Nvidia (NVDA) — 2027Q2 FY2027 Earnings Call Analysis

Guided 70% A Year Out, A Quarter Of It On Their Own Balance Sheet

They guided a full year out for the first time, then told you roughly a quarter of it depends on NVIDIA credit support.

Thesis: NVDA owns the bottleneck across the whole chain: foundry, CoWoS, HBM, and now power and shell. Demand runs near 100% against 70% supply, and management is pricing through the memory squeeze because customers have no alternative. The one genuine flaw is that roughly a quarter of FY2028 routes through NVIDIA's own balance sheet. Growth risk is capped and quantified; the margin trough is guided, dated and recovering by Q1. Own it into the price realization.

Verdict: LONG — Conviction: MEDIUM

Catalyst: Q1 FY2028 executed price increases landing in the P&L, Vera Rubin moving to 20% of DC revenue in Q3, and hyperscaler CapEx re-accelerating in Q4 as stated by management.

Key Risk: Colette: AI labs leveraging NVIDIA's balance sheet contribute 'roughly a quarter of our business next year' with credit enhancement on nearly 2GW. If that demand doesn't convert to cash, or no second buyer exists at the same price, backlog quality breaks.

The Tell: Colette volunteered, unprompted: 'we expect demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year,' then pre-empted the obvious objection with 'we know some will call this circular financing.' CFO quantifying her own credit exposure before anyone asked is the tell.

Detected Patterns

Beat and Raise Machine: Revenue $96B more than doubled YoY with a fourth consecutive acceleration, Q3 guided to $108B (+13% seq), and a first-ever full-year guide of ~70% growth. Every number moved up.
Pricing Power Signal: Memory costs 'exceeded our prior expectations and are headed even higher,' and management is raising prices in Q1 to offset while still guiding 70% growth. Customers pay because compute is fully utilized across every cloud they serve.
Backlog Fortress: Cloud industry backlog >$2T, top-five hyperscaler CapEx $800B (2026) to $1.3T (2027), AWS committing 2M additional GPUs through FY2029, OpenAI 12GW through 2030.
Rising Customer Switching Costs: Revenue per gigawatt climbs $18B (Hopper) to $25B (Blackwell) to $40B (Vera Rubin). Customers buy GPU, CPU, NVLink, three networking types, systems, algorithms and software as one stack.
Backlog Quality: Colette states AI labs on NVIDIA's balance sheet are ~a quarter of FY2028, with credit enhancement on nearly 2GW. She pre-empts the criticism: 'we know some will call this circular financing.'
Capacity Ceiling: Supply remains a bottleneck 'at least through the end of fiscal year 2028' at 70% of demand. Entire supply chain running flat out, yields still to improve. Growth is capped by capacity, not orders.
High Utilization as Ceiling: 'NVIDIA Compute is fully utilized across every cloud we serve' is presented as demand proof. It also means zero incremental volume without new capacity coming online.
Export Control Revenue Loss: China is under 1% of data center revenue, dilutive to GM, and explicitly zero in the forward outlook. No replacement market identified.

Friction Level: MODERATE_FRICTION — Both sides accept the supply constraint and the 70% FY2028 guide. Disagreement is whether the ~25% of FY2028 backed by NVIDIA's balance sheet converts to cash, and whether the 300bps GM compression is actually offset by Q1 price increases customers haven't signed.

Report not found

The report data is no longer available. Please return to the archive.

Guided 70% A Year Out, A Quarter Of It On Their Own Balance Sheet | Silicon Signal