Opinion
The biggest opportunity is not in AI infrastructure. It's in applications.
The next generation of AI winners may be companies that use the new infrastructure not just to sell software, but to rebuild entire service industries with software-like economics.
Tech giants continue pouring unprecedented capital into power, chips, data centers and foundation models for AI. But as history suggests, the largest long-term value may in fact be sitting layers higher: in the companies turning the infrastructure into usable applications. The industry is already being shaped by companies staking a claim in the market by building directly on top of the infrastructure now being laid. These are the products that people and businesses actually use, and they are the real AI revolution.
Jensen Huang, the CEO of NVIDIA, describes AI as a “five-layer cake.” At the bottom is energy. Above it are chips. Then comes the infrastructure: data centers, networking, cooling and cloud systems. Above that are models. And the top layer? The proverbial cherry on top: applications. The first four layers make AI possible, but the fifth answers the only relevant question: What can we do with it?
Today, most of the hype, and an extraordinary amount of capital, is focused on the foundational layers. We’re building power plants and grid capacity, racing to secure GPUs, constructing enormous data centers and funding increasingly powerful models. And while all of this is necessary, it isn’t the end product, but rather the road to get us there. What Huang calls “the largest infrastructure buildout in human history” is necessary, but I believe the biggest opportunity sits higher in the stack: in applications. The hype around that layer has not yet reached anything close to its eventual scale. We often describe “hype” as a trend we wish we’d entered earlier or an upgrade we didn’t know we needed until it already existed. So if there is a next wave to catch, I believe this is it.
History offers useful clues, but nothing is truly comparable to the potential impact of AI or its economics. The possibilities are too broad, its rate of improvement too fast, and its ability to perform cognitive work makes it fundamentally different from previous technological revolutions, which focused on manpower rather than brainpower.
Still, tech revolutions tend to echo. Take the internet, for example. Investors poured billions into the infrastructure that would eventually make the internet possible. Companies raced to lay fiber-optic cables across continents and underneath oceans. Some built giant server farms to host the websites everyone assumed would dominate the economy. But where are those companies now? Many didn’t survive long enough to enjoy the web they helped create. The infrastructure survived; much of it was bought, reorganized and reused. But only later did the demand arrive, and the companies that became synonymous with the dot-com revolution were actually the ones built on top of it. The cherries.
Amazon answered, what can I get? Google answered, how can I find it? And Facebook answered, who else is here? Later, Netflix transformed entertainment, Airbnb took over accommodations, and countless other companies used the internet as an invisible springboard for a completely new customer experience.
The smartphone revolution followed a similar pattern. Telecom companies spent enormous sums upgrading mobile networks, while Apple and Google built the platforms. The application layer grew into entirely new categories that would have been difficult to imagine from the infrastructure alone: Uber, Instagram, WhatsApp, Spotify and more. The application layer didn’t ride the industry; it became the industry.
The infrastructure made the revolution possible, but the applications told us what for.
So how does this apply to AI and, more importantly, what does it mean for the future of high-tech? I believe one of the most interesting opportunities will be in vertically integrated AI companies: businesses that don’t simply sell AI tools to an industry, but rather use AI to rebuild the economics of the industry itself.
This is particularly powerful in legacy service sectors. In traditional service businesses, we often see scale as adding people. More customers require more employees, more revenue requires more hours, and so on. There is a natural limit to operating leverage, which compresses margins and makes hypergrowth difficult to achieve. But AI changes that equation. For the first time, a meaningful portion of knowledge-based tasks can be encoded into systems set to operate continuously, improve organically over time and serve dramatically more customers without a proportional increase in staffing. AI is creating the possibility of taking typically low-margin service industries and rebuilding them with software-like economics.
Legal services, accounting, insurance operations, healthcare administration, publishing and dozens of other specialized industries requiring enormous amounts of skilled human labor to perform structured, repeatable knowledge-based tasks might now produce the next generation of winners. And don’t bet on them looking like traditional SaaS companies selling tools to those industries, but rather like the industries themselves, rebuilt on a foundation of AI.
Imagine a law firm where software performs much of the legal workflow, including research and drafting. A publishing company where AI handles large parts of formatting, design and production. An insurance operation where AI processes what once required a team of employees. Companies can potentially combine something investors rarely get in traditional services: large existing markets, high growth, high margins and unprecedented scalability. In other words, AI can turn labor-scaled businesses into intelligence-scaled businesses. AI applications in their infancy today will become entire industries of tomorrow.
I’ve applied this same idea to book publishing, a centuries-old, labor-intensive industry. I wondered: why not build an AI-native operating system designed to automate and integrate large parts of the publishing process rather than simply adding an AI feature to the old model?
And I don’t think I’ll be alone. I believe thousands of entrepreneurs will apply this same idea across thousands of verticals. The infrastructure race is real, necessary and enormous. There will of course be major winners in energy, chips, data centers and models, but infrastructure buildouts eventually mature, while models become more capable and, over time, more commoditized.
What remains is the question every tech revolution must answer: what can we do now that we couldn’t before? My bet is that the largest chapter of the AI revolution will be written by the companies answering that question. And that chapter is only beginning.
Yehuda Niv, Founder and CEO of Spines, an AI-native book publishing operating system, Founder of Niv Publishing and an Author.
