Executive Summary

Investment markets are applying two different standards of evidence to AI. The market has been willing to believe in AI infrastructure because GPUs, data centers, AI servers, optical communications, liquid cooling, power equipment, and supply chain orders can be built, measured, and reflected in financial results. But when the discussion shifts to SaaS and AI applications, the market asks for clearer proof of commercialization, including enterprise willingness to pay, user habits, workflow change, pricing power, and profit improvement.

AI demand is not absent today. Rather, its sources are not yet the same as fully mature end demand. A large part of current demand still comes from the strategic expansion of platforms and model companies, as well as experimental demand from AI startups supported by capital. This demand is real and can create cloud, compute, and infrastructure revenue, but it still needs to prove whether it can turn into stable usage, retention, and paid adoption.

For this reason, SaaS has become an important checkpoint for whether AI end demand is truly maturing. If AI is genuinely entering daily enterprise work, it should gradually appear in paid AI add-ons, usage frequency, renewal rates, seat expansion, usage growth, and gross margin improvement. In other words, the market believes in buildable AI, but it is still waiting for AI that enterprises and users are willing to keep paying for.

Introduction

Over the past year, an interesting asymmetry has gradually become clearer in the way investment markets respond to AI.

The market has been willing to believe in AI infrastructure. GPUs, data centers, AI servers, optical communications, liquid cooling, power equipment, and related supply chains have become some of the easiest AI narratives for capital markets to understand. Large technology companies continue to increase capital expenditure, data center construction keeps expanding, and supply chain companies have begun to reflect stronger orders and revenue growth.

But when the question turns to SaaS and AI applications, the market has been much more cautious. Many software companies continue to launch AI assistants, AI workflows, and enterprise AI features. They also keep emphasizing usage, adoption, and productivity gains. Yet relatively few have received long-term market trust and premium valuations.

This difference may not only reflect different positions in the industry. It may also show that the market is using different standards of evidence to evaluate AI.

For infrastructure, the market can see evidence that can be built, measured, and placed into financial models. For applications, the market is waiting for something harder to prove in advance, including user behavior, enterprise willingness to pay, workflow change, and the pricing power of software companies. In other words, the market believes in buildable AI, but it is still waiting for AI that customers will pay for.

Why the AI Infrastructure Narrative Is Easy to Believe

AI infrastructure has attracted strong market interest not only because it is important, but also because it is easier to understand. When large technology companies increase capital expenditure, the market can see it. When cloud platforms sign long-term compute contracts, the market can see it. When NVIDIA reports revenue growth and supply chain orders rise, the market can see it. When demand increases for data centers, power equipment, optical communications, liquid cooling, and AI servers, the market can see it as well.

These signals are relatively direct, and they can also show up more quickly in near-term financial results. This makes the AI infrastructure narrative easier to believe. It turns a highly uncertain AI future into investments, orders, capacity, and revenue that can be measured today.

This does not mean infrastructure demand is imaginary. On the contrary, this demand is very real. Large platforms do need more compute. Model companies do need a larger compute base. Enterprises and developers also need cloud services, APIs, and data infrastructure to use AI.

But the sources of this demand are not the same as a fully mature end market. A large part of the demand is actually coming from the strategic expansion of platforms and model companies. Cloud platforms do not want to lose their infrastructure position in the next stage of AI competition. Model companies need more compute to support training, inference, reasoning, and agentic workflows. Large technology companies also need to integrate AI capabilities into existing products, enterprise services, and advertising systems, so future user entry points are not redefined by other platforms.

In other words, the growth of AI infrastructure reflects real orders and capital spending. At the same time, it also reflects the anxiety of platform competition, the belief of capital markets, and the effort by large technology companies to defend future entry points.

AI Startups Sit in the Experimental Layer

Beyond large platforms and model companies, AI startups are also an important source of current demand. This layer is less visible than capital expenditure from large technology companies, and it does not directly reflect stable end demand in the way mature enterprise customers might. Yet AI startups are consuming large amounts of compute, APIs, GPU cloud capacity, data tools, vector databases, development platforms, and MLOps services. These expenditures create real cloud revenue and also support demand for model APIs and infrastructure tools.

But AI startup demand has one important feature. It is real, but not yet necessarily stable. Many AI startups are still searching for product-market fit. They are launching products quickly, testing user responses, adjusting business models, and using funding to support early growth. These activities create infrastructure demand, but they do not necessarily mean the end market has already formed a structure of sustained payment.

In this sense, AI startup demand looks more like an experimental layer between the supply side and the end market. It converts capital market belief in AI into cloud and infrastructure spending. But it still needs to prove whether these experiments can eventually turn into stable revenue, retention, and willingness to pay.

AI Demand Comes From Three Different Layers

To understand current AI demand more precisely, it may be useful to divide it into three layers.

The first layer is strategic demand from platforms and model companies.

This is currently the strongest layer. Large technology companies, cloud platforms, and model companies are investing heavily in advance so they do not miss the next stage of AI platform competition. The core of this demand is not only current usage, but also the competition for future platform position.

The second layer is experimental demand from AI startups.

This layer creates significant demand for compute and tools, but it still depends partly on venture funding and market confidence. It is real demand, but it also carries a higher degree of uncertainty.

The third layer is end demand from enterprises and consumers.

This is the most important layer, and also the one that most needs to be verified. The key questions are whether enterprises are willing to pay for AI tools over the long term, whether consumers will form daily usage habits, whether AI will truly enter workflows, and whether AI can create sustainable revenue, efficiency gains, and profit improvement.

AI demand is not absent today. A more precise way to describe it is that demand is already strong, but its sources are not yet fully mature end demand. A large part of today’s demand still comes from platform defense, model competition, and the experimental expansion of AI startups supported by capital.

The Market Is Applying Two Standards to AI

This also explains why the market currently finds it easier to believe in AI infrastructure than to fully believe in SaaS or AI applications.

For infrastructure companies, the market sees more concrete evidence. Large platforms are increasing capital expenditure. NVIDIA is delivering revenue growth. Data center construction continues to move forward. Demand is rising for optical communications, liquid cooling, power equipment, and server supply chains. These forms of evidence can be measured and tracked.

For SaaS companies, however, the market is asking for a different kind of proof. Can AI features really make customers willing to pay more? Can AI improve ARPU? Can AI increase retention? Can AI drive seat expansion? Can AI enter daily enterprise workflows, rather than staying at the level of trials, demonstrations, or short-term interest? More importantly, can AI improve the profitability of software companies, rather than only increasing inference costs and product development costs?

This is why the market is using two different standards. For infrastructure, the market is willing to believe in future demand first. For SaaS, the market wants evidence of commercialization first.

This is not a market contradiction. It shows that the market understands the uncertainty is different on each side. AI infrastructure is a construction problem. AI applications are a value-conversion problem. Construction problems are easier for capital markets to price in advance. Value-conversion problems require more evidence from real usage and financial results.

SaaS Is a Checkpoint for End Demand

If we want to know whether end demand for AI is truly maturing, SaaS is an especially important area to watch. SaaS companies are closer to enterprise workflows, and they are also closer to the question of whether enterprises are willing to pay for AI over the long term.

If AI is truly starting to change daily enterprise work, it should gradually appear in several signals.

• Enterprises are willing to buy AI add-ons.
• Users use AI in their workflows every day.
• AI features lead to higher renewal rates.
• Enterprises are willing to increase seats or usage.
• The gross margin pressure of AI products can be absorbed by pricing and efficiency improvements.

If these signals begin to appear, it would suggest that AI demand is starting to move from the infrastructure layer into the real enterprise usage layer. But if SaaS companies can only point to higher AI usage without proving willingness to pay, revenue contribution, and profit improvement, the market will likely remain skeptical.

This is why AI infrastructure can remain strong while software stocks stay under pressure. The market does not doubt AI itself. It is uncertain which software companies AI will actually make more valuable.

Conclusion

The current AI cycle has a clear reflexive structure. When the market believes AI is the next long-term growth story, large technology companies find it easier to invest more capital. When large technology companies invest more capital, NVIDIA, data centers, power equipment companies, and supply chain companies see stronger demand. When these companies deliver stronger financial results, the market becomes more confident in the AI infrastructure story.

AI startups are also part of this cycle. When capital markets believe in AI, startups can raise funding more easily. When startups receive funding, they buy cloud services, GPU compute, APIs, and development tools. These expenditures further support revenue for cloud providers, model companies, and infrastructure supply chains.

This cycle can reinforce itself, but it needs continuous new evidence. If end demand catches up, the cycle may become a true long-term growth cycle. If end demand does not catch up, or if commercialization moves too slowly, the market may begin to question whether earlier investment was too early, too large, or too concentrated.

This is why the current AI cycle may not be a simple demand boom. It looks more like an advance buildout driven by platform competition, startup experimentation, and capital market belief.

The market believes in infrastructure because it can turn the future of AI into orders, capital spending, and supply chain revenue that can be measured today. But whether the AI economy can ultimately work still depends on the harder parts to quantify in advance. Enterprises and users must form stable usage habits, keep paying, and allow that usage to support sustainable revenue and profit.

In other words, the market believes in buildable AI. But it is still waiting for AI that customers will pay for.