Cheaper AI shifts value from models to business workflows

By Daniel Okafor

China’s STAR50 technology index fell 40% in less than four months, erasing about $480 billion in market value, according to Reuters in October.

Reuters also reported that AI-related stocks had reached 42% of Chinese market capitalization and 57% of trading turnover across mainland China and Hong Kong.

What does China’s correction signal?

The correction does not prove that AI has stopped creating value. It tests an old industry assumption: better models would create more valuable companies.

Much of the selloff may reflect valuations running ahead of financial performance. The Entrepreneur contributor sees a deeper change. Model capability keeps improving, while access becomes cheaper and competition among providers intensifies.

That helps businesses using AI. It complicates the case for companies investing billions in chips, data centers and foundation models. The industry can create enormous economic value without model builders capturing most of it.

Why does model convergence matter?

The pace of model development remains extraordinary. But the gap between leading models is shrinking, according to the source.

The 2026 Stanford AI Index found that Anthropic, xAI, Google and OpenAI were all within 25 points of each other on the Arena leaderboard in March. Stanford also described the performance difference between leading U.S. and Chinese models as having “effectively closed.”

Convergence changes purchasing decisions. Buyers look beyond the technically strongest model toward cost, reliability, speed and integration with existing workflows.

The contributor’s internal team once centered its model choices on output quality. It now asks: “What is the cheapest reliable way to complete this particular task?”

That practical question can rule out expensive models for document extraction or repetitive classification. Complex analysis may justify a stronger model. Some workflows use several models for different steps.

Once models become components rather than products, AI economics change.

Where can application companies capture the value?

Falling task costs can expand activity. Customer service teams handle more conversations. Developers run more coding agents. Marketers analyze more campaigns and test more experiments. Small businesses automate administrative work.

The economic value of those activities can rise even as the payment to model providers falls for each task. Business owners gain margins or automate work that was too expensive. Software companies can make AI features viable for thousands of smaller customers, not only large enterprises.

Model providers face a harder defensibility problem. A technical lead can disappear within months. Customer relationships, distribution, proprietary data, context and detailed knowledge of how work gets done are harder to replicate.

Most customers do not want artificial intelligence. They want an outcome:

• Respond to a customer.
• Prepare a proposal.
• Reconcile an expense.
• Schedule employees.
• Create a campaign.
• Understand why revenue changed.

The model is only one component in delivering that outcome. At an AI-native company, the contributor says, the difficult work involves connecting AI to the right data, supplying context, building the workflow and deciding where a person must intervene.

An established workflow can let a company apply each model improvement faster than a new entrant that must first understand the customer. The contributor sees durable advantages developing in the application layer.

How does AI change software economics?

Traditional SaaS benefited from very low marginal costs. Opening a dashboard five times or 50 times did not materially change the infrastructure economics. Subscription pricing trained customers to choose a plan and pay monthly for access.

AI introduces meaningful compute costs for document generation, conversation analysis, reasoning across data and autonomous workflows. A customer who uses the product heavily can become significantly more expensive to serve.

That changes the meaning of growth. One million AI actions in a month say little if each action costs more than the value it creates. Greater adoption can magnify losses rather than profits.

The contributor recommends measuring cost per successful outcome:

• How much does the task cost to complete?
• How often does the AI succeed without intervention?
• How much human work remains?
• Does the task create revenue, save time or improve customer experience?

These measures provide more information than the number of prompts or tokens consumed.

How should companies price AI products?

Not every software company must abandon subscriptions. But each must connect pricing with usage, customer value and infrastructure costs. Traditional subscriptions may still work for some products.

Others may adopt credits, usage allowances or a combination of subscription and consumption pricing. Heavy usage that reduces profitability eventually requires higher prices or tighter allowances.

Falling inference costs can produce the opposite effect. Previously expensive capabilities can be included more broadly, improving the product without damaging margins. That mechanism could spread lower model prices across more software use cases.

What should founders build next?

The contributor remains optimistic about AI applications. More competition among foundation models should produce better models, lower prices and more choices for companies building on them.

Stanford’s narrowing performance gap suggests that competition will increasingly turn toward cost, reliability and usefulness in specific domains. Founders should avoid making one model their entire competitive advantage.

They should instead:

• Identify workflows where the company has genuine expertise.
• Measure the cost of completing tasks rather than celebrating usage.
• Build systems that can adopt cheaper or better models when they appear.
• Focus on what happens after a model produces an answer.

The largest opportunities may sit outside the model itself. They may sit in the customer relationship, proprietary context, workflow and distribution required to turn intelligence into a useful result.

China’s correction matters even to founders uninterested in Chinese technology stocks. The signal is not that AI is losing value. It is that creating value with AI and capturing that value are separate achievements. As intelligence becomes cheaper and more widely available, that distinction will decide which AI businesses endure.