Tokenomics 2030: The Infrastructure Behind the Intelligence Economy
A Sandisk outlook explores what the rapid growth of AI tokens could mean for the infrastructure powering the emerging intelligence economy
Transformative technologies create a new way to measure output. The industrial economy counted units produced. The digital economy measured transactions, compute and data. The emerging intelligence economy is beginning to coalesce around another unit: the token.
Tokens are the basic units AI models process and generate. They are not intelligence themselves, and not every token produces equal value. But they provide a practical way to measure the enormous flow of computational work behind AI.
Sandisk’s Tokenomics 2030 outlook estimates that AI systems will generate approximately 150 quadrillion tokens in 2026—roughly five billion every second.1 By 2030, the model indicates annual output could approach 10 quintillion tokens, or approximately 317 billion tokens per second.1.
That is an outlook representing intelligence moving from something people occasionally access toward a more prevalent interaction supporting people, businesses and autonomous systems.
Why Sandisk Is Studying Tokens
Why is a flash memory and storage company studying tokens? Sandisk designs and builds part of the infrastructure that turns data into intelligence. If tokens are the measure of what AI infrastructure ultimately produces, we want to understand not only how quickly token output may grow, but what that growth demands from the systems underneath it.
Those systems are composed of more than just raw accelerators. They depend on available power, compute performance, memory bandwidth, networking, storage and software working together.
That perspective inspired Eric Spanneut, Senior Vice President of Product Management at Sandisk, to assemble a team to create the Tokenomics model.
We built the Tokenomics model because tokens give us a way to connect the economic potential of AI with the physical infrastructure required to produce it. The question is not only how much intelligence people may want. It is how much intelligence the global infrastructure base could realistically support.
How We Built the Tokenomics Model
Many discussions about AI growth begin with demand: How many people will use AI? How many queries will they make? How much work will enterprises assign to AI?
Tokenomics adds another perspective by looking at the infrastructure available to generate those tokens. The Sandisk model considers the power directed toward AI inference, the size and capabilities of the accelerator fleet, compute and memory-bandwidth performance, networking and storage, and potential improvements in system utilization and software efficiency.
At a high level, the model establishes an estimated 2026 baseline and then looks at how a larger, more capable and more efficient infrastructure base could expand potential token output through 2030.
Power is the starting constraint, but power itself does not generate a token. You need compute, memory, networking, storage and software working together. Token growth ultimately depends on how efficiently the industry can convert infrastructure and power into useful intelligence.
This is why efficiency matters alongside infrastructure expansion. More capacity can increase token supply, but so can better utilization, more efficient data movement and reducing work the system has already performed.
What Does Token Growth Mean for Storage?
One of the first questions we get is: Does more token output mean more storage demand?
The answer is yes—but the relationship is more complex than applying a simple tokens-to-exabytes multiplier.
A token is a unit of AI processing, not a fixed-size storage object. Storage requirements depend on factors including workload type, model architecture, context length, data movement and reuse, infrastructure efficiency and, increasingly, how AI systems manage KV Cache during inference.
KV Cache creates a particularly interesting connection. As AI moves toward longer context and more persistent, agentic workflows, systems must retain more information from previous computation. Keeping that state in scarce high-bandwidth memory may not always be practical or economical.
Tiered architectures can place data across HBM, DRAM and enterprise SSDs according to its performance requirements. Enterprise SSDs can provide a larger-capacity tier for selected KV-Cache data, helping preserve previously computed information rather than repeatedly recreating it.
There isn’t a universal token-to-exabyte conversion. But as AI generates more tokens and maintains more context, data placement and reuse become increasingly important. The opportunity for flash is to help the system keep expensive compute and high-bandwidth memory focused on the work that benefits most from them.
The connection between tokens and storage is therefore less about a fixed capacity ratio and more about how efficiently infrastructure can help keep data available to the compute resources producing those tokens.
Scaling the Infrastructure of Intelligence
The journey from 5 billion to 317 billion tokens per second1 requires progress across the entire AI infrastructure stack. Accelerator, memory technology, storage device, or software optimization can’t support the growth of the intelligence economy on its own.
Tokenomics gives Sandisk a way to connect the unit of output, the token, with the various resources required to produce it.
As AI moves from answering isolated questions toward supporting continuous workflows and autonomous action, the ability to store, stage, access and reuse data becomes increasingly central to infrastructure productivity.
Read the Tokenomics 2030 Outlook report here:
TOKENOMICS 2030: THE RISE OF THE INTELLIGENCE ECONOMY
Forward Looking Statement:
Information above is based on Sandisk’s internal analysis and estimates that reflect current assumptions and expectations and are intended for illustrative purposes only.
This presentation includes forward-looking statements within the meaning of federal securities laws, including forecasts and expectations regarding AI, data infrastructure, related market opportunities, and the potential role and benefits of Sandisk’s NAND technology in such markets. These forward-looking statements are based on management's current expectations and are subject to risks and uncertainties that could cause actual results to differ materially from those predicted. Key risks and uncertainties include: adverse changes in global or regional economic conditions; volatility in demand for our products; difficulties or delays in product ramps, manufacturing or other supply chain disruptions; the impact of competitive products and pricing; our development and introduction of products based on new technologies and management of technology transitions; and other risks and uncertainties listed in the Sandisk’s filings with the Securities and Exchange Commission, including our Annual Report on Form 10-K filed with the SEC on August 17, 2026. You should not place undue reliance on these forward-looking statements, which speak only as of the date hereof, and we undertake no obligation to update or revise these forward-looking statements, except as required by law.
Author
Jeff Fochtman
November 12, 2026
[5 min read]