The world is too loud. Read what matters.

The a16z Show

AI Makes Venture's Power Law Steeper, Not Gentler

Of 3,000 venture firms, only 20 delivered 3x net returns for twenty straight years. For the first time, AI lets money convert directly into compute, and compute directly into product advantage — so the power law gets steeper, not flatter.

Venture CapitalAI AllocationPower LawLP PerspectivePrivate Equity

The video won't play here. Listen to the audio instead:

Three frontline LPs/GPs lay out the allocation logic of venture in the AI era: why money can compound directly into advantage for the first time, and why LP and GP incentives point in opposite directions. High information density, but the second half is worth more than the opening.

The argument · tap a timestamp to hear it

2:04

For the first time, money converts directly into product advantage

David George says the classic way to destroy a startup used to be giving it too much money: it would hire a thousand people, creating coordination problems, expense problems and warring priorities. This time is different — you can throw money straight into compute, and compute makes the product and the business better. This is a new form of ‘increasing returns to scale’ in the AI era: the old mechanisms of brand, reputation and accumulated resources are still there, but capital itself has become a self-reinforcing factor of production for the first time. So a more extreme power law doesn't surprise him.

— David George
4:04

AI hit in four years what SaaS took fifteen to reach

Ram's comparison: AI has already reached $100 billion in revenue; SaaS took 15 years to reach the same point, AI took only 4, and demand penetration is nowhere near done. He argues the root of ‘you can pour money into these companies’ is that there is no ceiling on inference demand. More critically, AI is attacking every face of GDP at once — transportation, labor, services, capital, coordination; no technology paradigm has ever hit so many parts of a $30 trillion GDP simultaneously. So for allocators this shouldn't be a satellite position; it should be core, even super-core.

— Ram
6:07

AI's TAM isn't software, it's the tasks being performed

Take healthcare: healthcare IT spends only $60 to $100 billion a year, but AI is attacking the value of actual labor and tasks — claims, billing, administration — which is a trillion-dollar industry. So AI's TAM can be more than 10x larger than traditional SaaS or healthcare IT. Ram says he has ‘chronically misjudged how big these outcomes could be,’ and David admits the same. Another number: the US economy spends about 40x more on labor than on software. But David stresses this doesn't mean labor disappears — it means labor gets reinvented.

— Ram / David George
11:14

Of 3,000 venture firms, only 20 sustained 3x

Ram's team looked at data on 3,000 US venture firms: over the past twenty years only 20 consistently delivered 3x net returns — under 1%. And you don't need seven or eight funds over twenty years to qualify; three to four funds at 3x net TPPI is enough. What those 20 share: they got the category-defining company in every vintage. The comparison point is Cambridge's data — over the past decade venture's average net return was only 1 to 2x, worse than private equity, worse than public markets, and your money is locked up for ten years. So whether you're in those 20 firms decides almost everything.

— Ram
13:18

The middle layer of funds is being squeezed from both ends

Ram and David repeatedly bring up the ‘death of the middle.’ Two ends survive: one is the very early, deeply vertical specialist fund, which moves fast and can grab the share that big funds also want; the other is the platform mega-fund that covers the full lifecycle from seed to IPO. The funds in the middle — neither early enough nor full enough — are the hardest to compete from. David quotes Endowment Eddie: interest in big funds is founder-driven, not LP-driven — founders want the brand that reduces outcome risk, scales, and brings customers and hires; LPs just follow along slowly.

— Ram / David George
21:43

AI makes it harder to tell what real traction is

Ram says AI makes their job harder than ever: rounds are bigger and faster, and traction in the industry has become confusing. A company comes out of an accelerator, hits $5 million ARR in a month, but has no renewal cycle yet, and raises at a huge multiple — and these companies may be in the same cohort singing to each other, or it isn't even ARR but some month multiplied by 12. Only one in nine such companies is genuinely special. So judging real traction can't rely on financial analysis; it requires understanding customers and talking to them directly. He cites Harvey: commercially strong early, but actual usage wasn't high, until reasoning models arrived and usage and engagement flipped completely.

— Ram
26:56

GP and LP incentives point in exactly opposite directions

Ram says a GP gets fired for missing the next Facebook, the next Uber — that's an ‘error of omission,’ and it's fireable. But an LP only gets fired for backing the wrong thing; not investing costs them nothing. So if you missed the frontier models but are roughly at benchmark, or even slightly below, your job is safe. For many LPs the upside itself isn't that attractive, so the ‘you'll miss the next generation’ pitch doesn't land with them. Ram thinks what LPs should actually do is three things — access, selection, sizing: if only 20 of 3,000 firms perform, you should concentrate in those 15 to 20. When he sees someone allocated across 50, 60, 70 venture funds, he finds it hard to believe that portfolio beats the average.

— Ram
38:20

Bolting AI onto a company doesn't make it Amazon

Ram punctures AI-era private equity in one line: putting Sears on a website doesn't make it Amazon — you have to build Amazon's logistics from scratch. He gives the counterexample: a company's first move is AI customer service, because it's the easiest fruit to pick, but without rebuilding around the workflow, customers used to dealing with humans churn fast, and every point of NPS lost ties directly to revenue decline, which stacked on debt spirals downward. Intercom is the positive case: bring the founder back, rebuild the entire business, build AI-native products, then sell it — that's an almost ‘suicidal’ transformation of the existing business, extremely hard to pull off in private equity.

— Ram

In their own words · checked verbatim

But right now, especially with the labs, for the first time in my career, you can take capital and throw it at a company, and it compounds their advantage.

David George3:04

We've reached $100 billion in revenue in AI. It took SaaS 15 years to get to the same point. AI did that in four years.

Ram4:04

We've looked at the data of 3,000 venture capital firms in the US. Only 20 have achieved consistent 3x net returns over the last two decades.

a GP can get fired for missing out, you know, the next, you know, Facebook, the next Uber, right? Like that is the error of omission and like that is fireable. But LPs on the flip side only get fired if you invest into a mentor.

just because you put Sears on a website, didn't make it Amazon, right? You have, to have the benefit of building Amazon from the studs logistically to make it Amazon. It's not just the website.

We are nowhere on robotics, but I think robotics is going to be bigger than the language stuff. And I think it's going to happen in the next 10 years.

David George45:28

the bottleneck in AI today is not demand. It's on the supply side. So you got energy, the grid data center, then you got chips, then you got frontier models and apps.

Figures

US venture firms delivering 3x net returns for twenty straight yearsOnly 20 out of 3,00011:14
Revenue scale AI has reached$100 billion (SaaS took 15 years, AI took 4)4:04
US economy's labor spend relative to software spendAbout 40x6:07
Average venture net return (Cambridge data, past decade)1 to 2x12:14
Median US enterprise AI spend$12 per employee per month36:14
Top 1% of AI spend in the dataset$7,000 per employee per month36:14
Healthcare as a share of US GDP18%45:28
2021-2022 software LBO deal volume$200 to $300 billion, of which over $200 billion was extracted38:20

Glossary

power law
The distribution rule in venture where a few winners capture the vast majority of returns.
death of the middle
Mid-sized funds that are neither early enough nor full enough get squeezed from both ends.
TPPI
Total Paid-in to Paid-in, a multiple measuring what a fund has actually distributed to LPs.
vintage
The year a fund was raised, used to compare performance across the same period.
ARR
Annual Recurring Revenue, subscription revenue annualized.
LBO
Leveraged Buyout, acquiring a company with a large amount of debt.

How to listen

Who it's for

LPs doing asset allocation, GPs out fundraising, and founders who want to understand how AI is changing venture's return structure.

Skip

The opening 0:00-2:04 is show promos and guest introductions, skippable; around 4:04 there's some chit-chat about tenth-anniversary merch.