For the last two years, the AI conversation has revolved around one simple belief.
The biggest AI labs build the best models.
The best models charge premium prices.
And enterprises will happily keep paying because there are no meaningful alternatives.
That assumption is now facing its biggest test yet.
As OpenAI and Anthropic move closer to IPOs that could value each company at well over $800 billion, a wave of Chinese AI companies is quietly changing how businesses think about AI spending.
This isn’t just about who has the smartest model anymore.
It’s becoming a story about performance per dollar, and that changes everything.
A policy decision created an unexpected opening
A few weeks ago, the US Commerce Department asked Anthropic to restrict access to its most advanced AI models, Claude Mythos and Claude Fable, for users outside the United States.
The concern was national security.
These models are exceptionally capable at identifying software vulnerabilities, and regulators wanted tighter control over who could access those capabilities.
From a security perspective, the move made sense.
Commercially, however, it arrived at a difficult moment.
Companies that suddenly lost access to premium Anthropic models didn’t stop using AI.
Instead, many began testing alternatives.
Some of those alternatives happened to come from China.
AI has become one of the biggest cost items for enterprises
Until recently, AI spending was viewed as part of innovation budgets.
Today, it is becoming a significant operating expense.
During the latest earnings season, several large technology companies acknowledged that AI costs are putting pressure on profitability.
Among them were:
- Meta
- Shopify
- Spotify
Shopify summed it up neatly by saying that increasing large language model costs had offset some of its efficiency gains.
That may sound like ordinary corporate language, but it reflects a much larger trend.
Companies are discovering that running AI at scale is far more expensive than expected.
According to CloudZero, nearly half of surveyed companies were already spending over $100,000 every month on AI during 2025.
Naturally, finance teams have started asking an obvious question.
Are we getting enough value for what we’re paying?
The price gap is becoming difficult to ignore
Independent benchmark firm Artificial Analysis compared the costs of running leading AI models across identical evaluation workloads.
The differences were striking.
Approximate benchmark costs:
- Anthropic Claude: $4,811
- OpenAI ChatGPT: $3,357
- DeepSeek: $1,071
- Moonshot Kimi: $948
- Zhipu GLM: $544
That means the most expensive frontier models can cost several times more than some Chinese competitors for similar workloads.
Price alone doesn’t decide enterprise software purchases.
But when companies are spending millions annually on AI, those differences quickly become meaningful.
Chinese AI models are moving into the mainstream
Just a year ago, Chinese models represented only a tiny fraction of developer usage on platforms like OpenRouter.
Today, they account for the majority of requests.
That kind of shift rarely happens by accident.
Several Chinese companies have released increasingly capable models over the past few months, including:
- DeepSeek
- Zhipu AI
- Moonshot AI
- Xiaomi
Rather than relying on one breakthrough product, China now has multiple companies improving models simultaneously.
That makes the competitive landscape much harder to ignore.
Performance is getting close enough for many businesses
The conversation is no longer about whether Chinese AI models can compete.
The more relevant question is where they are “good enough.”
For many enterprise tasks such as:
- Coding assistance
- Customer support
- Internal knowledge search
- Document summarisation
- Workflow automation
Companies increasingly believe lower-cost models deliver sufficient quality.
They may not outperform the very best frontier models on every benchmark.
But if they solve 95% of everyday business problems at a fraction of the cost, many CFOs will see that as a compelling trade-off.
Businesses are already changing how they use AI
Another major shift is happening inside enterprise AI architecture.
Instead of sending every request to the most expensive model, companies are becoming smarter about routing workloads.
A growing approach is known as the advisor model.
Here’s how it works:
- A lower-cost model handles routine requests.
- Only difficult or complex tasks are forwarded to premium models such as OpenAI or Anthropic.
- The expensive model is used only when necessary.
This dramatically reduces overall AI spending without significantly affecting user experience.
From the customer’s perspective, it’s an efficient optimisation.
From the frontier AI provider’s perspective, it reduces high-margin usage.
Even software companies are helping customers spend less
Cost optimisation is becoming its own business opportunity.
Companies such as Figma are introducing tools that help customers reduce unnecessary AI token usage.
Enterprise customers have also become much more disciplined.
The pattern appears to follow three stages:
- Initial experimentation
- Rapid expansion of AI usage
- Careful optimisation once costs become visible
Many businesses now seem to be entering that third stage.
Instead of asking how much AI they can use, they’re asking how little they need to spend to achieve the same results.
Trust still gives American AI companies an advantage
Price isn’t the only factor.
Highly regulated industries still care deeply about:
- Data privacy
- Security
- Compliance
- Model governance
- Legal accountability
Banks, governments, defence organisations and healthcare providers are generally more cautious about adopting foreign AI infrastructure.
For these customers, American providers still enjoy a significant trust advantage.
Companies like Cohere have built much of their business around exactly this market.
However, regulated industries represent only one portion of global enterprise demand.
Outside those sectors, purchasing decisions often become much more price sensitive.
China is also preparing its AI companies for public markets
The competition isn’t limited to technology.
China is also building financial infrastructure to support its AI industry.
The Shanghai Stock Exchange recently proposed easier listing requirements for AI companies through its STAR Market.
The goal is straightforward.
Provide domestic AI firms with easier access to capital so they can continue investing aggressively.
At the same time, Chinese regulators have also promised stricter enforcement against AI-driven market manipulation and illegal trading practices.
This shows that China is thinking about AI as both a technology race and a capital markets race.
The US is responding in different ways
American companies aren’t standing still.
Several strategies are already emerging.
Nvidia is promoting downloadable AI systems that enterprises can run on their own infrastructure.
New startups are building open-weight American models that give businesses more flexibility without relying on Chinese platforms.
Rather than competing only on intelligence, these companies are trying to compete on:
- Cost
- Deployment flexibility
- Enterprise trust
- Data control
That suggests the AI market may eventually split into multiple segments instead of being dominated by a handful of frontier providers.
What this means for the upcoming IPOs
None of this means OpenAI or Anthropic are suddenly weak businesses.
Far from it.
Both continue to build some of the world’s most advanced AI systems.
Demand for cutting-edge models remains extremely strong.
The real question is whether today’s premium pricing can remain intact over the next decade.
Their IPO valuations assume several things continue to hold true:
- They remain technology leaders.
- Customers continue paying premium prices.
- Competitors struggle to catch up.
- Market share remains relatively stable.
The competitive environment now looks more complicated.
Lower-cost Chinese models are improving rapidly.
Enterprises are becoming far more disciplined about AI spending.
New routing techniques mean premium models are increasingly reserved only for the hardest tasks.
That doesn’t eliminate demand for frontier AI.
It changes where and how often those models are used.
The bigger picture
The AI race is no longer simply about building the smartest model.
It is becoming a contest between:
- Capability
- Cost
- Trust
- Accessibility
- Ecosystem
For investors evaluating future AI IPOs, that distinction matters.
Technology leadership still commands a premium.
But sustained pricing power is much harder to defend once credible alternatives emerge.
OpenAI and Anthropic remain leaders in frontier AI.
However, China’s growing ecosystem suggests that the next phase of competition may be decided less by who builds the absolute best model and more by who delivers the best value at scale.
For anyone following the upcoming AI IPO wave, that’s a trend worth watching closely.