Nvidia has put a striking number behind the scale of the artificial intelligence infrastructure boom: the cloud industry’s backlog has now crossed $2 trillion, according to CFO Colette Kress during the company’s second-quarter fiscal 2027 earnings call. The figure represents the enormous pipeline of infrastructure demand building across major cloud providers as they race to add computing capacity for AI workloads.
The $2 trillion backlog helps explain why Nvidia expects its own revenue to grow by approximately 70% in fiscal 2028, despite saying that its outlook is supply constrained. Kress said customer forecasts point to Nvidia’s growth potentially doubling next year, but the company expects to deliver around 70% growth because supply remains a bottleneck. Nvidia CEO Jensen Huang later said revenue would be significantly higher without those constraints.
Nvidia CFO Highlights $2 Trillion Cloud Backlog
The $2 trillion figure was highlighted by Kress during the earnings call as evidence that demand for AI computing remains exceptionally strong.
She said the cloud industry now has more than $2 trillion in backlog, while capital expenditures by the five largest hyperscalers are expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027.
AI Infrastructure Numbers At A Glance
| Metric | Figure |
|---|---|
| Cloud industry backlog | >$2 trillion |
| Top-five hyperscaler CapEx, 2026 | ~$800 billion |
| Top-five hyperscaler CapEx, 2027 | ~$1.3 trillion |
| Nvidia Q2 FY2027 revenue | $96.2 billion |
| Nvidia Q2 Data Center revenue | $89 billion |
| Nvidia FY2028 revenue growth outlook | ~70% |
| Nvidia supply outlook | Constrained |
The scale of these figures suggests that AI infrastructure spending is moving from a short-term investment cycle toward a multi-year buildout.
What Does A $2 Trillion Backlog Mean?
A backlog represents demand and commitments that have yet to translate fully into completed infrastructure and revenue.
The figure should not be interpreted as $2 trillion of Nvidia orders.
Instead, Kress was referring to the broader cloud industry’s backlog, which reflects the infrastructure commitments and demand visibility of cloud providers.
That distinction is important.
Cloud Customer Demand
↓
AI Service Demand
↓
Hyperscaler Capacity Commitments
↓
Cloud Industry Backlog
↓
Data Center Construction
↓
AI Compute Deployment
↓
Nvidia GPU + System DemandNvidia benefits because a large portion of that infrastructure ultimately requires advanced accelerators, networking, CPUs and other components from the AI hardware ecosystem.
$2 Trillion Backlog Supports Nvidia’s 70% Growth Outlook
The backlog is particularly significant because Nvidia has provided an unusually strong forward-looking forecast.
Kress said Nvidia expects revenue to grow by approximately 70% in fiscal 2028, while emphasizing that this is a supply-constrained outlook.
Wall Street had been expecting roughly 44–45% growth, making Nvidia’s preliminary forecast substantially more bullish than the market’s previous assumptions.
Nvidia Growth Expectations
| Forecast | Revenue Growth |
|---|---|
| Wall Street FY2028 expectation | ~44–45% |
| Nvidia FY2028 preliminary outlook | ~70% |
| Unconstrained potential | Significantly higher |
Huang said Nvidia’s revenue would be significantly higher than 70% if the company were not constrained by supply.
Customer Demand Could Support Even Faster Growth
One of the most important comments from the earnings call was that Nvidia’s customers are forecasting growth that could effectively double next year.
Kress said customers’ forecasts point to Nvidia’s growth doubling, but the company expects to deliver approximately 70% because of supply limitations.
This creates an unusual situation in the semiconductor industry.
Nvidia is not struggling to find buyers.
It is struggling to produce enough computing capacity to satisfy potential demand.
Customer Forecasts
↓
Potential Growth
~100%
↓
Nvidia Supply Capacity
↓
Supply Bottlenecks
↓
Delivered Growth
~70%The difference between potential demand and deliverable supply is effectively the company’s current growth ceiling.
Nvidia’s Entire Supply Chain Is Under Pressure
Huang said “our entire supply chain is challenged”, describing a market in which suppliers are operating at full capacity.
The bottleneck extends beyond Nvidia’s own manufacturing operations.
Advanced memory, packaging, components, networking equipment and data-center infrastructure all have to scale simultaneously for Nvidia to ship complete AI systems.
Nvidia Supply Chain
| Component | Importance |
|---|---|
| GPUs | AI training and inference |
| High-bandwidth memory | Feeds AI accelerators |
| CPUs | AI server and agent workloads |
| Networking | Connects large GPU clusters |
| Advanced packaging | Integrates compute and memory |
| Power infrastructure | Enables data-center deployment |
| Data-center capacity | Houses AI systems |
The constraint therefore cannot be solved simply by manufacturing more GPUs.
Memory Has Become A Major Bottleneck
Nvidia’s CFO also highlighted extreme pricing conditions in memory.
Kress said memory price increases had exceeded Nvidia’s previous expectations and were expected to rise further into the following year.
Nvidia expects tighter memory supply to remain a bottleneck at least through the end of fiscal 2028.
Memory Pressure On Nvidia
AI Demand Surges
↓
More GPUs Required
↓
More HBM Required
↓
Memory Demand Rises
↓
Supply Tightens
↓
Memory Prices Increase
↓
Nvidia Margin PressureThe company expects gross margins to decline in the near term before recovering toward approximately 72–73% in fiscal 2028 as price increases take effect.
Hyperscalers Are Spending At An Extraordinary Pace
The $2 trillion backlog is accompanied by unprecedented capital expenditure.
Nvidia expects the five largest hyperscalers to spend nearly $800 billion in 2026 and approximately $1.3 trillion in 2027.
That would represent a roughly 62.5% increase between the two years.
Hyperscaler CapEx Growth
| Year | Estimated CapEx | Change |
|---|---|---|
| 2026 | ~$800 Bn | — |
| 2027 | ~$1.3 Tn | ~62.5% |
2026
~$800 Billion
↓
2027
~$1.3 Trillion
↓
~$500 Billion Additional SpendingSuch spending provides a powerful demand signal for the broader AI infrastructure supply chain.
More Compute Is Translating Into More Revenue
Kress emphasized that Nvidia’s hyperscale customers are seeing strong financial performance and that additional compute capacity is translating into higher revenues.
Nvidia’s Q2 hyperscale revenue reached $49 billion, up 13% sequentially.
The company’s argument is straightforward: when cloud providers add Nvidia computing capacity, they can sell more AI services to customers.
Hyperscale Revenue
| Metric | Q2 FY2027 |
|---|---|
| Hyperscale revenue | $49 Bn |
| Sequential growth | 13% |
| Total Data Center revenue | $89 Bn |
| Data Center sequential growth | 18% |
This creates a feedback loop between AI demand, cloud capacity and Nvidia’s hardware sales.
Nvidia Data Center Revenue Hits $89 Billion
Nvidia’s Data Center business generated $89 billion in Q2 fiscal 2027, increasing 18% sequentially.
The company said strong contributions came from both hyperscalers and its ACIE business, which includes NeoCloud, industrial and enterprise customers.
Data Center now accounts for more than 92% of Nvidia’s total quarterly revenue.
Nvidia Q2 FY2027 Revenue
$96.2 Billion
↓
Data Center
$89 Billion
↓
~92.5% of Total RevenueThe concentration demonstrates just how closely Nvidia’s financial performance is linked to AI infrastructure investment.
The $2 Trillion Backlog Extends Beyond Big Tech
Nvidia’s AI opportunity is no longer limited to Amazon, Microsoft, Google and Meta.
The company said demand is broadening across:
- Hyperscalers
- AI laboratories
- AI-native startups
- NeoCloud providers
- Enterprises
- Sovereign customers
The ACIE business, which includes NeoCloud, industrial and enterprise customers, generated $40 billion in Q2, up 138% year over year and 25% sequentially.
Nvidia’s Expanding Customer Base
| Customer Category | Role In AI Buildout |
|---|---|
| Hyperscalers | Massive cloud infrastructure |
| AI labs | Frontier-model training and inference |
| NeoClouds | Specialized AI capacity |
| Enterprises | Corporate AI deployment |
| Sovereigns | National AI infrastructure |
| AI startups | New model and application workloads |
This diversification is important because it reduces Nvidia’s dependence on any single category of customer.
NeoCloud Capacity Is Expanding Rapidly
Nvidia said its NeoCloud partners are expected to exit 2026 with approximately 8 gigawatts of installed capacity, compared with around 3 gigawatts at the end of 2025.
That represents an increase of approximately 167%.
End-2025
~3 GW
↓
2026 Expansion
↓
End-2026
~8 GWThe growth reflects increasing demand from enterprises, AI startups and sovereign customers that may not want to build their own large-scale AI data centers.
AWS Adds 2 Million Nvidia GPUs
One of the clearest examples of the infrastructure buildout came from Nvidia’s expanded partnership with Amazon Web Services.
AWS plans to deploy an additional 2 million Nvidia GPUs, beginning in the current quarter and continuing through the second quarter of fiscal 2029.
The deployment will also include Nvidia Vera CPUs, with some systems incorporating the Rubin architecture.
AWS-Nvidia Expansion
| Particular | Details |
|---|---|
| Additional Nvidia GPUs | 2 million |
| Deployment begins | Current quarter |
| Deployment period | Through Q2 FY2029 |
| CPUs | Nvidia Vera |
| New platform | Vera Rubin |
| Physical AI | Amazon warehouse robots |
The AWS commitment provides a concrete example of how the broader AI infrastructure spending cycle can translate into Nvidia hardware demand.
Vera Rubin Could Accelerate Revenue Growth
Nvidia expects its next-generation Vera Rubin platform to become an important driver of growth.
The company said Rubin is expected to account for approximately 20% of Data Center revenue in Q3. Hyperscale growth is then expected to accelerate again in Q4 and into fiscal 2028 as Rubin supply increases.
Nvidia’s Platform Ramp
| Platform | Strategic Role |
|---|---|
| Hopper | Earlier AI generation |
| Blackwell | Current major platform |
| Vera Rubin | Next-generation AI infrastructure |
| Vera CPU | AI server and agent workloads |
| Groq 3 LPX | High-interactivity inference |
The increasing revenue generated per unit of data-center power is also becoming an important part of Nvidia’s strategy.
Nvidia Says Revenue Per Gigawatt Is Rising
Huang highlighted how Nvidia’s full-stack systems are generating increasing economic value from each unit of data-center power.
He said revenue opportunity per gigawatt has increased from approximately $18 billion with Hopper to $25 billion with Grace Blackwell and about $40 billion with Vera Rubin.
Revenue Opportunity Per Gigawatt
| Nvidia Platform | Revenue Opportunity/GW |
|---|---|
| Hopper | ~$18 Bn |
| Grace Blackwell | ~$25 Bn |
| Vera Rubin | ~$40 Bn |
This means Nvidia does not necessarily need data centers to expand power capacity at the same rate as revenue.
It can generate more computing and economic output from the same power footprint as each generation becomes more productive.
Why The $2 Trillion Number Matters
The backlog matters because it provides a window into future infrastructure spending.
If cloud providers already have more than $2 trillion in backlog, their incentive to continue expanding computing capacity is significant.
However, the figure should not be treated as guaranteed revenue for Nvidia.
The backlog belongs to the cloud industry and reflects commitments across different services and infrastructure categories.
What The $2 Trillion Does And Does Not Mean
| Interpretation | Correct? |
|---|---|
| Nvidia has $2T of orders | No |
| Cloud industry backlog exceeds $2T | Yes |
| It indicates strong future infrastructure demand | Yes |
| All backlog will become Nvidia revenue | No |
| It supports Nvidia’s demand thesis | Yes |
| It eliminates AI spending risks | No |
This distinction is essential when interpreting the number.
The AI Infrastructure Cycle Could Last For Years
The combination of a $2 trillion cloud backlog, rising hyperscaler CapEx and Nvidia’s 70% fiscal 2028 revenue outlook suggests that the current AI infrastructure cycle could continue for several years.
Nvidia is effectively arguing that the industry is still in the early stages of building the computing capacity required for AI.
AI Adoption
↓
More Users
↓
More AI Queries
↓
More Inference
↓
More Compute
↓
More Data Centers
↓
More Nvidia Systems
↓
More AI CapacityAs AI becomes more deeply embedded in enterprise software and consumer applications, the amount of compute required could continue to rise.
AI Agents Could Further Increase Compute Demand
Huang has also emphasized the growing role of AI agents.
Unlike traditional software applications that may execute a relatively small number of computing operations per user interaction, autonomous agents can perform multiple steps continuously.
That could dramatically increase inference demand.
The implication is that AI adoption does not necessarily mean a fixed amount of compute per user.
More capable AI systems can actually increase the amount of computing required.
Nvidia Still Faces Major Risks
The $2 trillion backlog is bullish, but it does not eliminate the risks surrounding Nvidia and the AI infrastructure cycle.
The major risks include:
- Memory shortages
- Rising component costs
- Data-center construction delays
- Power constraints
- Export restrictions
- Customer concentration
- Custom AI chips
- AI investment returns
- Potential overbuilding
Nvidia itself expects gross margins to decline toward 71–72% in fiscal Q4 before settling around 72–73% in fiscal 2028 because of memory-cost pressures.
The Bigger Picture
Nvidia CFO Colette Kress’s reference to a more than $2 trillion cloud-industry backlog provides perhaps the clearest numerical snapshot yet of the scale of the AI infrastructure buildout. At the same time, the five largest hyperscalers are expected to spend nearly $800 billion on capital expenditure in 2026 and $1.3 trillion in 2027. These investments are being driven by the need to expand computing capacity for AI workloads across cloud services, frontier models, enterprise applications and sovereign AI infrastructure.
The most important point for Nvidia investors is that the $2 trillion figure is not Nvidia’s backlog. It represents the broader cloud industry’s backlog and therefore should not be converted directly into Nvidia revenue. However, it provides important context for Nvidia’s unusually bullish outlook. The company expects approximately 70% revenue growth in fiscal 2028 even though it says supply will remain constrained through at least the end of that fiscal year. Without those constraints, Huang said Nvidia’s growth would be significantly higher.
Looking Ahead
The next phase of the AI boom will depend on whether hyperscalers can turn their enormous infrastructure commitments into profitable AI services. Nvidia’s argument is that the economics are already strong: its hyperscale customers are seeing accelerating revenue and expanding margins, while the demand for Nvidia compute remains high enough that capacity is fully utilized across the clouds it serves. The additional 2 million GPUs planned by AWS and the rapid expansion of NeoCloud capacity provide concrete evidence that the infrastructure race is continuing.
For Nvidia, the challenge is increasingly one of execution rather than demand generation. The company must secure enough memory, manufacturing capacity, networking components and power-efficient infrastructure to capture more of the demand represented by the cloud industry’s $2 trillion backlog. If supply constraints ease while AI demand continues accelerating, Nvidia could potentially outperform its already aggressive 70% fiscal 2028 revenue outlook. If infrastructure spending slows or AI economics disappoint, however, the same backlog could prove less valuable than its headline number suggests
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
