> ## Content Index > Fetch the complete content index at: https://www.earningsdriver.com/llms.txt > Use this file to discover other available public pages before exploring further. # Nvidia Sells Chips. It Earns on Being Hard to Leave. - URL: https://www.earningsdriver.com/nvidia-sells-chips-it-earns-on-the-lock-in/ - Published: 2026-09-23T01:54:22.000Z - Updated: 2026-09-27T14:55:53.000Z - Author: Heejung Chang **In one line:** Nvidia sells GPUs, but what protects its profit is CUDA — the software lock-in that makes them hard to leave. Watch the gross margin, not the headline revenue. **At a glance** - **Driver:** pricing power, protected by the CUDA software lock-in - **The one number:** gross margin (Q2 FY2027: 75.0%; Q3 FY2027 guidance: 74.0% ± 0.5 pt) - **Biggest risk:** the largest customers building their own chips, especially for inference - **Reports:** roughly late February, May, August, and November On August 26, Nvidia reported another record quarter: revenue of $96.2 billion, up 106% from a year earlier (Q2 FY2027, ended July 2026). The argument only grew louder: is the demand real, or the top of a bubble? The story everyone tells is simple. AI needs chips, and Nvidia makes the chips. That explains the **revenue**, most of it **$89 billion of data-center sales**. It does not explain the **profit**: a **gross margin of 75%**, a level companies that sell hardware rarely reach. Something protects that price. That something is the real business. ## What it sells vs. what it earns on Nvidia owns no chip factory. The physical chips are built by someone else — mostly TSMC in Taiwan. Nvidia keeps the design, the software, and the customer relationships: it is a **fabless designer**. It still sells the chip — but what it charges a premium for is *the design and the ecosystem around it*. A raw chip can be copied, and a rival can even build a faster one. What cannot be copied cheaply is **CUDA** — the software layer Nvidia has built over nearly two decades, that its customers have written years of code against. Around six million developers work in it, and the leading AI frameworks were tuned to it first. Switch to a rival's chip and you rewrite code, retrain engineers, and abandon years of tooling. That **switching cost** is what lets Nvidia hold its price while everyone else competes on specs. ## What moves the profit Nvidia reshapes the basic profit line (**price × quantity, minus cost**) into its own: > **Profit ≈ chips sold × (price − what each chip costs to build) − design R&D** The handful of things that move it: - **Price per chip.** The design-and-software premium, protected by the CUDA lock-in, by staying the performance leader, and for now by scarce supply. Not disclosed. Like volume and build cost below, it shows up in data-center revenue and the gross margin. - **Volume shipped.** Huge, but for now capped by how much TSMC can build and package, so Nvidia allocates. - **Build cost per chip.** The foundry, the memory and the packaging. It is the one large cost on an otherwise asset-light base. - **Design R&D.** The price of staying ahead each generation. ## Reading the earnings: what to watch, and how Nvidia reports quarterly: the results release, the 10-Q/10-K and the earnings call, all at investor.nvidia.com. Sources that work for any company are gathered in [Where to find the numbers](https://www.earningsdriver.com/numbers/). - **Data-center revenue and its growth from the prior quarter.** This is the main gauge of the AI boom. *Read it:* accelerating growth means the boom is intact. Slowing growth is the first warning sign, and it shows long before annual growth looks weak. It can mean tight supply, new limits on sales to China, or cooling demand, and management usually says which on the call. *Latest:* $89.0 billion in Q2 FY2027, up 18% from the prior quarter and 117% from a year earlier. - **Gross margin and next quarter's margin guidance.** This is where pricing power shows. *Read it:* holding in the low-to-mid 70s means pricing power is intact. Scarce supply helps today; the real test of CUDA is whether the margin holds once supply catches up. Drifting toward the 60s signals discounting, a lower-margin product ramp or competition. Rising volume can hide a price cut in revenue, but not in the margin. One-off charges such as inventory write-downs are flagged here. *Latest:* 75.0% in Q2 FY2027, with Q3 guided to 74.0% ± 0.5 point. - **Next-quarter revenue guidance.** Nvidia gives an explicit forecast, and the stock moves on guidance against expectations, not on the quarter just reported. *Read it:* a guide below consensus, even after a beat, is the real signal. *Latest:* Q3 FY2027 guidance is $108.0 billion ± 2%. - **Purchase commitments and inventory** (balance-sheet notes). This is the supply bet. *Read it:* rising commitments mean confidence. Inventory climbing while revenue slows is the classic digestion signal, the pattern of 2022\. *Latest:* inventory was $31.6 billion on July 26, 2026, up from $21.4 billion in January. - **Customer concentration** (10-Q/10-K). *Read it:* when a few buyers make up a large share of sales and receivables, and those buyers are also building their own chips, the price is at structural risk. *Latest:* on July 26, 2026, five direct customers accounted for 70% of what Nvidia was owed (22%, 14%, 13%, 11% and 10% of accounts receivable). ## Leading indicators (outside Nvidia's filings) These often move before Nvidia's own numbers — the earliest read on demand and supply. - **Hyperscaler capital spending** — Microsoft, Alphabet, Amazon, and Meta; their AI capex largely becomes Nvidia's revenue, though a growing share goes to their own chips, and their guidance moves a quarter or two ahead. *(demand)* - *When:* quarterly, with each company's earnings (a tight window in late Jan / Apr / Jul / Oct). *Where:* their results releases and calls — the capex line in the cash-flow statement, and capex guidance on the call. - **HBM memory makers** — SK hynix, Micron, Samsung; the high-bandwidth memory co-packaged with each Nvidia AI chip. Whether it is sold out, and how fast they add capacity, gates supply. *(supply)* - *When:* quarterly earnings, plus interim capacity and pricing commentary. *Where:* their earnings calls; memory-market trackers (e.g., TrendForce). - **TSMC's advanced-packaging (CoWoS) capacity** — a physical ceiling on how many AI chips can actually ship. *(supply)* - *When:* monthly sales (around the 10th of each month), plus quarterly earnings and capex guidance. *Where:* TSMC's monthly revenue report and quarterly call / IR. - **GPU cloud rental prices and lead times** — a near-real-time thermometer: scarce and rising means demand is hot; softening prices are the first sign it is cooling. *(demand)* - *When:* continuous (moves week to week). *Where:* GPU cloud providers' pricing pages (the big clouds, plus specialists like CoreWeave and Lambda) and GPU-market trackers. - **Power and data-center availability** — a growing constraint on how fast chips can be put to work; grid-connection queues and the pace of data-center construction cap how fast chips can actually be deployed. *(deployment)* - *When:* quarterly and slower. *Where:* data-center REIT earnings (Equinix, Digital Realty), grid-operator interconnection-queue data, and industry research. ## What Wall Street is asking On every earnings call the same handful of questions come up. What each is really probing: 1. **“How much of this demand is real consumption, not build-ahead?”** — are the chips being put to work now, or bought early and stockpiled to lock up scarce supply? Both look identical in this quarter's revenue, but they split next: chips in real use get reordered; a stockpile pauses new orders until it is worked off. That pause is the sudden drop Nvidia hit in the 2022 bust. 2. **“Where does gross margin go from here?”** — can Nvidia keep charging its premium? Margin slips when it has to discount against a credible rival, or when a new chip line ramps at lower early margins. A falling margin is the moat leaking, and a pricing problem shows there before it reaches revenue. 3. **“How are you holding inference as customers build their own silicon?”** — AI work splits in two: *training* a model (building it) and *inference* (running it, over and over). Nvidia dominates training, where CUDA runs deepest; the cloud giants' own chips (Google's TPU, Amazon's Trainium and Inferentia) are used especially for inference, a large, growing and repetitive volume. The question asks how fast that in-house silicon is taking the part Nvidia is most exposed on. 4. **“Is supply still the constraint — packaging and memory (CoWoS, HBM)?”** — is Nvidia selling everything it can make, held back only by supply, or has demand itself softened? If supply is the cap, there is a backlog of unmet orders — a good problem. If demand is the cap, growth is topping out. 5. **“How much revenue is exposed to China and export controls?”** — US rules restrict which chips Nvidia may sell to China. The question sizes how much revenue those rules put off-limits, and what a further tightening — or a carve-out — would remove or add overnight. Nvidia's Q3 FY2027 outlook assumes no data-center compute revenue from China. Most of these surface in the **margin** before they surface in the revenue. ## A note on valuation What decides Nvidia's stock is less this quarter's profit than its **durability** — how many years the 70%-plus margins hold before custom chips or a rival erode them. ## The shape Nvidia is a **chokepoint**: most AI models are trained on its chips, and the switching cost keeps them there. That is what lets it set its price rather than take it. The edge lives in *being hard to replace*, not in the chip alone. It is the same skeleton as any company that owns the one thing others cannot easily copy. *(See:* [*TSMC Builds Other People's Chips. It Earns on Being Almost Impossible to Replace.*](https://www.earningsdriver.com/tsmc-builds-other-peoples-chips-it-earns-on-being-almost-impossible-to-replace/)*)* **New to reading a business this way?** If parts of this were hard to follow — or you want the basic ideas, and how profit is built across many different industries, not just this one — that is what the book works through, one industry at a time: [**Earnings Driver**](https://www.earningsdriver.com/book). --- *Sources: Nvidia Q2 FY2027 results (quarter ended July 26, 2026, reported August 26, 2026) and Form 10-Q; Nvidia FY2025 Form 10-K (CUDA developers). All at investor.nvidia.com.* *Educational only — not investment advice. Figures as of the periods noted.*