Ideas
Capital now compounds AI labs' advantage
Within any AI category the power law is extreme and the winner captures most share and market cap, but AI will massively expand the number of categories, so investors should back the leading company in every credible category.
Frontier AI leaders are underrepresented
The power law is now systemic in technology investing: the three frontier companies SpaceX, OpenAI, and Anthropic represent $3.5-5T of potential enterprise value, and many LPs and institutional allocators lacked exposure to them.
AI should be core allocation
The AI bottleneck is now supply-side, not demand: energy, grid, data centers, chips, frontier models, and apps are the chain; the U.S. is strong in chips and onwards but weak in speed to power, permissioning, transmission, and regulation, making next-generation chips, memory, data centers, and energy infrastructure large opportunities.
Healthcare AI TAM is tenfold SaaS
AI's TAM can be more than 10x traditional SaaS or healthcare IT because it goes after actual labor and task value; in healthcare, claims, billing, and administration alone represent a trillion-dollar industry.
AI stack growth is not zero-sum
The AI stack should not be viewed as zero-sum: frontier labs, open-source models, and application-layer companies can all grow as the market expands, so investors should avoid assuming one layer must cannibalize another.
Concentrate in top venture firms
Consistency in venture is extremely rare—only 20 of 3,000 U.S. VC firms achieved consistent 3x net returns over two decades—so LPs should concentrate capital in the top 15-20 firms and get access, selection, and sizing right rather than spreading across 50-70 managers.
Concentrate in top venture firms
Consistency in venture is extremely rare—only 20 of 3,000 U.S. VC firms achieved consistent 3x net returns over two decades—so LPs should concentrate capital in the top 15-20 firms and get access, selection, and sizing right rather than spreading across 50-70 managers.
Avoid middle; back specialist or scaled VC
There is a death of the middle in venture: specialized early-stage funds with deep domain expertise and large multi-stage platforms with operating resources can win, while mid-sized firms struggle because founders choose partners who can de-risk outcomes and provide brand signaling.
Avoid middle; back specialist or scaled VC
There is a death of the middle in venture: specialized early-stage funds with deep domain expertise and large multi-stage platforms with operating resources can win, while mid-sized firms struggle because founders choose partners who can de-risk outcomes and provide brand signaling.
Pre-seed funds complement large venture firms
Pre-seed funds with sub-$100M vehicles can coexist with large firms: large firms often wait for more certainty and lead later rounds, while small pre-seed funds can take a clip or two earlier, carve out a niche, and still have a right to win.
Early-stage edge powers late-stage venture
The strongest late-stage franchises are attached to an early-stage franchise; that relationship allows them to size 5-10% of a late-stage fund into category-defining companies, whereas de novo late-stage funds struggle to write large $500M checks.
Legal AI adoption inflected post-reasoning models
Real AI traction should be judged by customer demand and usage texture, not just cohort or renewal data; Harvey is a positive example because usage inflected after reasoning models improved and clients began demanding law firms use the product.
Favor AI-accelerating SaaS, avoid legacy
In public SaaS, only 15-20 companies still trade above 10x revenue, and most are showing AI-driven growth acceleration; investors should favor AI-tied software with acceleration while non-AI legacy software remains challenged.
Enterprise AI adoption remains very early
Enterprise AI diffusion is still very early and therefore very bullish: the median U.S. company spends $12 per employee per month on AI, the top 1% spends $7,000, and only 10-30M users are active versus 1.5B knowledge workers.
Legacy software LBOs face credit stress
Pre-ChatGPT private equity software LBOs are challenged: 2021-22 software deals were done at 25-32x EBITDA with over $200B of debt, and as public comps fell toward 2x revenue, leverage ratios rose; non-AI-resilient software faces both equity and credit stress.
Robotics will exceed language AI
Robotics is barely penetrated and will be bigger than language AI, with the opportunity likely to develop over the next 10 years.
Autonomy and robotaxis remain early
Autonomy and robotaxis are still almost untouched—fewer than 10,000 Waymos live in the U.S.—leaving large open space for new companies to build and create value.
This a16z video, published September 10, 2026,
features David George, Jen Kha, Aram Verdiyan
discussing AI-SECTOR, OPENAI, ANTHROPIC, SPCX, Healthcare AI administration, AI ecosystem, Top-tier venture capital funds, Average venture capital funds, Mid-sized venture funds, Large multi-stage venture firms, Specialized early-stage venture funds, Pre-seed venture funds, Multi-stage venture firms with early-stage franchise, Harvey, Legacy SaaS, Enterprise AI, Private equity software LBOs, Software private credit, ROBO, ARKQ.
17 trade ideas extracted by AI with direction and confidence scoring.
Speakers:
David George,
Jen Kha,
Aram Verdiyan
· Tickers:
AI-SECTOR,
OPENAI,
ANTHROPIC,
SPCX,
Healthcare AI administration,
AI ecosystem,
Top-tier venture capital funds,
Average venture capital funds,
Mid-sized venture funds,
Large multi-stage venture firms,
Specialized early-stage venture funds,
Pre-seed venture funds,
Multi-stage venture firms with early-stage franchise,
Harvey,
Legacy SaaS,
Enterprise AI,
Private equity software LBOs,
Software private credit,
ROBO,
ARKQ
