Is the 70% growth target still conservative? NVIDIA investor meeting "reveals all": Business scale may double by FY28 if not restricted by supply!

Nvidia’s growth outlook for the coming years may be far more aggressive than the market anticipates.

According to Wind Chaser Trading Desk, JP Morgan noted in its latest report that at a recent investors’ conference, Nvidia’s management made it clear that the FY28 YoY growth framework of 70% is not the demand-side ceiling; if not for supply constraints, business growth could have exceeded 100%. This means that the core issue currently limiting Nvidia’s growth has shifted from “whether demand can be sustained” to “whether capacity can keep up.”

Even more notable is that AI demand itself is still expanding rapidly and that the structure of this demand is changing. About 18 months ago, Nvidia’s training and inference revenues each contributed about half; now, inference revenue has surpassed training and is expected to continue rising in proportion. At the same time, revenue from AI computing infrastructure contributed by emerging cloud service providers has already exceeded 50%, with growth momentum shifting from traditional hyperscale cloud vendors to a broader AI computing ecosystem.

On the supply side, whether Nvidia can achieve greater growth hinges on key factors. Management identified advanced wafers and memory as the two most critical bottlenecks, and they continue to coordinate with TSMC, Micron, SK Hynix, and Samsung to expand supply. Especially given the ongoing tightness in HBM, if the supply of key components improves, Nvidia’s previously pent-up order demand due to capacity limitations could be further released.

The change in customer structure is also alleviating concerns about Nvidia’s overreliance on a few major customers for growth. OpenAI and Anthropic currently account for about 20% of Nvidia’s business in terminal consumption terms, and by FY28 this could rise to around 25%. However, revenue from emerging cloud service providers for AI computing infrastructure now accounts for more than 50%.

With sustained demand expansion, higher proportion of inference, and further diversification of customers, Nvidia’s growth story is shifting from a simple “training compute cycle” to a broader AI infrastructure cycle.

70% Growth for FY28 Is Not the Ceiling—Supply Is the Largest Constraint

The report points out that Toshiya Hari, Nvidia’s VP of Investor Relations and Strategic Finance, said that the company’s proposed 70% YoY growth framework for FY28 is not driven by a single customer or business, but is propelled by joint demand from hyperscale cloud providers, emerging cloud service providers, AI labs, sovereign AI, enterprises, and on-premise deployments.

One important reason for providing a multi-year growth framework in advance is the significant gap between market consensus and the company’s internal assessment. If this gap persists, it can hinder capacity planning by supply chain partners.

Of note, Hari stated directly that if there were no supply constraints, Nvidia’s business growth could have exceeded 100%.

In other words, the 70% growth target is more like the company’s publicly confirmed baseline under current supply conditions, not the true demand-side ceiling. As capacity continues to be released, Nvidia’s actual growth could significantly surpass this figure.

Inference Revenue Has Surpassed Training—AI Demand Moves from “Card Purchases” to Continuous Operation

The income structure of training versus inference was also a key topic at this meeting.

Because Nvidia GPUs have strong workload-switching capabilities—for example, Grace Blackwell can be used for both model training and rapid inference—the company finds it difficult to split these two segments precisely.

However, management provided an important reference point: about 18 months ago, training and inference revenues were each around 50%; now, inference revenue has surpassed training, and the gap is expected to continue widening.

This means Nvidia’s demand structure is changing. Training remains a key driver for AI infrastructure expansion, but as model sizes increase and AI applications proliferate, inference is becoming a more persistent source of compute demand and is expected to further enhance Nvidia’s revenue stability.

Advanced Wafers and HBM Remain Two Major Supply Bottlenecks

Despite strong demand, Nvidia’s biggest current challenge remains its supply chain.

Hari specifically emphasized that the main constraints to meeting next year’s demand are advanced wafers and memory. For advanced wafers, Nvidia mainly relies on TSMC; HBM and other high-bandwidth memory involves Micron, SK Hynix, and Samsung.

Nvidia continues to communicate with TSMC and the three major memory suppliers, focusing on boosting key component supply capabilities.

This also means that as AI demand continues to rise, HBM capacity expansion remains a key prerequisite for Nvidia’s growth potential to be realized. For HBM suppliers like Micron and SK Hynix, visibility of Nvidia-driven demand remains high.

Customers No Longer Rely Heavily on Hyperscale Cloud Providers—Emerging Cloud Providers Now Contribute Over Half

Nvidia’s customer structure is also shifting.

Hari revealed that OpenAI and Anthropic currently account for about 20% of Nvidia’s business in terminal consumption terms, potentially rising to about 25% by FY28. However, since the two primarily acquire compute power through cloud and emerging service providers, this figure does not equate to Nvidia’s direct customer revenue share.

What’s truly noteworthy is that revenue from AI Computing Infrastructure (ACIE) contributed by emerging cloud service providers has already surpassed 50%.

This means Nvidia’s growth sources are expanding from a handful of hyperscale cloud providers toward emerging cloud providers, model companies, and enterprise clients. The continued diversification of customers and compute demand has also, to some extent, reduced risk associated with revenue concentration among a few customers.

Open Source vs. Closed Source Is Not an “Either-Or”—Lower Model Costs Actually Stimulate Compute Demand

On the development of open and closed source models, Nvidia’s management maintained their previous position: the two modes will coexist long-term; technological progress in AI does not mean one will replace the other.

Nvidia itself uses both OpenAI, Claude and other closed-source models, as well as hybrid approaches with open and closed source models in key workloads such as chip design.

At the same time, management indicated that model builders’ gross margins are improving. As Nvidia GPUs continue to advance, the computing cost per token keeps dropping, thus improving business economics for model companies.

This could generate a new demand loop: Lower computing costs → improved model company profitability → accelerated AI application adoption → inference demand growth → further compute purchases.

Is the Financing Model Sustainable? Nvidia Directly Addresses “Circular Financing” Concerns

On the market’s concerns regarding financing, management introduced three main arrangements: revenue-sharing agreements with some emerging cloud providers, the PORTS-Pike data center park plan, and the $500 billion private capital platform involving Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR, and others.

The core mechanism of the revenue-sharing agreement is: Nvidia helps lock in the base compute price, and when compute leasing rates exceed the agreed threshold, the company can share upside revenue. This way, in addition to direct hardware sales, Nvidia has the opportunity to achieve recurring revenue from compute infrastructure operations.

Addressing market worries about “circular financing,” management responded that the scale and ceiling of such financing are both controlled and fundamentally backed by strong end demand, ecosystem returns, and the creditworthiness of ultimate compute buyers.

Judging from the information disclosed at this conference, Nvidia’s real challenge is no longer whether there is demand, but whether it can quickly turn demand into supply. If bottlenecks such as wafers and HBM continue to ease, the previous 70% FY28 growth framework may indeed have room for significant upward revision.