All news

Venture capital in 2026: record money, fewer hands, smaller teams

Thin grooves in a pale stone slab run to the edge of a dark graphite-blue pool with one cobalt reflection

2026 brought more venture money than any year before, and most of it went to a few dozen companies. Here is what that means for AI-native founders.

I'm Max Levoshin, founder of Fanaura, a private family office and group of companies. We invest our own capital and the capital of our LPs, usually through a separate SPV for each deal, and most of our first checks go into AI-native companies at pre-seed and seed. Every investment decision we make passes through me, so I read the venture market the way a founder reads a cap table: where the money sits, who controls it and what it costs.

The third quarter of 2026 is now fully reported, and yesterday Anthropic, one of the two largest AI labs, made its listing documents public. It is a good moment to step back. Below is what I see in the numbers, in the latest accelerator batches and in the theses of the largest investors, where capital is going, and what it means if you are building an AI-native company today.

The biggest year on record, in very few hands

In the first nine months of 2026, startups worldwide raised roughly $700 billion. In the United States the figure reached $515.8 billion, about 44% above the full-year record of 2021. The first and second quarters of this year are the two largest quarters venture has ever had.

The third quarter cooled. Global funding came in at about $160 billion, a quarter below the second quarter and still about half above the same quarter of 2025. Rates did part of that work: the 10-year Treasury spent the autumn near 5%, close to its highest level in almost two decades, and a 5% risk-free rate resets every hurdle rate a venture investor works with.

Under the totals the picture is sharper. AI took 82.7% of every venture dollar invested in the US this year. Deals of $100 million and more were about 5% of US transactions in the third quarter and 75% of the money. More companies raised rounds of $1 billion or more in that single quarter than in any quarter before. Venture in 2026 is a market of a few dozen companies with balance sheets the size of national budgets, plus a long tail of everyone else.

The door at the bottom is wide open

The part most commentary misses: the long tail is growing. More US companies received their first institutional round in the first nine months of 2026 than in all of 2021, about 8,600 of them. AI accounts for roughly 45% of US deals, far below its share of dollars. Most AI-native companies funded today are small.

Entry prices are rising while checks are not. The median US seed round this year is $3.0 million, the only stage where the typical round got smaller, at a median pre-money valuation of $18.3 million. For AI companies the median seed is $4.5 million at $21 million pre-money. At Series A the median pre-money is $60.3 million across the market and $78 million for AI. Between rounds, AI companies step up 2.2 times in valuation against 1.5 times for everyone else, the widest gap since ChatGPT launched.

The message is simple. You need less money than five years ago to prove a lot more, and investors pay for the proof. A team of five with agents in every workflow now ships what a team of thirty shipped in 2021.

Fewer funds, bigger funds

The money behind seed rounds comes from venture funds, and that source is thinning. US venture firms raised $108.5 billion in new funds over nine months, already more than in all of 2025. Funds of $500 million and above took 78% of it while making up 6% of the funds closed. In the first half of the year, three firms raised almost half of all new US venture capital.

At the other end, only 81 first-time funds closed in the US this year, the fewest in more than a decade. The number of emerging managers closing any fund at all fell from 927 in 2022 to 211 this year, and the median US fund is now $10 million. Since 2022, limited partners have put about $202 billion more into venture than they got back, and they answer by backing the largest names.

In practice, the firms with the most money write the largest checks, and the largest checks do not go into a $3 million seed. The space between a friends-and-family round and a large Series A is served by fewer specialist funds than at any point in the last ten years. Family offices, operator angels and corporate investors are filling it. That is where we work.

What gets funded now

The theses the largest investors published this year share one idea: sell the work, not the tool. Businesses spend about six dollars on services for every dollar they spend on software, and AI-native companies now go after the larger budget. Some of the biggest firms have built vehicles around the idea. One raised about $2 billion for a holding company that buys accounting and IT firms and rebuilds them around agents, and Anthropic set up a services venture of about $1.5 billion with financial partners. Founders who start from tasks their customers already pay someone else to do have the shortest path to revenue.

The second shift happened over the summer: AI moved into the physical world. In the winter and spring batches of the best-known accelerator, about one company in eight built robots, drones, chips, factories or energy systems. In the summer and fall batches it was close to one in four, and most of the standouts of the September demo day made hardware. Robotics startups raised $9.9 billion in the third quarter, more than twice the figure of a year earlier.

The third is infrastructure for agents: identity, payments, memory, security and sandboxes for software that acts on its own. In the latest cohort of one large firm's accelerator, developer tools and AI infrastructure make up 28% of the companies.

The traction bar has risen with all of this. At the March demo day of the best-known accelerator, 14 companies already had more than $1 million in annual recurring revenue, and the batch grew revenue by 14% a week on average. AI-native companies with $1 million to $20 million in revenue grow about 256% a year, against 90% for software companies that added AI to an existing product. For a Series A, a B2B AI company now needs $2 million to $4 million in ARR, net revenue retention above 120% and a gross margin moving toward 60%.

Teams are smaller than ever. The median team in the latest accelerator batches is two people, and about a third of new startups have a single founder. Revenue per employee is now one of the first numbers an investor asks for.

What investors pass on

The turn came in February. After a run of launches from the frontier labs, software and IT services stocks lost close to $1 trillion in value at the worst point of the sell-off, and private investors drew their own conclusions. Per-seat pricing is the first casualty, because seats lose meaning when a few agents do the work of a department. Thin interfaces over a single third-party model, clones of CRM systems and task trackers, generic productivity tools and integration connectors that open protocols turn into commodities follow close behind. When labs ship plugins, thin software layers lose their premium. The question in every first meeting is whether the next model release makes the product better or makes it unnecessary.

Revenue quality is checked line by line. Free pilots, contracts that are signed but not live and annual run rates extrapolated from one strong week no longer pass as ARR. Investors also check whether AI sits at the core of the product: of roughly 9,000 venture-backed companies that call themselves AI companies, about two in five show no sign that AI is central to what they sell.

Margins are the next filter. The average AI product runs at a gross margin of about 53%, against roughly 80% for classic software. That works if the trend points to 60% and above and the price follows the work done rather than the tokens consumed. A product that resends a huge context with every call, or depends on a single model provider, fails the test.

Consumer AI without retention is the last red flag. About 21% of AI app subscribers are still paying after a year, against 31% for other apps, so a strong launch month no longer raises a round. In crypto, interest has moved from tokens to stablecoins and payments.

The US leads, Europe accelerates, Latin America wakes up

The US still takes well over half of global venture money, and in the first half of the year about two thirds of US dollars landed in the San Francisco Bay Area. If you are raising a nine-figure round for a frontier model, that is still the room.

Europe had its strongest quarter in four years: about $25 billion in the third quarter, up roughly 77% on a year earlier, with AI taking a record 75% of the region's money. The growth leans on a handful of very large rounds, including the largest venture round a European company has ever raised, $3.5 billion for Mistral in Paris. Money is also spreading beyond London to Paris, Munich and Berlin. Public capital is joining: in August the European Commission completed the legal set-up of a new public scaleup fund, with a target of €5 billion for scaleups in artificial intelligence, quantum, biotechnology and clean technologies.

Latin America is small and moving the right way. Startups in the region raised about $1.3 billion in the third quarter, up 26% on a year earlier, and angel and seed rounds rose 62% year on year. Mexico has overtaken Brazil in venture dollars this year, and the largest seed round in the region's history, $85 million, went to an AI-native wealth adviser. From our office in Asunción I see the same thing in applications: strong technical teams, lower costs and much less competition for the first check.

Exits: the giants list, everyone else sells

Liquidity in 2026 has two speeds. At the top, SpaceX raised $75 billion in June in the largest IPO ever, and yesterday Anthropic made its registration statement public, the step before a Nasdaq listing under the ticker ANT. Further down, the window is narrow: most venture-backed IPOs in the third quarter were biotech companies, and the average new listing ended the quarter flat.

For everyone else the exit is a sale. Acquirers have spent at least $120 billion on venture-backed companies this year, on pace to match last year's record, and SpaceX's $60 billion all-stock purchase of Anysphere, the maker of the Cursor coding tool, is the largest acquisition of a startup ever made. At the other end, software companies without an AI core are being sold at a fraction of their peak: Airtable, valued at $11.7 billion in 2021, was sold in the third quarter at an enterprise value of about $1.3 billion, and Miro, once valued at $17.5 billion, went for about $1.4 billion.

The secondary market shows the same split. Secondary volume across private markets set a first-half record of about $120 billion. Venture fund stakes trade at around 79% of their stated value on average, yet stakes in the leading AI companies change hands close to full price while weaker names sell at deep discounts. There is very little in between.

What I take from all this

Raise for proof, not for runway. Checks at the bottom of the market are small and entry prices are high, so the round has to buy one clear milestone: paying customers, usage that holds, a margin that survives the compute bill. A seed round planned around burn ends in a bridge.

Price for the work, not the seat. Investors now compare AI companies with the labour budget they replace, not with the software budget. If your product finishes a task, charge for the finished task, and show what each one costs you to deliver.

Keep the team small and make it a strength. Revenue per employee has become one of the first numbers in a pitch. Five strong engineers with agents in every workflow read as discipline, and forty people at seed read as a plan from 2021.

Own something that compounds. A model anyone can rent is not a moat. Proprietary data, a workflow your customers live in, a regulated licence or a distribution channel others cannot copy is. This year investors pay a premium for exactly these.

Build for a buyer, not only for a bell. For almost every company outside the top few dozen, the exit is a sale. Acquirers pay for technology that slots into products they already sell. Knowing who those buyers are from day one shapes what you build.

Look beyond the Bay Area. The largest rounds stay in San Francisco, and the talent does not. Europe has its strongest venture market in years, and Latin America is growing from a low base with strong engineers and little competition for the first check.

Where we fit

At Fanaura, AI-native companies come first, alongside games and interactive entertainment, technology, blockchain and digital assets, real estate and public and crypto markets. We work from San Francisco, Vilnius and Asunción, invest our own capital and the capital of our LPs through a separate SPV for each deal, and never charge founders anything.

We look for companies where the model does the core work, a small technical team ships every week, and customers pay for results at a price that covers compute. You can read how we work with founders before you write to us. If that describes what you are building, the Apply link below takes about five minutes. If we see a fit, we get in touch.

For founders

Building an AI-native company where the model does the core work? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.