ByteDance brought in nearly 75% of Nscale’s 2025 sales, the FT reports

Nscale's prospectus for its planned New York listing disclosed significant customer concentration risk, stating that a substantial portion of revenue comes from a limited number of clients. The company warned that losing, or seeing materially reduced spending from, one or several top customers would adversely affect its business, creating a key risk factor for prospective investors.

Analysis

Customer concentration is more consequential for an AI-infrastructure issuer than for a typical software IPO because the associated capacity is largely fixed-cost and often financed ahead of utilization. A single customer reducing GPU reservations can create a double hit: lower revenue plus under-absorbed power, depreciation and financing costs. The market should therefore value Nscale on contracted backlog quality, minimum-spend protections, contract duration and customer credit—not headline revenue growth; absent those disclosures, a premium to more diversified compute peers is difficult to justify.

The second-order risk is bargaining power. Large AI labs and hyperscalers can use alternative GPU-cloud providers, direct OEM allocations and in-house data-center builds to renegotiate pricing at renewal, while a concentrated supplier has limited ability to redeploy specialized capacity rapidly. That dynamic can turn apparent backlog into lower-margin revenue over the next 12-24 months, particularly if GPU supply loosens or inference shifts toward lower-cost hardware.

Near term, concentration alone is unlikely to be a durable catalyst without evidence of a delayed deployment, lower committed utilization or weaker renewal economics. The contrarian case is that a small set of investment-grade customers with take-or-pay contracts can make concentration economically safer than a broad base of uncommitted demand. The key diligence item is whether top-customer revenue is secured by enforceable minimum commitments extending beyond the debt amortization profile of the underlying capacity.

Key Decisions for Investors

• Do not participate in the Nscale IPO at an undifferentiated AI-infrastructure multiple until the prospectus quantifies top-customer concentration, take-or-pay coverage, contract tenor and capacity-level debt obligations. Reassess after initial results or the first post-listing lockup window, when utilization and pricing disclosures can be tested.
• Set an IPO watch trigger: consider a tactical short or put structure only if the stock trades above GPU-cloud peers on EV/revenue while a top customer exceeds roughly 25-30% of revenue without minimum-spend protection. Thesis horizon: 3-9 months; cover on verified multi-customer backlog diversification or sustained positive operating cash flow.
• For existing AI-compute exposure, favor diversified or vertically integrated alternatives over concentrated pure-play capacity providers; use CRWV and NBIS as relative-value monitoring comparables rather than assuming all GPU-cloud revenue deserves the same multiple. A narrowing valuation gap would require evidence that Nscale's contracted revenue converts to cash after power, depreciation and interest expense.
• Watch quarterly indicators rather than headline bookings: revenue per deployed GPU, committed versus utilized capacity, gross-margin progression, customer prepayments and debt-service coverage. A material guidance cut, receivables build, or renewal repricing would falsify the bullish contracted-demand narrative quickly.