AI fundraising in 2026: what US VCs actually fund
Three weeks ago a founder showed us a deck where the product was a Slack bot that summarized meeting notes using GPT-4. The pitch was "AI for sales teams." Sequoia passed in 48 hours. The reason: thin wrapper, no moat, Salesforce will ship this feature in Q3.
That's the AI fundraising market right now. AI startups are raising larger rounds than the broader seed market—AI/ML seed rounds averaged $4.6M in March, versus $3.1M for all US seeds—but the bar is wildly uneven depending on what you're building. Series A valuations for AI companies are up 23% year-over-year according to Qubit Capital, with median AI Series A posts exceeding $50M. But most of that capital went to a small number of frontier model labs and infrastructure plays.
This is a working brief on how US VCs sort AI deals, where the money is going, and what you need to show if you're raising.
How US VCs bucket AI companies
Most funds sort AI deals into four categories, each with a different funding bar.
Frontier model labs. OpenAI, Anthropic, Mistral, x.AI. Funded at billion-dollar posts even pre-revenue. This category is closed unless you have ex-DeepMind or ex-OpenAI co-founders and access to a frontier compute cluster. You're not raising here unless you already know you are.
AI infrastructure. Vector databases, fine-tuning platforms, observability tools, evals, model deployment, inference optimization. This is where a lot of the 2024-2025 capital went. $4-8M seed rounds at $20-40M posts are still achievable if you're solving a real infrastructure problem. Examples: companies building better retrieval systems, tools that reduce inference cost, platforms that let enterprises fine-tune models without hiring an ML team.
AI-native applications. Products that couldn't exist without LLMs. Cursor is the canonical example—it's not "code editor with AI features," it's "AI that happens to live in an editor." Perplexity is another. These raise well if the product is genuinely differentiated, but the bar is high. Expect $3-6M seed at $15-25M post if you have real traction.
AI features bolted onto existing categories. "AI for legal," "AI for sales," "AI for HR." This is the largest bucket and the hardest sell. Incumbents like Salesforce, Workday, and HubSpot are already shipping AI features. Unless you have a clear wedge—proprietary data, deep workflow integration, a specific persona the incumbent can't serve—US VCs will pass. We see this every quarter: a founder builds a solid product, gets early traction, then hits a wall when the incumbent announces their AI roadmap.
Why "thin wrapper" kills deals
The most common rejection reason we see: "This is just a wrapper on Claude." If your product is a prompt layer on top of a foundation model API, investors discount it heavily.
The defenses that work:
Custom data. You have proprietary data the foundation model doesn't, and you're using it for retrieval or fine-tuning. A legal tech company with 10 years of case outcomes indexed and structured has a moat. A legal tech company that sends a case summary to GPT-4 does not.
Workflow integration. Your product lives inside a workflow the customer can't replicate by opening ChatGPT. If a user can get 80% of your value by copy-pasting into Claude, you don't have a product.
Evals and guardrails. You're validating outputs against domain-specific benchmarks. A healthcare AI that hallucinates is worthless. If you've built evals that catch hallucinations before they reach the user, that's real engineering.
Compound product. You're not one LLM call. You're 10+ AI features that together form a system. The system is the moat, not any individual feature.
Capital efficiency is the new bar
The 2024 playbook—raise $20M, figure out PMF later—is dead. Investors now ask about burn multiple even at seed. They want to see your path to Series A metrics without needing a bridge round.
For AI companies, this means three things:
Burn multiple matters. If you're burning $300K/month and growing 8% MoM, you're not getting a Series A. Plan for 15%+ monthly growth or show a clear path to profitability.
Compute spend projections. Investors will model your inference costs. If your unit economics depend on GPT-4 API calls and you're not planning to fine-tune a smaller model or switch to open-source, they'll assume your gross margins stay at 50%. That's a problem. Plan for 70%+ gross margins at scale or have a story for how you get there.
Gross margin trajectory. AI products starting at 50% gross margin get flagged. Show how you'll improve margins as you scale—switching to cheaper models, batching inference, fine-tuning on your own data.
Team pedigree matters more for AI
US VCs weight AI team backgrounds more heavily than they do for B2B SaaS. The signal hierarchy:
Ex-OpenAI, Anthropic, DeepMind, FAIR, Google Brain. Highly fundable. If you were a research scientist or engineer at a frontier lab, you can raise a seed round on a deck and a demo.
Top university AI programs. CMU, MIT, Stanford, Berkeley, Toronto, ETH Zurich. Strong signal, especially if you published at NeurIPS or ICML.
FAANG senior IC or director with ML experience. Acceptable if paired with domain expertise. A Google L6 who built ranking systems can raise for an AI product in a space they know well.
Strong technical founder, no AI lab pedigree. Workable if the product execution is exceptional. We've seen founders with backend engineering backgrounds raise for AI infra plays because they solved a real problem well.
Non-technical founder building AI. Hard sell unless you have a strong technical co-founder. Investors assume you'll get out-executed by someone who can read the research papers.
Geography still matters
SF and NYC dominate AI fundraising. Within SF, the concentration is even tighter—Hayes Valley and SoMa are the AI epicenter. Boston has a strong ecosystem around MIT alumni. Austin and Seattle have smaller but real AI communities.
If you're outside the US and building AI, plan to spend significant time in SF during your raise. Remote-only AI fundraising is harder than remote-only B2B SaaS fundraising. Investors want to see you at the AI events, the dinners, the office hours. It's not fair, but it's real.
AI-focused funds worth knowing
Beyond the generalist funds we covered in our top US seed VCs piece, these funds are AI-heavy or AI-specific:
AI Grant (Daniel Gross, Nat Friedman) focuses on frontier AI. Conviction Partners (Sarah Guo) backs AI infrastructure and applications. Radical Ventures funds AI infra and applied AI. Costanoa Ventures writes checks for enterprise AI. SignalFire has a data-driven, AI-friendly thesis. Felicis Ventures backs AI tooling and applications. Initialized has a strong AI infra portfolio. a16z's Infrastructure team applies the firm's partnership model to AI infra specifically.
What your deck needs to show
Four slides matter more for AI than for non-AI companies.
Compute cost slide. Show your unit economics including inference cost per user or per transaction. Most AI decks skip this. Investors will model it themselves and assume the worst if you don't show it.
Eval slide. How do you measure quality? If you're building a product that generates outputs, you need evals. Without them, your product claims are unverifiable. Show the eval framework, the benchmarks, the accuracy metrics.
Defensibility slide. Why does your moat survive when GPT-5 ships? The answer can't be "we're first to market." It needs to be workflow integration, proprietary data, network effects, or a compound product that's hard to replicate.
Talent slide. If your team has AI lab backgrounds, lead with it. If you don't, lead with domain expertise and show why you can out-execute the AI lab alumni who will enter your space.
If you're building something that's not AI
The bar is higher. VC attention is concentrated in AI, and many partners have shifted to AI-heavy theses. Non-AI categories need to clear a higher bar to compete for partner time.
The fix: lean into a clear non-AI thesis where AI hype is less dominant. Regulated industries, deep tech, climate, biotech. Or find the funds that are explicitly skeptical of the AI froth. They exist—some Tier 1 funds are avoiding AI deals because they think the market is overheated.
We've seen non-AI founders succeed by targeting the partners who are tired of seeing AI pitch decks. If you're solving a hard problem in a space that doesn't benefit from LLMs, say so clearly. Don't add "AI-powered" to your pitch because you think it helps. It doesn't.
Next steps
If you're raising for an AI company, figure out which of the four buckets you're in. Frontier model labs are closed. Infrastructure is still fundable but competitive. AI-native apps need exceptional execution. AI features on existing categories need a moat the incumbent can't replicate.
Build the compute cost slide. Get evals that show your product works. Identify the 8-12 funds that match your specific category. If you're non-AI, find the 5-10 funds least dominated by AI thesis.
If you want help calibrating your AI fundraising strategy, book a discovery call.
