Every capital cycle has an address. In railroads it was a handful of banks; in the PC era it was a short stretch of Sand Hill Road. In AI, the address is a surprisingly small list of venture firms whose checks decide which model labs get compute, which application companies get distribution, and which infrastructure picks up the rest. If you are a founder raising for an AI company, this list is your map. If you are an investor or operator, it is something better: a live readout of where the most informed money believes the value will settle.
This article profiles seven firms that are genuinely active in AI, not merely exposed to it. For each: what the firm is, its signature AI positions, what those positions signal, and who should be pitching them. Then the more useful layer: how to read the seven portfolios together, and the mistakes founders make when they treat this list as a leaderboard instead of a map.
How to read “active in AI”
Two different games run under one label. The first is the model-lab game: multi-billion dollar rounds into frontier labs, where checks are enormous, access is scarce, and the bet is on compute plus research talent compounding. The second is the application and infrastructure game: seed through growth rounds into companies that sit on top of the models or sell tools to those who do. A firm can be elite at one and absent from the other. Knowing which game a firm actually plays matters more than its logo, because it predicts check size, diligence style, and what the partner across the table will consider proof.
The seven firms
1. Andreessen Horowitz: the index position
The largest platform firm in venture has effectively indexed the AI stack: a repeat OpenAI backer, lead of Mistral’s Series A, and early money in ElevenLabs, alongside a long tail of application and infrastructure positions. The signal in the portfolio is breadth as strategy; the firm is betting the transition itself rather than a single winner. Best fit: founders who want the biggest platform machine behind them, including marketing, talent, and go-to-market support, and who can tolerate being one position among many.
2. Sequoia Capital: the compounding franchise
Venture’s most durable franchise plays both games. It is a repeat OpenAI backer and was one of the co-leads of Anthropic’s Series H, while its application bets, including Harvey in legal and Glean in enterprise search, share a pattern: proprietary workflow and distribution rather than thin wrappers on someone else’s model. The signal: Sequoia believes applications win on the workflow they own. Best fit: application founders with real usage who can show a moat that is not the model.
3. Khosla Ventures: the conviction check
Khosla wrote the first institutional check into OpenAI, a reported $50 million in 2019, when the consensus considered it philanthropy with extra steps. That is the firm’s whole character: earliest, riskiest, most contrarian, with a deep-tech appetite that extends to fusion and robotics. The signal: conviction before evidence, sized to matter. Best fit: research-heavy founders raising before there is a product, who need an investor comfortable being wrong publicly for years.
4. Lightspeed Venture Partners: the scale-up lead
Lightspeed has become the firm that leads the round after the breakout: it led Anthropic’s reported $3.5 billion Series E in early 2025 and has been an early and repeated backer of Mistral. A global partnership gives it reach across US and European AI. The signal: heavy conviction in frontier labs, entered at scale rather than at inception. Best fit: companies with breakout metrics raising large growth rounds from a lead that can anchor them.
5. Thrive Capital: the concentration thesis
Thrive runs the opposite of an index: few positions, enormous sizing, and repeated doubling down, most visibly as lead of OpenAI’s $6.6 billion October 2024 round. The signal: portfolio concentration as a deliberate strategy, closer to a crossover fund’s conviction than classic venture spray. Best fit: category leaders who want a lead investor that keeps buying in every subsequent round instead of diluting attention across hundreds of names.
6. Menlo Ventures: the ecosystem builder
Menlo led one of Anthropic’s major 2023 rounds and then did something more interesting: it launched the Anthology Fund, a $100 million vehicle with Anthropic to back companies building on Claude. The signal: the firm is investing in a model ecosystem the way earlier funds invested in an operating system. Best fit: founders deliberately building on a specific frontier platform who want capital that comes with ecosystem access rather than neutrality.
7. Radical Ventures: the specialist
Radical is the rare firm that has been AI-only since before it was fashionable, with early positions in Cohere and autonomous-driving company Waabi, and a partnership stacked with technical operators. The signal: judgment at the research layer, where generalists must borrow conviction. Best fit: technical founders whose edge is the model or method itself, who want investors able to read the paper and not just the pitch.
The map in one table
| Firm | Center of gravity | Signature AI position | Best fit |
|---|---|---|---|
| Andreessen Horowitz | Full stack, all stages | OpenAI, Mistral Series A lead, ElevenLabs | Founders wanting the platform machine |
| Sequoia Capital | Applications plus labs | OpenAI, Anthropic Series H co-lead, Harvey | Application founders with workflow moats |
| Khosla Ventures | Earliest deep tech | First VC check into OpenAI, 2019 | Pre-product research founders |
| Lightspeed | Growth-stage labs | Anthropic Series E lead, Mistral | Breakout companies raising big rounds |
| Thrive Capital | Concentrated late stage | OpenAI October 2024 round lead | Category leaders wanting a repeat buyer |
| Menlo Ventures | Model ecosystem | Anthropic rounds, Anthology Fund | Founders building on Claude |
| Radical Ventures | AI-native early stage | Cohere, Waabi | Technical founders at the research layer |
What the seven portfolios say together
Read as one dataset, the list makes three statements. First, the lab game has consolidated: the frontier rounds now route through a handful of firms with the fund sizes to matter, which is why the same names recur across OpenAI and Anthropic’s cap tables. Second, the smart application money is converging on a single filter: own the workflow and the distribution, because the model layer beneath you is a rented advantage. Third, ecosystem strategies are emerging, with Menlo’s Anthology Fund the clearest case of venture capital organizing itself around a platform the way it once organized around iOS.
For investors, that concentration cuts both ways. The firms with the best information are visibly all-in, which is signal; they are also structurally unable to be bearish, which is noise. The discipline for separating the two is the same one that applies to any headline valuation, and it is the entire subject of reading an AI IPO like an operator.
A worked example: picking a lead for an AI application Series A
Suppose you run an applied AI company: real revenue, a vertical workflow product, models rented from a frontier lab. Which of the seven do you actually pitch?
Cross off the mismatches first. Khosla is built for pre-product research risk you no longer represent. Thrive and Lightspeed lead rounds a stage or two later than yours. Radical wants the edge in the model itself, and yours is in the workflow.
That leaves three real conversations. Sequoia if your moat story is distribution and workflow ownership, because that is the pattern its application bets already follow. Menlo if you are committed to Claude, where the Anthology Fund turns your platform dependence into an asset instead of a diligence question. Andreessen Horowitz if what you most need is the platform machine and you are confident you will not get lost in the index. Three targeted pitches beat thirty generic ones, and the targeting logic came entirely from reading positions, not brands. Build that reading discipline once and reuse it everywhere; the general version is in why research architecture comes before investing decisions.
Common mistakes
Pitching the brand instead of the thesis. Founders queue for the most famous logo when the honest fit is two rows down the table. The firm’s existing positions tell you what it believes; pitch into that belief or expect a polite pass.
Treating lab investors as application investors. A firm that writes billion-dollar checks into frontier labs is not automatically the right seed lead for your vertical SaaS product. Same sector, different game.
Ignoring portfolio conflicts. Concentrated AI portfolios mean your likeliest competitor may already be in the room. Check the portfolio before you send the deck, not after.
Reading valuation as validation. A famous firm’s mark on a hot AI company is a negotiated number, not a verdict; the gap between private marks and clearing prices is where discipline lives. If you use AI tools to research any of this, make the model argue the bear case too; the method is in the minimum viable AI investing workflow.
FAQ
Which venture capital firms are most active in AI? The most active names in frontier AI include Andreessen Horowitz, Sequoia Capital, Khosla Ventures, Lightspeed, Thrive Capital, Menlo Ventures, and AI-only specialist Radical Ventures. They play different games: some lead multi-billion dollar model-lab rounds, others focus on application and infrastructure companies built on top.
How should a founder choose which AI investors to pitch? Match the firm’s demonstrated positions to your stage and your edge. Research-stage founders fit conviction investors like Khosla; workflow application companies fit Sequoia’s pattern; companies building on Claude fit Menlo’s ecosystem strategy. Three pitches aimed at demonstrated beliefs outperform a spray of thirty.
Are AI-only venture funds better than generalists for AI startups? Neither dominates. Specialists like Radical Ventures offer technical judgment at the research layer, which matters most when your edge is the model or method. Generalist platforms offer distribution, talent networks, and follow-on depth, which matter most once your edge is commercial.
Do these firms invest outside Silicon Valley? Yes. Lightspeed and Andreessen Horowitz hold significant European AI positions, most visibly in Mistral; Radical Ventures has roots in Toronto’s research community. The lab game concentrates capital in a few firms, not in one zip code.