Unique Research · Insight / AIGC monthly · October 20, 2023 · Author Jerry · Review Qiuping · Digest of September essays from Sequoia & a16z
Since last summer, generative AI has been in a high-speed phase. ChatGPT made clear that AI is a transformative force for industries and competition. September brought a cluster of industry essays from Sequoia and a16z on economics, consumer use, platform shifts, and healthcare. This chapter summarizes the dry goods.
a16z Martin Casado — “Computing’s third era” and creation’s marginal cost → 0
- For ~10 years, AI beat humans at entity recognition; Meta, Google, Netflix already grew market value with ML.
- Most true-AI use cases compete with humans. Brains evolved for berry-picking and lion-dodging — the infamous AI mediocrity spiral: products that look smart in demos, then underperform expensive human alternatives.
- Third era of computing: chips drove compute’s marginal cost to ~0; internet drove distribution’s marginal cost to ~0; generative AI drives creation’s marginal cost to ~0.
a16z Olivia Moore — 90% of genAI apps grow on organic traffic and monetize
- Most leading genAI apps were built from scratch; ~80% of top sites did not exist a year ago.
- ChatGPT alone is ~60% of Top-50 monthly traffic (~#24 site globally) — still small vs mainstream apps.
- General LLM chatbots ~68% of Top-50 user traffic; AI companions and content tools (image, pro writing) rising.
- Early “winners” exist, but most categories still open.
- GenAI rewrote acquisition: little paid growth; free organic; users pay — ~90% of companies already have revenue; almost all subscription. Avg ~$21/month (~$252/year).
- Still browser-first; mobile-native genAI is the next platform bet.
OpenAI CTO Mira Murati × Martin Casado — redefining human–digital interaction
- Inflection point: AI redefines how people interact with digital information.
- More data + compute → stronger models; AGI may need other breakthroughs.
- Users often need a right-sized model, not the strongest — cheaper; building great products on models is hard.
- Text is strong today; multimodal (image/video) next.
- RLHF for reliability; web access for freshness and hallucinations.
- Vision: ensembles of agents users collaborate with; a platform where developers build agents.
Sequoia — Generative AI’s Act Two
- GenAI as a platform shift / Cambrian explosion / generation’s space race — innovation density not seen since early internet.
- Enablers: 60 years of Moore’s Law (exaflops), 40 years of internet data (COVID-accelerated), 20 years of mobile/cloud (a supercomputer in every pocket).
- AI researchers went from garage hackers to teams commanding billions in compute.
- Hype excess: every inbox pitched “AI Salesforce / Adobe / Instagram” — unsustainable fundraising, talent wars, GPU grabs.
- Yet startups already did >$1B genAI revenue in months — faster start than early SaaS.
- Act Two: apps that use foundation models as part of fuller solutions to human problems — plus a new LLM developer stack of compute/tools for production.
- Sequoia’s earlier piece: some forecasts wrong (pace exceeded expectations), some right (new track, killer apps appearing, developers central, product form still evolving).
a16z — Let’s Get Personal: AI unlocks online consumer services
- Internet made services cheaper/online; AI lets startups scale experiences that copy the best of 1:1 human service (tutoring, coaching, doctors).
- Survey: 57% prefer face-to-face mainly for customization, expertise, or personal relationship — all areas genAI can address.
- Opportunity may be 5–20× today’s online consumer services; biggest upside in high-touch verticals e-commerce hasn’t penetrated (vs transactional utilities/flights).
Sequoia — AI’s $200B Question
Nvidia’s Q2 beat → endless GPU demand narrative. Thumb rule: $1 GPU ≈ $1 data-center opex. ~$50B GPU run-rate → ~$100B DC spend; with 50% end-user margin → ~$200B revenue needed. Even generous AI revenue from OpenAI (~$1B), Microsoft Copilot (~$10B), Google/Meta/Apple (~$10B each), and Oracle/ByteDance/Alibaba/Tencent/X/Tesla (~$5B each) leaves a >$125B annual hole — largely for startups to fill by turning infra into end-customer value.
a16z — What Builders Talk About When They Talk About AI
Voices from OpenAI, Microsoft, Anthropic, Databricks, Character.AI, Roblox, insitro, Figma: start of computing’s third era; transformers ~6 years old, ChatGPT <1 year; creation’s marginal cost → 0; “Wright brothers’ first flight” moment; GitHub Copilot as template for AI amplifying cognitive work; foundation models changing behavior across art, design, games, medicine, code, relationships.
Sequoia — Bringing Generative AI to Healthcare
- U.S. healthcare >$4T spend; ~$300B administrative — repetitive, labor-heavy.
- Near-term: back-office automation + frontline workflow efficiency; aspiration: better patient outcomes.
- Use cases Sequoia sees: clinical documentation, patient engagement, prior auth, and more daily.
- Past decade of health software faced thin provider margins and concentrated payers (top 5 >50% share). GenAI’s immediate, visible value helps break the “poverty trap” of buyers who won’t pay for long-dated efficiency promises.