đź’° SF Tech Week VC Panel Insights: The Great Valuation Reset of 2025

General
4

Reading this thread while the “Technical Due Diligence in 2025” session is wrapping up, and everything connects.

How VCs are Evaluating Technical Teams Differently

Panel had:

  • Technical DD partner from Sequoia
  • CTO-in-residence at a16z
  • Technical advisor who evaluates startups for VCs
  • Engineering leader from a unicorn who went through 5 funding rounds

The opening: “What technical metrics actually matter when deciding to invest?”

The answer has changed DRAMATICALLY in 2025.

The Old vs New Technical DD Checklist

2021 Technical DD:

  • Do you have a technical co-founder? âś“
  • Is your code on GitHub? âś“
  • Can you ship features fast? âś“
  • That’s it. Here’s $10M.

2025 Technical DD:

  • Architecture: Can you scale to 10x revenue without rewriting? (show me the diagrams)
  • Unit economics: What’s your compute cost per user? (show me the math)
  • Security: Do you have SOC 2? (show me the report)
  • Team: Do you have senior engineers or just juniors? (show me the LinkedIn profiles)
  • Technical debt: How much will it cost to fix? (show me the backlog)
  • Dependencies: Are you over-reliant on one vendor? (show me your stack)

The Sequoia DD partner: “In 2021, we funded technical vision. In 2025, we audit technical execution.”

How This Maps to @finance_fred’s 64% AI Funding Stat

The a16z CTO-in-residence explained the technical concentration risk:

Problem: 80% of AI startups use the same stack:

  • OpenAI GPT-4 or Anthropic Claude for LLM
  • Pinecone or Weaviate for vector database
  • LangChain for orchestration
  • AWS or Google Cloud for infrastructure

If you’re 100% dependent on OpenAI, what happens if:

  • Their API goes down? (It will)
  • They increase prices 10x? (They might)
  • They launch a competitor to your product? (They probably will)

The VC question: “What’s your differentiation if OpenAI can replicate your product in 6 months?”

Most founders don’t have a good answer.

The Infrastructure Cost Reality

This maps perfectly to @sales_jenny’s revenue metrics:

The technical advisor shared a brutal example:

AI startup financials:

  • ARR: $2M
  • Gross margin: 40%
  • Why so low? OpenAI API costs eat 60% of revenue

Traditional SaaS financials:

  • ARR: $2M
  • Gross margin: 80%
  • Infrastructure costs: 20% of revenue

VC reaction: “Your margins are terrible. How do you ever get profitable?”

The unicorn engineering leader: “We spent 18 months rewriting our entire AI stack to use open-source models. Went from 40% gross margin to 75%. That’s the only reason we raised Series C.”

The 30% Valuation Compression = Technical Trade-offs

@finance_fred mentioned seed valuations down 30%.

From a technical perspective:

With $10M seed (2021):

  • Hire 10 senior engineers at $200K each = $2M/year
  • Build robust, scalable architecture
  • Have time to refactor and optimize

With $7M seed (2025):

  • Hire 7 engineers (mix of senior and mid-level) = $1.4M/year
  • Build “good enough” architecture
  • Accrue technical debt that you’ll pay for later

The a16z CTO: “Startups raising smaller rounds are building faster but messier. They’ll hit scaling issues at $5-10M ARR that will require expensive rewrites.”

The Open Source vs Closed Source Decision

This is tomorrow’s big debate topic, but it came up in technical DD discussion:

VCs now ask: “Are you using open source models or closed source?”

Open source (Llama, Mistral):

  • :white_check_mark: Lower costs (self-host)
  • :white_check_mark: Data privacy (never leaves your infrastructure)
  • :white_check_mark: Customizable (fine-tune for your use case)
  • :cross_mark: More engineering effort
  • :cross_mark: Worse performance than GPT-4/Claude

Closed source (OpenAI, Anthropic):

  • :white_check_mark: Best performance
  • :white_check_mark: Easy to integrate
  • :cross_mark: Expensive at scale
  • :cross_mark: Data privacy concerns (@security_sam’s compliance issues)
  • :cross_mark: Vendor lock-in risk

The Sequoia DD partner: “We prefer companies using open source at the core with closed source as fallback. Shows technical sophistication.”

The Cybersecurity Funding Surge - Technical Angle

@finance_fred mentioned $4.9B to cybersecurity in Q2.

The technical advisor explained why security is fundable:

Security startups have clear technical moats:

  • Proprietary threat detection algorithms
  • Years of training data (attack patterns)
  • Integration complexity (hard to replicate)

AI startups without moats:

  • Using off-the-shelf foundation models
  • Thin wrapper around OpenAI
  • Easy to replicate

The a16z CTO: “If a mid-level engineer can rebuild your product in a weekend using OpenAI’s API, you don’t have a technical moat. You have a marketing problem.”

Brutal but true.

How the 12-Year Exit Timeline Changes Technical Strategy

@finance_fred mentioned 12-year exits.

From a technical perspective:

Technical decisions you make in 2025 will matter until 2037.

The unicorn engineering leader: “We chose MongoDB in 2015. Still using it in 2025. Made the right architectural choices early because we’re stuck with them.”

Questions VCs are asking:

  • Will your technology stack be relevant in 10 years?
  • Are you betting on technologies that might be dead? (looking at you, crypto startups from 2021)
  • Can you migrate to new technologies without rewriting everything?

The Technical Talent Crisis

Both @finance_fred (64% of funding to AI) and @sales_jenny (hiring challenges) mentioned talent.

The numbers are INSANE:

ML Engineer salaries (SF Bay Area, 2025):

  • Entry-level: $180K-250K
  • Mid-level: $250K-400K
  • Senior: $400K-600K
  • Staff/Principal: $600K-1M+

Meanwhile, your seed round is 30% smaller.

The math doesn’t work.

What startups are doing:

  • Hiring remotely (pay $150K for talent in Europe/Asia that would cost $400K in SF)
  • Using contractors for non-core work
  • Leveraging AI tools to boost productivity (GitHub Copilot, Cursor)
  • Focusing team on differentiated work, buying commodity infrastructure

The a16z CTO: “Your 5-person team needs to ship what used to take 15 people. The only way is better tools and ruthless focus.”

The Technical Advice for Founders Raising

Based on the panel:

Before you pitch VCs, have answers to:

  1. “What’s your cloud spend per customer?”

    • If you don’t know, you’re not ready to raise
    • If it’s >20% of ARPU, you have a problem

  2. “Show me your architecture diagram”

    • Can you scale 10x without major rewrites?
    • What are your single points of failure?

  3. “What’s your vendor lock-in risk?”

    • Are you 100% dependent on OpenAI?
    • What’s your mitigation strategy?

  4. “How much technical debt do you have?”

    • Be honest about what needs to be refactored
    • Show you understand the problem

  5. “What happens if your tech lead leaves?”

    • Is all the knowledge in one person’s head?
    • How are you documenting?

The Controversial Framework: Technical Leverage

The Sequoia DD partner shared their “technical leverage” framework:

High technical leverage (VCs love this):

  • 5 engineers supporting $10M ARR = $2M per engineer
  • Automated infrastructure
  • Self-serve product (no professional services)
  • Open source foundations with proprietary layer

Low technical leverage (VCs avoid this):

  • 20 engineers supporting $5M ARR = $250K per engineer
  • Manual processes
  • Requires implementation team for each customer
  • Fully reliant on third-party APIs

They literally calculate revenue per engineer as a key metric now.

How This Connects to @product_david’s PMF Discussion

@product_david mentioned you need PMF in 6-9 months.

From a technical perspective: You can’t over-engineer early. Ship fast, refactor later.

BUT - @finance_fred’s data shows you might not raise Series A for 24 months (vs 18 months in 2021).

So you need to find the balance:

  • Fast enough to find PMF in 6-9 months
  • Solid enough to scale to $5-10M ARR without major rewrites
  • Efficient enough to survive on smaller seed round

The unicorn engineering leader: “This is the hardest part. Move fast but don’t break things TOO badly.”

My Action Items as CTO

After hearing VCs, sales, product, and now technical DD:

  1. Audit our OpenAI dependency - 80% of our product uses OpenAI. Need to build open source fallback.

  2. Calculate revenue per engineer - If we’re below $1M per engineer, we’re overstaffed or undermonetized.

  3. Document architectural decisions - VCs will ask “why did you choose X?” Need good answers.

  4. Plan for 10-year technology horizon - 12-year exit timeline means our 2025 tech decisions matter until 2035.

  5. Hire for leverage, not headcount - One senior engineer with AI tools > three junior engineers.

Questions for CTOs and Technical Founders

  1. What’s your revenue per engineer? Are you hitting $1M+ per engineer?

  2. How dependent are you on one vendor (OpenAI, AWS, etc)? What’s your mitigation plan?

  3. For AI companies: What’s your technical moat? Can mid-level engineer replicate your product?

  4. What’s your gross margin? If it’s <70%, why?

Tomorrow: The open source vs closed source AI debate. I’m expecting this to be THE most important technical discussion of SF Tech Week.

Sources:

  • SF Tech Week “Technical Due Diligence in 2025” panel (Day 2)
  • Sequoia, a16z technical advisors’ frameworks
  • Unicorn engineering leader’s case studies
  • My own experience going through 3 funding rounds as CTO