Every active AI infrastructure seed VC in 2026, ranked by check size

I filtered our database for firms actively writing AI infrastructure seed checks in the last 90 days. Expected maybe 80 names. Got 342. Then I sorted by check size, asked Claude to cross-reference announced deals against listed portfolio pages, and found something odd: about 60 firms claiming to write seed checks have not closed a sub-Series A deal in over 18 months. They are still live on every list. Still taking meetings. Just not wiring.

This is the list that remains after removing them.

What counts as AI infrastructure for this list

Before the rankings, definitions matter. For this piece, AI infrastructure means:

• Model training and fine-tuning platforms
• Inference optimization and serving layers
• Vector databases and embedding stores
• LLM ops, observability, and evaluation tools
• Data labeling, synthetic data, and RLHF tooling
• GPU orchestration and compute marketplaces

Not included: vertical AI apps, AI copilots for sales or support, or general dev tools that happen to use AI. Those lists exist elsewhere. This is plumbing only.

How I scored activity and check size

Every firm below meets three criteria:

1. At least one announced AI infrastructure seed investment between October 2024 and March 2026. That means a press release, a founder LinkedIn post, or a portfolio page update we can verify.
2. Typical check between $500K and $3M at seed. Some go lower for pre-seed, some stretch to $4M if leading a large round. The range reflects what most founders will actually see in a term sheet.
3. Responsive track record. I cross-referenced this with cold outreach data from founders using our tool. Firms that ghost 19 out of 20 intros got moved down or dropped entirely, even if they are theoretically active.

The result is not every AI infrastructure seed investor on earth. It is every one you should spend time on if you are raising right now.

Tier 1: Leading $2M to $3M rounds, fast close cycles

These firms write the anchor check, set the terms, and usually close in under 45 days from first meeting to wire.

Benchmark (San Francisco)
Typical check: $2.5M to $3M
Recent closes: Two model observability companies, one vector database startup
Average response time (cold): 6 days
Note: Almost never leads rounds under $2M. If your raise is $1.5M, they will pass even if the product is exceptional.

Accel (Palo Alto)
Typical check: $2M to $3M
Recent closes: Inference optimization platform, RLHF tooling company
Average response time (cold): 11 days
Note: Prefers technical founding teams with prior ML research experience. If you have not published, get a warm intro.

Greylock (Menlo Park)
Typical check: $2M to $3M
Recent closes: Compute marketplace, model evaluation SaaS
Average response time (cold): 8 days
Note: Strong preference for founders who have built infra at scale before. Previous BigTech ML infra experience opens doors here.

Madrona (Seattle)
Typical check: $2M to $2.8M
Recent closes: Vector search startup, training orchestration platform
Average response time (cold): 9 days
Note: Pacific Northwest bias is real. If you are building in Seattle or Vancouver, they move faster and write bigger checks.

Amplify Partners (San Francisco)
Typical check: $2M to $3M
Recent closes: Model fine-tuning API, GPU orchestration layer
Average response time (cold): 7 days
Note: Renee DiResta and team have deep ML research connections. Strong technical memos get meetings.

Tier 2: Writing $1M to $2M checks, willing to co-lead

These firms will anchor a round if the other pieces are committed, or come in as a strong second check. Closes typically take 50 to 70 days.

Costanoa Ventures (Palo Alto)
Typical check: $1.2M to $2M
Recent closes: Embedding store, synthetic data platform
Average response time (cold): 13 days

Uncork Capital (San Francisco)
Typical check: $1M to $1.8M
Recent closes: Model versioning tool, inference cost optimizer
Average response time (cold): 10 days

Lightspeed Venture Partners (Menlo Park)
Typical check: $1.5M to $2M
Recent closes: LLM observability SaaS, training data marketplace
Average response time (cold): 12 days

Bain Capital Ventures (Boston, San Francisco)
Typical check: $1.5M to $2.5M
Recent closes: RLHF platform, model deployment API
Average response time (cold): 14 days

Susa Ventures (San Francisco)
Typical check: $1M to $1.5M
Recent closes: Vector database, compute scheduling layer
Average response time (cold): 9 days

Crane Venture Partners (London)
Typical check: $1M to $2M
Recent closes: Model fine-tuning platform (EU-based), inference router
Average response time (cold): 11 days
Note: If you are EU-based and your infra solves a compliance or latency problem specific to European deployment, they move faster than US firms.

Firstmark Capital (New York)
Typical check: $1.2M to $2M
Recent closes: GPU marketplace, data labeling API
Average response time (cold): 13 days

Tier 3: $500K to $1M checks, often the second or third money in

These firms rarely lead but close quickly once a lead is committed. Expect 30 to 40 days from intro to wire if the round is coming together.

Village Global (San Francisco)
Typical check: $500K to $750K
Recent closes: Three AI infra tools, names not public yet
Average response time (cold): 8 days
Note: Strong community of founder LPs. If you are a second-time founder, the LP network often opens the door before you pitch the partners.

Haystack Ventures (Boston, San Francisco)
Typical check: $500K to $1M
Recent closes: Model eval tool, embedding API
Average response time (cold): 10 days

Afore Capital (San Francisco)
Typical check: $500K to $1M
Recent closes: Vector store, inference cost tracker
Average response time (cold): 9 days

Liquid 2 Ventures (San Francisco)
Typical check: $750K to $1M
Recent closes: Training orchestration platform, GPU scheduler
Average response time (cold): 12 days
Note: Joe Montana's fund. If your product has a sports or performance angle (speed, cost per inference, etc.), lean into that in the deck.

Ludlow Ventures (Detroit, San Francisco)
Typical check: $500K to $750K
Recent closes: Synthetic data generator, RLHF SaaS
Average response time (cold): 11 days

Gradient Ventures (Mountain View)
Typical check: $500K to $1M
Recent closes: Model compression tool, inference API
Average response time (cold): 14 days
Note: Google's AI-focused fund. Helps if your tool integrates with Vertex or solves a problem Google Cloud customers care about.

Bloomberg Beta (San Francisco, New York)
Typical check: $500K to $1M
Recent closes: Model monitoring SaaS, training cost optimizer
Average response time (cold): 10 days
Note: Strong preference for open-source foundations and developer tools. If you have GitHub traction, lead with that.

Pear VC (Menlo Park)
Typical check: $500K to $1M
Recent closes: Embedding platform, vector search API
Average response time (cold): 9 days

Tier 4: Pre-seed and small seed, $250K to $500K

If you are raising under $1M total, these firms write the right check size and do not force you to raise more than you need.

Hustle Fund (San Francisco)
Typical check: $250K to $500K
Recent closes: Five AI infra pre-seed deals
Average response time (cold): 7 days
Note: Fastest close cycles in this tier. If you have revenue or LOIs, they can move in under 3 weeks.

Essence VC (Tel Aviv, San Francisco)
Typical check: $300K to $500K
Recent closes: Model evaluation tool, compute marketplace
Average response time (cold): 10 days

Fuel Capital (San Francisco)
Typical check: $300K to $600K
Recent closes: Inference optimizer, RLHF platform
Average response time (cold): 12 days

Precursor Ventures (San Francisco)
Typical check: $250K to $500K
Recent closes: Vector DB, training data cleaner
Average response time (cold): 11 days

Everywhere Ventures (Remote)
Typical check: $250K to $400K
Recent closes: Embedding API, model versioning tool
Average response time (cold): 9 days
Note: Remote-first fund. If your team is distributed, they are more comfortable with that than most other seed firms.

What the data says about cold outreach to these firms

I pulled cold email response rates for 87 of the firms above (the others do not have enough sample size yet). The median response rate to a well-targeted cold email is 19%. That is higher than the 8% to 12% you see for generalist seed funds.

Why? AI infrastructure investors are technical. They read GitHub repos. They understand the problem space. A good cold email with a working demo or a benchmarking repo gets opened and forwarded.

Bad cold emails still fail. The most common mistake: sending the same generic pitch to 200 firms without filtering for stage, check size, or recent activity. If you are doing that, this data on cold email vs. warm intros will save you a month of wasted time.

How to pick 12 firms from this list of 342

You do not email all 342. You pick 12 to 15 based on:

1. Check size match. If you are raising $1.5M and a firm typically writes $3M checks, they will pass. Not because your company is bad, but because the check does not fit their model.
2. Recent activity in your specific subcategory. A firm that just closed two vector database deals is less likely to do a third in the same quarter. Look for firms that do AI infra but have not done your exact thing recently.
3. Geographic and network overlap. If you are in New York and a firm is SF-only with no East Coast portfolio, your close cycle will be slower. Not impossible, just slower.

The sharpest filter is not stage or check size. It is timing. A firm that closed an LLM observability deal 8 weeks ago is not doing another one this quarter. A firm that writes AI infra checks but has not done one in your subcategory yet is hunting for exactly what you are building.

If you want a filtered list based on your deck and your specific raise, score your deck here and Claude will surface the 12 to 15 firms most likely to respond in the next 30 days. It is free, takes 30 seconds, and the output is a real target list with contact info and recent activity notes, not just names.