Is customer concentration a genuine risk for a fast-scaling company?

Brendan Foody · Sep 15, 2025

High customer-revenue concentration is acceptable when the concentrated customers are the best customers in the world; building a business that creates enormous value for them matters more than diversification.

Nvidia has similar concentration and is worth trillions of dollars — empirical evidence that leaning into a handful of top customers works.

Scope: Mercor's concentration breakdown is similar to Nvidia's

19:54 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Jonathan Siddharth · Dec 1, 2025

Revenue concentration among a handful of frontier labs is not a serious risk in this market

NVIDIA reportedly gets ~39% of revenue from two clients and ~50% from four; the big eight will spend enormous sums on compute, energy and data, with Stargate alone around $100B a year

Scope: acknowledges the concentration level is 'a little weird' and that things could change

37:54 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Anastasios Angelopoulos · Aug 3, 2026

Silicon Valley investors are far too averse to revenue concentration in startups.

Highly successful companies like TSMC also have concentrated revenue.

0:00 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Anastasios Angelopoulos · Aug 3, 2026

Silicon Valley investors have become far too squeamish about revenue concentration; it is not a valid criticism of the data providers

Many hundred-billion-dollar public businesses are highly revenue concentrated — TSMC, two-customer businesses, government suppliers like Anduril — and they do great

43:49 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Harry Stebbings · Aug 3, 2026

Venture investors have become wusses about revenue concentration and should instead embrace it

44:45 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Andrew Feldman · Mar 24, 2025

Revenue concentration in the G42 deal is both a strength and a weakness, because being a strategic partner is a learned skill that becomes replicable across other large partners

You catch three large customers by catching one first; having worked at being a strategic partner, it's now a muscle they can replicate with a dozen other companies

Scope: both strength and weakness

40:45 20VC: AI Chip Wars: How Cerebras Plans to Topple NVIDIA's Dominance | Why We Have Not Reached Scaling Laws in AI | What Happens to the Cost of Inference | How We Underestimate China and Shouldn't Sell To Them with Andrew Feldman

Andrew Feldman · Mar 24, 2025

Serving a partner at G42's scale has hardened Cerebras' software, manufacturing and supply chain and made it materially better prepared to serve hyperscalers, sovereigns and other massive partners

They deployed tens of exaflops, scaled manufacturing 2x and 5x, prepared supply chain partners, and learned from mistakes in a first big Middle East relationship

41:13 20VC: AI Chip Wars: How Cerebras Plans to Topple NVIDIA's Dominance | Why We Have Not Reached Scaling Laws in AI | What Happens to the Cost of Inference | How We Underestimate China and Shouldn't Sell To Them with Andrew Feldman

Andrew Feldman · May 26, 2026

Customer concentration in a large deal is not the risk investors think it is; the only way to win many large customers is to win the first one and build the organizational and supply-chain muscle

Cerebras went from a $1B G42 deal to a $20B+ deal with a different customer in a year; winning and keeping one large customer builds the capability to win the next

22:08 20VC: Cerebras CEO on the Future of Data Centres, Token Costs and Memory | We are Not in an Infra Bubble & Dario Got a Bad Deal with Elon for Compute | Should US Companies Sell to China & Why Most Layoffs are AI Washed with Andrew Feldman

Matthew Steckman · Mar 23, 2026

The hardest part of the government business is extreme revenue concentration — only about 20 of roughly 600 annual contracts are materially sized — so every one of those deals is existential risk and the real challenge is what you do in the gaps between them.

Large wins are few and far between, so each is business-defining whether it goes right or wrong.

15:19 20VC: Inside Anduril's $20BN Army Contract & Why Anduril Must Go Public | Why 99% of Drone Companies Will Die | Why There is Never an Ethical Question of How Anduril Products are Used with Matthew Steckman, President @ Anduril

Matthew Steckman · Mar 23, 2026

The defense companies VCs fund that will most go to shit are those whose business depends on capturing a single large program.

By the nature of how defense acquisitions occur you have to be wide; if there's only one program and you don't win it, you're not a good company.

47:13 20VC: Inside Anduril's $20BN Army Contract & Why Anduril Must Go Public | Why 99% of Drone Companies Will Die | Why There is Never an Ethical Question of How Anduril Products are Used with Matthew Steckman, President @ Anduril

Jack Zhang · May 27, 2025

Airwallex's jump from zero to $1BN in transaction volume in nine months came from riding two hypergrowth customers' global expansion rather than from broad customer acquisition

Onboarding a massive tuition-payments company and Shein in January 2018 meant their volume scaled with those customers' own global growth

49:39 Riding hypergrowth customers concentration drives explosive scaling

Harry Stebbings · Nov 29, 2023

Startups should avoid customer concentration in a single segment such as high-growth startups, and narrowing focus increases quality

A macro recession will hit startup spending and funding, so a customer base concentrated there is fragile

13:11 Segment concentration in cyclical customers like startups is a fragile risk

Roman Chernin · Jun 8, 2026

Nebius's long-term strategy should be the most diversified customer portfolio possible rather than dependence on a handful of hyperscale customers

Hyperscale customers like Meta and Microsoft bring their own full software stack and need only physical infrastructure, so there is very little additional value a provider can add above raw capacity, and that customer population is tiny

20:32 Hyperscale concentration leaves no room to add value

Matan Grinberg · Jun 13, 2026

Extreme customer concentration (~90% of revenue from one customer) is bad for investors because of the added risk, though such businesses can reach a steady state

There is a permanent sword of Damocles over the business — the risk never goes away even if the revenue is stable

69:02 Permanent sword of damocles risk even at steady state