Cold email template AI VC, infrastructure seed (2026)
Cold email template AI VC founders use at seed, opener grounded in model architecture and a concrete benchmark, with 9 to 12 percent reply rates.
cold-email-ai-seed.txt
Subject: {{BENCHMARK_METRIC}} on {{MODEL_OR_WORKLOAD}}, and a question about {{PORTFOLIO_CO}}
Hi {{FIRST_NAME}},
We hit {{BENCHMARK_METRIC}} on {{MODEL_OR_WORKLOAD}} last week, running on {{COMPUTE_SETUP}}. {{TECHNICAL_WEDGE_ONE_LINE}}.
{{DEPLOYMENT_WIN}}: {{CUSTOMER_NAME}} moved {{WORKLOAD_DETAIL}} onto us in {{TIMEFRAME}}, cutting {{COST_OR_LATENCY_METRIC}}. That is the reason I am writing you specifically, your note on {{PORTFOLIO_CO}} framed the same infra problem we are solving from the other side.
Raising a {{ROUND_SIZE}} seed, {{COMMITTED_AMOUNT}} committed from {{LEAD_OR_ANGELS}}. Deck: {{TRACKABLE_DECK_LINK}}. Free Tuesday or Thursday next week, 20 minutes, your calendar?
{{YOUR_NAME}}
{{COMPANY}} , {{ONE_LINE_DESCRIPTION}}
Variables · fill before sending
- FIRST_NAMEPartner's first name, spelled correctly, no "Hi team"
- BENCHMARK_METRICA concrete, verifiable number (e.g. "2.1x inference throughput vs vLLM baseline", "MMLU 71.3 at 7B", "p95 latency 38ms at 400 RPS")
- MODEL_OR_WORKLOADThe specific model family or workload (e.g. "Llama-3.1-70B", "mixture-of-experts routing", "vector search over 10B embeddings")
- COMPUTE_SETUPThe hardware or runtime, named (e.g. "8xH100 single node", "TPU v5e pods", "CPU-only on Graviton")
- TECHNICAL_WEDGE_ONE_LINEWhat you do architecturally that is different, in one sentence, no adjectives
- DEPLOYMENT_WINLabel for the concrete deployment story (e.g. "First production deployment", "Migration off Triton")
- CUSTOMER_NAMENamed customer or logo. If under NDA, use "a Fortune 500 insurer" style, never "a major enterprise"
- WORKLOAD_DETAILWhat workload they moved and at what scale (e.g. "40B tokens/day of retrieval")
- TIMEFRAMEHow fast the migration or deployment happened (e.g. "nine days", "one sprint")
- COST_OR_LATENCY_METRICThe outcome number (e.g. "inference spend by 62%", "p99 from 340ms to 90ms")
- PORTFOLIO_COA relevant recent investment from this partner in AI infra, ML tooling, or data
- ROUND_SIZETarget round size (e.g. "$3.5M")
- COMMITTED_AMOUNTSoft-circled or committed amount with source (e.g. "$1.2M")
- LEAD_OR_ANGELSNamed lead or notable angels (e.g. "a former Anthropic research lead")
- TRACKABLE_DECK_LINKA DocSend, Pitch.com, or similar trackable link, never a PDF attachment
- YOUR_NAMEYour first name only in the sign-off
- COMPANYCompany name
- ONE_LINE_DESCRIPTIONSix-word description (e.g. "inference runtime for small-model teams")
How to use it
- Lead with one verifiable number, not three. Partners scan the first two sentences. A single sharp benchmark beats a paragraph of stacked claims.
- Name the hardware and the baseline. "2.1x faster" is noise. "2.1x vs vLLM on 8xH100" is a claim a partner can verify or ask about.
- Do not write "AI-powered", "leveraging LLMs", or "cutting-edge". These are the exact phrases AI-infra partners use as filters. Be architectural, not adjectival.
- Attach a trackable deck link, not a PDF. OpenVC's cold email playbook flags the trackable link as table stakes; it also tells you who opened and re-opened.
- Match the partner's actual thesis. Reference a portfolio company they personally led or posted about, not a fund-wide logo. Generic "I saw your fund invests in AI" reads as a blast.
- Skip the Calendly link in the first email. Propose two specific windows. Calendly in a cold email reads as low-effort and drops reply rates, per OpenVC's 13 cold email examples teardown.
- Keep it under 150 words. This template runs 120. If your filled version is over 180, you are re-pitching the deck inside the email.
- Send Tuesday or Thursday, 9am partner-local. AI partners get 30 to 50 of these a week, volume has climbed since AI captured roughly 31% of all venture dollars in 2024. Time-of-day is one of the few free levers left.