Summary

  • Four direct customers represented 29%, 25%, 15% and 13% of Astera Labs' fiscal-Q2 2026 revenue: 82% in aggregate. Five direct customers had represented 90% in Q1.
  • That eight-point decline cannot establish end-customer diversification. Astera Labs says certain direct customers manufacture for end customers, which can move production among those partners without changing underlying demand.
  • The latest annual look-through was more concentrated: one end customer supplied more than 70% of 2025 revenue and the top three approximately 86%. Those historical annual figures are not a current-quarter estimate.
  • Revenue doubled year on year and Q3 guidance implies another sharp sequential rise. Receivables and inventory also expanded, making product conversion, payment and cash separate tests from the identity of the invoice recipient.

Astera Labs invoiced four customers for 82% of its fiscal second-quarter revenue. The natural reading is that four buyers supplied most of the demand. It is also the reading the filing explicitly warns against.

The four anonymised direct accounts supplied 29%, 25%, 15% and 13% of the US$392.4 million quarter. In the preceding quarter, five accounts at or above the disclosure threshold supplied 29%, 21%, 16%, 12% and 12%, or 90% together.

Eight percentage points disappeared from the disclosed concentration and one letter fell below 10%. That could be genuine broadening. It could also be the same economic demand passing through a different factory. The Q2 Form 10-Q says certain customers in the table are manufacturing partners buying for end customers. Those end customers may shift production volumes among partners from one period to the next.

The direct-customer table therefore answers a precise accounting question: which legal counterparties did Astera Labs invoice? It does not, by itself, answer the strategic question: how many independent companies controlled the system designs and budgets behind those invoices?

The same demand can change letters

Imagine an end customer moving a server programme from one contract manufacturer to another. The first partner's purchases from Astera Labs fall and the second partner's rise. Customer A may lose share while Customer B gains it, even if the end customer's total deployment plan is unchanged.

The reverse is also possible. Several direct accounts can buy for one economic principal, making the invoicing table look broader than the demand that caused it. One direct account can also aggregate several programmes. The letters reveal neither case.

This is why the Q1-to-Q2 movement needs a neutral interpretation. Five disclosed direct customers and 90% became four and 82%. The filing does not allocate that eight-point change between new end demand, production reallocation and customers below the 10% threshold. Calling it diversification would claim information the table does not contain. Calling it entirely cosmetic would do the same.

The historical comparison reinforces the point. In Q2 2025, five direct accounts represented a disclosed 87%. Their letter shares changed markedly by Q2 2026. That movement can matter for shipping, credit and collection even when it says less about the ultimate buyer.

The annual report keeps a second ledger

Astera Labs' 2025 Form 10-K makes the distinction explicit. In the financial statements, a customer is the party directly invoiced, primarily an end customer's manufacturing partner or a distributor. Elsewhere, the company estimates end-customer demand using information attached to purchase orders.

That look-through produced a much more concentrated annual picture. One end customer represented more than 70% of 2025 revenue; the top three represented approximately 86%.

Those percentages should not be carried forward mechanically. They cover fiscal 2025, not Q2 2026, and the company does not publish the same end-customer bridge each quarter. Their value is definitional. They prove that the direct-account table and the economic-principal table can be materially different.

The end-customer universe is structurally narrow. Astera Labs focuses on hyperscalers, leading AI accelerator vendors and system manufacturers. A chip may be ordered by a manufacturing partner, incorporated into a platform for one of those end customers, and deployed in a data centre owned or operated by another party. Each layer is commercially real. Only one layer decides the architecture that creates Astera Labs' design slot.

Billing geography has the same identity problem

Q2 revenue by billing address was US$152.7 million in China, US$113.7 million in Singapore and US$105.8 million in Taiwan. Only US$5.5 million was billed to the United States, with US$14.7 million elsewhere.

That is not a map showing where the AI racks were installed or where their budgets were approved. Astera Labs says the geography is based on customer billing addresses and includes manufacturing partners and distributors.

The table is operationally useful. It shows where invoices, receivables and supply-chain counterparties are concentrated. It can illuminate trade, logistics and collection exposure. But turning 95% of Q2 billings in China, Singapore and Taiwan into a claim about final demand would repeat the direct-customer error at geographic scale.

The distinction matters because a single end customer can redesign its production network without redesigning its AI system. Billing may move across borders while the ultimate buyer, platform and deployment programme remain the same.

Three product families do not prove three principals

The current operating evidence is strong. Revenue grew 104% year on year and 27% sequentially. Astera Labs attributes the increase to more units across Aries, Scorpio and Taurus, alongside higher average selling prices from hardware modules and Scorpio mix.

The Q2 results release guides Q3 revenue to US$540 million to US$560 million. At the US$550 million midpoint, that would be another 40% sequential increase. Management expects the Scorpio X production ramp to make Scorpio the company's largest product family one quarter earlier than previously expected.

This broadening is economically meaningful. Aries retimers, Scorpio switches and Taurus signal conditioners solve different connectivity problems inside an AI system. More products can increase content per rack and protect a supplier if one component category matures.

It is not the same as broadening the group of economic principals. One hyperscaler programme can use several Astera Labs products. A new product family can deepen share of wallet at an existing end customer rather than add a new one.

The distinction cuts both ways. Concentrated end demand does not erase product success, and product success does not erase concentrated end demand. The right analysis keeps both facts visible.

Orders narrow the distance but do not remove it

The annual report says a design win occurs after customer validation and an order. It also says a design win does not guarantee sales. A customer's system plan can be delayed or cancelled, and master agreements generally do not require a minimum quantity.

That produces a sequence of control. The end customer selects a system architecture and expected volume. A manufacturing partner receives production allocation. It places a purchase order with Astera Labs. Astera Labs commits supply, ships a product and recognises net revenue. The direct customer then pays the receivable.

Each step is stronger evidence than the one before it, but control changes hands along the way. Astera Labs controls product development, partner qualification and shipment. The end customer controls architecture and much of the production plan. The manufacturing partner controls the direct order and payment path.

Customer warrants sit alongside that chain. The Q2 filing describes warrants whose vesting depends on purchases by an unnamed Holder and affiliates. Astera Labs recorded US$10.2 million in Q2 as a reduction of revenue related to the warrants. The note does not identify the Holder, and it provides no basis for mapping the Holder to Customer A-F. The mechanism is relevant as a customer-purchase incentive, not as an identity clue.

Working capital is the third ledger

Revenue identity and cash identity are also different. Net receivables rose from US$83.2 million at year-end to US$134.8 million in March and US$192.5 million in June. The Q2 sequential increase was US$57.7 million, or 42.8%.

Four direct accounts represented 69% of net receivables at June: 30%, 17%, 12% and 10%. Three accounts represented the same 69% aggregate at year-end, but with different letters and shares. The company recorded no material credit-loss allowance at either date.

This is not evidence of bad debt. It is evidence that rapid shipment creates a separate collection exposure at the directly invoiced layer. A manufacturing partner can be the credit counterparty even when an end customer caused the order.

Inventory almost doubled from US$59.0 million at year-end to US$113.8 million, with US$98.7 million in work in progress. Astera Labs says inventory increased because of per-unit costs and a build for anticipated demand.

First-half operating cash remained strong at US$162.3 million, up US$16.4 million. Yet net income increased by US$150.4 million year on year while operating-asset and liability changes were US$184.0 million less favourable. Receivables accounted for US$124.0 million of that change and inventory US$38.6 million.

Liquidity of US$1.3 billion rules out an immediate financing-stress story. It does not remove the need to watch whether an accelerated Scorpio ramp becomes collected cash without a persistent expansion of receivables and work in progress.

Sources