How AI economics will change for startups and small businesses

Efficiency to develop new demands for hardware

Faster, cheaper hardware is eliminating bottlenecks and transforming the economics of AI. Tensordyne builds inference racks using 90% less power than other products. This is an advance that embodies Jevons’ insight. Lower costs mean organizations do more AI, not less.

“The Jevons Paradox is gaining momentum as the types of tasks that AI is actually doing are becoming more complex and intensive,” says co-founder Jill Backhus. “When ChatGPT was released a few years ago, there was only one input and one output. Now, a single developer’s agents can perform hundreds of tasks at once, each requiring more compute-intensive inference.”

Customers value speed and model quality

The move to agent systems has already increased compute usage per user and task, but hardware is struggling to keep up and many companies are facing rising costs even as they accelerate AI adoption.

“To overcome the Jevons Paradox, we need to significantly lower the cost of running AI workloads without sacrificing the speed and model quality that customers value,” Backhus said. “Software will be more efficient, but the biggest game changer will be the hardware: chips and racks.”

The Tensordyne Napier inference system uses less energy, space, and money per token and runs 13x more throughput than Nvidia at lower costs. Systems and the AI ​​models they provide are typically either fast and expensive, or slow and cheap. The company plans to eliminate that compromise and make running high-performance AI workloads less capital-intensive.

“Startups have a finite supply of capital to burn, so the value of each token is much higher,” Bakhus says. “If chips and data centers are more than 10x more cost-effective, the model will be cheaper to operate and profit margins for financially constrained early-stage companies will change overnight. That could be rocket fuel for growth.”

Lower costs could democratize AI research and development, unlocking use cases that are currently too expensive to test. For founders, the core lesson lies in the underlying economics. “If a state-of-the-art customer support model currently costs $5 per session, but it goes to $0.50 due to cheaper hardware, the number of companies adopting it will increase by 10 or even 100 times,” Backhus says.

Productivity that creates new possibilities for professional services

Low-cost AI only makes sense if companies can find new ways to use AI. Companies like Orbital are seeing just that. In fields like law and real estate, outcomes have long been tied to working hours, but AI does more than just save time. It enables tasks that were previously uneconomical or impossible. At Orbital, a real estate AI platform that supports 200,000 transactions a year in the US and UK, we see this happening every day.

“The developers who design these AI systems aren’t working fewer hours; they’re typically spending the same amount of time, or sometimes more. They’re just getting much more done,” says CTO Andrew Thompson. “Human capital and token capital allow us to produce more software per person.”

Software is inherently creative

The real question is what that additional capacity will enable. “The total production will certainly increase, but what’s more interesting is what you get out of that production,” Thompson says. “The single biggest predictor of a software team’s success is the rate at which it ships new things to customers. Software is inherently creative. Nothing is true until a customer actually uses the product.”

He also points out that building software is a process of discovery. When AI powers that discovery loop, you can ship more products and reach solutions faster. “These are the efficiency gains that people want when using AI to level up their work,” he says.

Salary growth skyrockets

Once a rarity, becoming a “10x engineer” is now achievable by mastering AI, allowing small teams to deliver what once required entire departments. “Fewer people producing the same or more output means more productivity per person, which means higher rewards for individuals who effectively leverage AI,” Thompson says. “I think that’s why salary growth is skyrocketing across various roles in the tech industry.”

He expects the Jevons effect to create a virtuous cycle of adoption, with increased consumption of AI tools as their usefulness increases, but there is some nuance to this.

“At that time, Intel kept producing faster and faster CPUs, but Microsoft Word never got the same speed up,” he says. “That’s because Word wasn’t static. As CPUs could handle more processing, Microsoft developers built increasingly rich features into Word that weren’t possible before. Hardware advances were absorbed through new capabilities, not speed.”

Commercial real estate due diligence has also been rewritten, allowing companies to review every document on every property and create a complete due diligence report as if they had reviewed every property manually. “This is a new service that law firms can sell to their clients, and they can sell work that didn’t exist before because it wasn’t economically viable,” Thompson says.

Accessibility to facilitate the infrastructure flywheel

Even as more companies adopt these AI-powered workflows, computing demand will not plateau. It accelerates. That’s where infrastructure providers like Verda come into the picture. Better hardware increases efficiency and makes computing more accessible, further accelerating the Jevons effect.

The AI ​​infrastructure provider has grown revenue 20x in two years, exceeded $100 million in annual occupancy, and raised $117 million in April. This growth is not only due to easier access, but also from an enhancement cycle where hardware improvements and more capable models advance simultaneously.

A big part of the growth story for Verda and the AI ​​industry as a whole is that models and their capabilities improve and move from near-sufficient to truly production-ready. CTO Arturs Poli said: “Rather than thinking of rapid growth purely as a byproduct of easier access to computing, we see hardware and models improving together. Accelerators are getting more powerful, and the models running on them are getting more accurate. These are closely related, a clear Jevons dynamic with growth that is very rapid and difficult to predict.”

Value and price formation

Democratizing access removes long-standing barriers in favor of large corporations. The combination of high-quality, open-weight models and more readily available computing opens the door to lower-cost agent workflows. It also makes it easier for enterprises and businesses to deploy new services that leverage AI computing, as cost and availability are no longer barriers to entry.

This flywheel is already reshaping value and price. Poli expects the trend of increased usage to accelerate with more efficient and diverse ways to harness intelligence.

And sustainability is deeply tied to this expansion.

“Sustainability is very important to us and we want to lead the way with our own practices,” he says. “Specifically, we’re looking at everything from generating new clean energy to using excess heat to heat our homes.”

If the companies developing AI get it right, the Jevons paradox could become one of the defining economic forces shaping the next phase of AI economics. By increasing efficiency, rather than simply lowering costs, AI has the potential to expand into markets, products, and services that were not previously economically justifiable.