated to "innovative growth" strategies across multiple managers found those managers had crowded into the same category. Yale's endowment and Harvard Management Company both spent the 2000s rebuilding allocation models after that cycle exposed how much manager overlap had been hiding inside supposedly uncorrelated mandates. Sound familiar?

I am not predicting AI infrastructure collapses the way telecom did. The demand signal for AI compute and inference today looks more durable than 1999-era bandwidth demand did, and the revenue growth at OpenAI and Anthropic is real, even if some of the ARR figures floating around younger AI startups deserve the skepticism Bonatsos and his co-panelists raised. What I am saying is that the mechanism of risk is the same one that burned sophisticated LPs before: capital concentrating into infrastructure buildout faster than anyone can verify the demand will hold at that scale.

The steelman, and why I am still cautious

Let me make the bull case as strongly as I can, because it deserves a fair hearing. AI infrastructure spending in 2026 is backed by real, measurable enterprise adoption, not speculative page-view metrics. OpenAI and Anthropic both report substantial and growing revenue, not just usage. Unlike dot-com-era telecom, where demand was a forecast, AI compute demand from enterprises, governments, and consumer products is observable today, in the form of API bills, GPU utilization rates, and inference costs that companies are paying right now, not promising to pay in five years. Firms like Andreessen Horowitz and Thrive Capital, two of the most active investors in this cycle, would argue the concentration reflects genuine winner-take-most dynamics in foundation models, where scale itself is the moat, and that spreading capital thin across dozens of also-ran model companies would be the real mistake.

That argument is not crazy. Scale economics in frontier AI are real. But here is why I remain cautious anyway. First, valuation and revenue quality are two different things, and the panel warning about ARR inflation should worry you regardless of which side of the AI-optimism debate you sit on. Inflated recurring revenue figures make mega-round valuations look safer than they are. Second, concentration risk does not require the AI thesis to be wrong to hurt LPs. It only requires a repricing, a slower-than-expected path to profitability, or a single disappointing model release from one of the five to seven companies absorbing most of the capital. Third, the emerging-manager funding collapse, down 35% year over year, tells you the venture business's ability to originate the next non-AI breakout, the next Airbnb, the next Stripe, the businesses that don't need a GPU cluster to exist, is being quietly defunded while everyone stares at the mega-round headlines. Rick Zullo of Equal Ventures and Nisha Dua of BBG Ventures, both active in the emerging-manager and non-AI seed space, have flagged this starvation dynamic as a structural problem for venture returns five to seven years out, not just a 2026 curiosity.

What accredited investors should actually do about this

You do not need to avoid AI exposure. You need to know exactly how much of it you already have and decide if that is the exposure you meant to take.

• Ask your GP for a look-through, not a label. "Diversified growth fund" is a label. Ask what percentage of deployed capital and follow-on reserves sits in AI infrastructure and foundation-model-adjacent names, directly or through co-investment rights.
• Separate infrastructure exposure from application exposure. A company using AI to solve a vertical problem in healthcare billing or logistics carries different risk than a company competing to build or serve the underlying models. Know which one you are funding.
• Underwrite revenue quality, not just revenue growth. Given the ARR inflation concerns senior VCs raised publicly this year, ask managers how they verify recurring revenue claims in AI portfolio companies before marking them up.
• Deliberately seek emerging-manager and non-AI allocations if you want real diversification. With emerging managers down 35% year over year in fundraising, the capital still flowing there is arguably getting better access and better terms precisely because everyone else stopped looking.
• Size your AI infrastructure bet like the concentrated bet it is. If you decide the AI mega-round thesis is worth backing, back it deliberately and size it as a concentrated position, not as the "safe, diversified" sleeve of your portfolio.

None of this means sitting out. It means going in with eyes open instead of discovering the concentration after the next repricing, the way plenty of telecom-era LPs discovered theirs only once the bankruptcies started. For more on how AIN thinks about sizing exposure across a venture-heavy portfolio, see our market analysis coverage, which tracks these allocation shifts quarter over quarter.

The $510 billion H1 2026 number is real, and it is a record. But a record built on 70% to 80% AI concentration and three-quarters of deal value in five names is not evidence that venture capital is thriving as a diversified asset class. It is evidence that venture capital has temporarily become an AI infrastructure fund with a diversified-sounding name on the door. You can decide that is a bet worth making. Just make sure you are making it on purpose.

Author Disclosure: Jeff Barnes, MBA has no personal position in any company, fund, or platform named in this article. Angel Investors Network has no current commercial relationship with any party mentioned. AIN provides marketing and education services, not investment advice. Past performance does not guarantee future results. All investments involve risk, including loss of principal.