AI capabilities are advancing rapidly. The harder question may be whether frontier companies can monetize them at the same pace.
Dario Amodei recently argued in We Must Pace the Frontier that frontier AI development should slow down enough to give safety systems more time to catch up. His call has also received public support from Sam Altman and Elon Musk.
His concern is clear. AI capabilities may be improving faster than our ability to understand, evaluate and control increasingly powerful systems. Amodei is not calling for AI development to stop. For safety reasons, he argues that the growth of frontier capabilities should be paced so that evaluation, alignment and control mechanisms have more time to catch up. [1]
But his argument raises another question:
What if frontier AI is also moving faster than its own business model?
AI demand is growing rapidly. Enterprise adoption is real. Revenue is rising.
At the same time, staying at the frontier is becoming extremely expensive.
A Reuters analysis of Microsoft, Alphabet, Amazon, Meta and Oracle estimated that between 2025 and 2027, every additional dollar of operating cash flow could come with about $1.57 of additional capital spending. Their combined capital expenditure could even exceed their combined free cash flow by 2027. [2]
These numbers cover the wider AI infrastructure build-out, not only frontier model training. Still, they show how capital-intensive the AI race has become.
This creates a simple question:
Can frontier companies turn new capabilities into durable revenue as quickly as they create them?
Enterprise adoption is not the final business model
Selling thousands of ChatGPT, Claude or Copilot licenses to a large company is meaningful.
But user licenses are probably not the biggest enterprise AI opportunity.
The larger opportunity starts when AI becomes part of daily business operations.
An AI agent could process an order, investigate a blocked delivery, check an invoice against an ERP system, support financial closing, test software or handle part of a customer service process.
At that point, AI is no longer only software used by employees. It becomes part of how the company operates.
A person may send a few prompts each day. A business process running continuously with AI could create thousands or millions of model calls involving reasoning, data access, API calls and validation.
That could create a much larger enterprise AI market.
But enterprises are not fully there yet.
Deloitte reported in 2026 that only 5% of surveyed organizations believed their business processes were highly ready for AI agents. Only 15% had scaled multi-agent systems across different business functions. [3]
McKinsey also found that AI adoption was growing quickly while many organizations were still trying to turn that usage into clear financial results. [4]
This does not mean enterprises need frontier development to slow down. Their systems, integrations and business processes will continue to improve either way.
The pressure may instead be on the companies building the frontier.
The market can move slower than the model cycle
A large company does not need to adopt every new frontier model immediately.
If a bank has built a reliable workflow around Model N, it may continue using it after Model N+1 appears. Stability, security and integration can matter more than always using the newest model.
Frontier companies face a different situation.
They may spend huge amounts developing Model N, release it and begin monetizing it. Before the market has fully used its economic potential, they may already need another large investment for Model N+1.
If model cycles remain very fast, frontier companies may face:
- shorter commercial lives for individual models
- more frequent training investment
- continuous product adaptation
- high infrastructure spending
- less time to build durable revenue around existing capabilities
The technology can keep improving and enterprise adoption can keep growing.
But the frontier company still has to finance the next step.
That may be the real tension.
The next enterprise opportunity may be around the model
Much of the next enterprise AI opportunity may be around the model rather than inside it.
A capable agent still needs safe access to ERP, CRM and other business systems. It needs the right permissions. Its actions need to be monitored and recorded. It must know when to stop, recover or ask a human for help.
This creates opportunities in three broad areas:
- Enterprise integration: connecting AI safely to business systems, data and workflows
- Trust infrastructure: authorization, audit, evaluation, monitoring and human control
- Agent operations: orchestration, cost control, model selection and reliability at scale
Platforms such as SAP, Salesforce, ServiceNow and Microsoft Dynamics could become important environments for this new AI layer.
A model may already be smart enough to understand a business problem. The harder part is often giving it safe and reliable access to the systems where the real work happens.
This is also why intelligence alone is not enough.
Generating code is different from changing a production system without supervision. Understanding why an order is blocked is different from taking the correct action inside an ERP system every time.
Making existing models more dependable may create as much enterprise value as making the next model even smarter.
More AI use does not guarantee more frontier revenue
Enterprise AI systems will probably use different types of models.
Difficult tasks may go to frontier models. Simpler tasks may go to smaller or cheaper models. Some workloads may use specialized or open-weight models.
This is good for customers because it reduces cost and increases flexibility.
But it also means that large growth in AI usage may not lead to the same growth in revenue for frontier model providers.
AI value creation and frontier company value capture are not necessarily the same thing.
As models become cheaper and easier to replace, this difference could become more important.
So what does pacing change?
This brings us back to Amodei.
If frontier development slows, safety teams clearly gain more time.
But frontier companies could also gain something economically valuable:
more time to monetize what they have already built.
A slower model cycle could extend the commercial life of existing models.
Enterprise customers would have more time to build production workflows around them.
More engineering effort could go into reliability, inference efficiency, integrations and lower operating costs.
Frontier companies could also have more time to generate returns from one major investment before making the next one.
This does not mean enterprise AI needs a slowdown. Enterprise systems, agent reliability and AI products will continue to improve anyway.
The question is different:
Can frontier companies continue increasing capability faster than the market can turn that capability into durable revenue?
From this perspective, pacing may not only give safety systems more time. It may also give frontier companies more time to capture economic value from capabilities they already have.
Of course, no company can easily slow down while competitors continue at full speed. Amodei also makes clear that his proposal does not ignore competition with China and the need to maintain technological leadership. [1]
So the safest pace and the economically sustainable pace may not always be the pace that competition allows.
The bigger question
For the last few years, the AI industry has focused heavily on how much smarter the next model will be.
That will remain important.
But another question is becoming harder to ignore:
Can frontier AI companies keep advancing faster than the market can monetize what they have already built?
The technology can succeed. Enterprise adoption can grow. AI can create enormous economic value.
Yet frontier companies may still have to invest in the next model before the market has fully realized the economic potential of the previous one.
That is what makes Amodei's argument interesting beyond safety.
If enterprise monetization continues to move more slowly than model development, how long can frontier companies sustain today's investment pace?
Sources
- Dario Amodei, We Must Pace the Frontier, September 2026.
- Reuters, AI investment boom puts Big Tech's free cash flow under pressure, July 22, 2026.
- Deloitte, AI Agents Are Only the Beginning: The Path to Agentic Transformation, August 2026.
- McKinsey & Company, The State of AI in 2026: On the Road to ROI, August 2026.
Author's note: This article was prepared with the assistance of AI. The main ideas, views and analysis are my own. AI was used to support research, structure and editing.