Hyperscaler exposure and the concentration question

1. The disclosure itself — what share of revenue rides on the top-N customers, and how to interpret the number.

   01

   Customer concentration

   ○Mark read

   How much of revenue comes from a few large customers. In tech, often phrased as 'hyperscaler exposure' for chip and infra suppliers.

   Customer concentration measures how dependent a company is on a small number of buyers. Broadcom and Nvidia disclose top-customer concentration in the 30–45% range for their AI silicon businesses — meaningful, and the bull case carries a single-customer-deceleration risk that diversified businesses don't.

   Concentration is not the same as fragility. A multi-year contract with a top-three hyperscaler is concentrated revenue, but it's also funded backlog. Read concentration alongside RPO and the customer's own capex disclosures.

   View in glossary →

2. Concentrated revenue with a fat backlog reads very differently from concentrated revenue without one.

   02

   RPO (remaining performance obligations)

   ○Mark read

   Contracted future revenue the company hasn't recognised yet. The forward look that backlog metrics give into cloud and enterprise-software trajectory.

   RPO — remaining performance obligations — is contracted revenue the company has signed but not yet recognised. For hyperscalers and enterprise SaaS, RPO is a stronger near-term growth read than current revenue: it tells you what's locked in, not what's already booked.

   Compare RPO growth to revenue growth. RPO meaningfully outrunning revenue means the next several quarters of acceleration are funded. RPO growth that lags revenue is the early warning of demand cooling.

   View in glossary →

3. For SaaS, the corresponding quality read on the existing customer base.

   03

   Net revenue retention

   ○Mark read

   Dollar-weighted expansion in an existing customer base over a year, net of churn. The SaaS-quality metric.

   Net revenue retention (NRR) is how much revenue a cohort of customers from a year ago is spending today, including upsells and downgrades, net of churn. 110% NRR means existing customers are spending 10% more than a year ago even after losses; 90% means the base is shrinking.

   A SaaS business with consumption-based pricing (Snowflake, Datadog) prints elevated NRR when usage grows — that's the bull case in one number. Look at the trend: NRR collapsing from 160% to 120% over two years is not a disaster, but it is the end of a hyper-growth phase.

   View in glossary →

4. And the other side of the trade — when the hyperscaler is buying GPUs, its own capex line is the AI suppliers' revenue.

   04

   Capex intensity

   ○Mark read

   Capital expenditure as a percentage of revenue. How much of every sales dollar the company is sinking into long-lived assets — data centres, fabs, factories.

   Capex intensity is capital expenditure divided by revenue. For platform tech in the AI build cycle it's the single most-watched line: hyperscalers running capex intensity in the mid-20s%+ (vs. historical 12–15%) are betting that future revenue justifies today's spend.

   High capex intensity is not inherently bad. The right way to read it is alongside the backlog (RPO) and the customer base — capex pulling forward to meet contracted demand reads very differently from capex pulled forward on hope.

   View in glossary →

Concentration is not the same as fragility. A multi-year contract with a top-three hyperscaler is concentrated revenue, but it is also funded backlog with a counterparty whose own capex disclosures you can read. The bear case isn't 'concentration' — it's 'concentration without contracted forward demand.' The numbers above tell you which one you're looking at.