— At Nvidia’s GTC keynote in San Jose, CEO Jensen Huang said demand for the company’s Blackwell and the newer Vera Rubin architectures is accelerating sharply. He told the audience that Nvidia saw roughly $500 billion of demand for those families through 2026 and now expects at least $1 trillion in orders through 2027. Huang presented the projection as a monetary reflection of booming AI infrastructure investment and highlighted Rubin’s performance advantages over Blackwell.
Key takeaways
- Nvidia reported about $500 billion in demand for Blackwell and Rubin chips through 2026, and Jensen Huang said he now sees at least $1 trillion through 2027.
- Vera Rubin, announced in 2024 and put into production in January, is described by Nvidia as 3.5x faster than Blackwell for model training and 5x faster for inference, with peak performance up to 50 petaflops.
- Nvidia indicated it expects to scale production in the second half of 2026 to meet growing orders, signaling a significant manufacturing ramp.
- The $1 trillion figure refers to demand/orders scale rather than guaranteed revenue in a single fiscal year; Nvidia framed it as cumulative through 2027.
- If realized, the projection implies accelerated data-center upgrades and major capital spending by hyperscalers, cloud providers, and enterprises adopting large-scale AI models.
- Market reaction will depend on supply execution, customer procurement timelines, and broader macroeconomic conditions affecting capex plans.
Background
Nvidia’s Blackwell architecture became the company’s flagship AI datacenter family after previous generations, and Rubin was unveiled in 2024 as its successor with a focus on higher training and inference throughput. The company announced Rubin production began in January 2026 and has publicly quantified performance gains versus Blackwell: approximately 3.5 times faster on training workloads and roughly 5 times faster on inference tasks, with peak figures reaching about 50 petaflops. Those performance claims are central to Nvidia’s argument that customers will accelerate hardware refreshes and place larger orders.
The semiconductor industry has seen rising demand for specialized AI accelerators since 2022, driven by large language models and generative AI applications that require vast compute and memory bandwidth. Cloud providers, hyperscalers and enterprise AI projects have been the primary customers for datacenter GPUs and accelerators, and supply constraints in prior years amplified backlogs and order pipelines. Nvidia’s scale and ecosystem — including software stacks and partnerships — give it a dominant position in that market, which frames why its projections carry weight with investors and customers.
Main event
Speaking from the GTC stage in San Jose, Jensen Huang revisited the demand figures Nvidia shared last year and offered an updated, larger projection. He recapped that the company saw about $500 billion in demand for Blackwell and Rubin through 2026 and stated that, looking forward, he now sees at least $1 trillion in orders through 2027. Huang emphasized the scale as evidence of a sustained, multi-year investment cycle in AI infrastructure.
Huang tied the projection to Rubin’s technical advances, noting the architecture’s higher throughput for both training and inference. He referenced Nvidia’s January production start for Rubin and reiterated the company’s plan to expand manufacturing capacity later in the year to satisfy customer demand. Executives framed the ramp as a coordinated effort involving foundry partners, logistics and supply-chain planning.
Onstage, Nvidia presented technical slides comparing training and inference performance between Blackwell and Rubin and highlighted customer interest from cloud providers and large enterprises. The company described Rubin as optimized for large-model workloads and positioned it as a driver of increased purchases as organizations seek to accelerate AI development and deployment.
Analysis & implications
If Nvidia’s $1 trillion demand projection holds, the company and its ecosystem could see a multiyear uplift in revenue and order flow. Large-scale orders would translate into extended supply commitments from foundries and increased capital expenditures by customers deploying clusters at hyperscale. That could benefit suppliers across the semiconductor supply chain but would also concentrate risk in Nvidia’s ability to convert demand into deliverable units on schedule.
For cloud providers and enterprises, a wave of Rubin and Blackwell deployments would accelerate model training cycles, shorten iteration times for large models, and expand inference capacity. Organizations that move quickly to secure hardware could gain competitive advantages in model development, but smaller firms may face higher costs or longer waits for capacity, potentially increasing reliance on cloud services and managed offerings.
From a market perspective, the projection may raise investor expectations for Nvidia’s multi-year growth but also heighten scrutiny of margins, inventory timing and capital intensity. Realizing $1 trillion of orders does not equal $1 trillion in immediate revenue; hardware delivery, customer recognition, and amortization all affect how and when sales appear on financial statements.
Comparison & data
| Metric | Value |
|---|---|
| Demand through 2026 (Blackwell + Rubin) | $500 billion |
| Projected demand through 2027 | At least $1 trillion |
| Rubin vs Blackwell (training) | ~3.5x faster |
| Rubin vs Blackwell (inference) | ~5x faster |
| Peak Rubin performance | Up to 50 petaflops |
The table summarizes Nvidia’s public performance claims and the demand figures discussed onstage. These numbers are cumulative estimates of customer interest and order intent rather than single-year revenue figures; supply timing and order firming will determine how much is recognized each quarter.
Reactions & quotes
Investors and industry watchers reacted quickly to Huang’s projection, noting both the market opportunity and the execution risk inherent in such a large demand backlog.
“Now, I don’t know if you guys feel the same way, but $500 billion is an enormous amount of revenue…right here where I stand, I see through 2027, at least $1 trillion.”
Jensen Huang, Nvidia CEO (GTC keynote)
Company comments framed the figure as demand-driven; analysts stressed the difference between demand signals and confirmed orders.
“Orders of this magnitude would accelerate AI infrastructure refresh cycles worldwide and strain supply chains unless production scales predictably.”
Industry analyst
Nvidia also publicly signaled production scaling plans for Rubin, which the company said would ramp in the second half of 2026 to meet demand.
“We are increasing production capacity and coordinating with partners to meet customer timelines in H2 2026.”
Nvidia spokesperson (company statement)
Unconfirmed
- Whether the $1 trillion figure reflects fully contracted, legally binding purchase orders or aggregated demand estimates has not been explicitly clarified by Nvidia.
- Exact quarterly timing and unit volumes for the production ramp in the second half of 2026 remain subject to supplier disclosures and future Nvidia updates.
- Customer deployment schedules and how much of the projected demand will translate into on-premises versus cloud-hosted capacity are not yet confirmed.
Bottom line
Jensen Huang’s $1 trillion demand projection signals a potentially transformative wave of AI hardware purchasing if supply and customer procurement align. The figure underscores strong interest in higher-performance accelerators like Vera Rubin but does not eliminate execution risk tied to production scale-up and order conversion.
For market participants, the projection raises the stakes: suppliers and partners must deliver capacity, customers must budget and schedule deployments, and investors will watch quarterly delivery and revenue recognition closely. The next validation points will be Nvidia’s production updates, order confirmations from major customers, and how quickly cloud and enterprise deployments accelerate over the coming quarters.