In recent years, The AI race is structured as a contest between large language model builders, with OpenAI holding a large early lead over rivals like Google and Anthropic. But that framework is starting to look outdated.
With the latest model release, that change has become harder to ignore. OpenAI's long-awaited GPT-5 was supposed to represent another leap forward. Instead, it arrived as a modest upgrade, reinforcing the sense that LLM progress was incremental.
Conversations are already underway in boardrooms and leadership teams. The question is not which model is best, but what effect will it actually have on your business?
The benchmarks most often used to rank these models focus on puzzle solving and abstract reasoning rather than whether the systems are reliable, cheap, secure, or easy to deploy within large organizations. That's what companies value most.
As the technological gap between major models narrows, the battle is shifting elsewhere. The AI race is currently being fought over who can drive adoption and use these tools on a daily basis.
Today, the capabilities of Google, OpenAI, and Anthropic appear to be much closer together than many expected two years ago, with no company clearly ahead in the way of settling the AI race. What increasingly separates the two is not model performance or consumer buzz, but the ability to deploy AI in large organizations.
This is important because benchmarking does not generate revenue, but adoption does. Billions of dollars are being borrowed from big tech groups to build AI infrastructure, but that investment only makes sense if companies actually use these tools. When pilots stall adoption, the economics quickly collapse, which is part of why investors are starting to ask harder questions.
AI frontrunner
By that measure, Microsoft is starting from a strong position, with tools already embedded in large enterprises through products like Office, Teams, and GitHub. But Microsoft's advantage is distribution, not ownership. The company has no control over the underlying models that power its AI tools, leaving it reliant on OpenAI, but the relationship looks more risky now that the ChatGPT maker is supplying models to Apple for use on the iPhone.
So it's no surprise that Microsoft has reportedly started paying for access to Anthropic's AI models.
Google, on the other hand, appears to be in an advantageous position thanks to its control of the “full stack,” from its proprietary models and productivity software to its cloud infrastructure and custom chips. The recently released Gemini 3 is widely seen as a leapfrog over OpenAI's GPT-5. highlights its benefits. Unlike Microsoft, Google controls the model, platform, and infrastructure it runs on. This is a strong position when race execution is critical.
For other players like Anthropic, the challenge is scale, not model quality. Claude is highly praised by corporate clients, especially when it comes to coding. But it will be difficult for larger rivals to translate that technological prowess into widespread adoption of their tools without the ability to reach and distribute consumers.
This problem is not unique to Anthropic. Expectations for AI were huge, far exceeding reality. An MIT study found that despite large-scale pilot projects, about 95% of companies still aren't seeing measurable benefits. This gap between hype and reward means that spending on AI is now under greater scrutiny.
Many organizations report that AI has helped them work faster, reduce time spent on daily tasks, and increase productivity for individuals. But most companies are still stuck there, using AI not to do a better job, but to do the same job faster.
Little attention has been paid to improving quality or stopping the spread of common “AI slop,” let alone using these tools for higher-value work, such as developing fresh products or finding new ways to create value for customers. There, AI will not only speed up tasks, but also begin to influence decision-making.
FOMO vs FOMU
Even in areas that are often cited as early adopters, such as consulting and banking, AI is still sitting on top of existing workflows. AI won't pay off unless companies change the way they work.
Many organizations are caught between two opposing forces: fear of missing out on AI opportunities (FOMO) and fear of disruption (FOMU). Huge sums of money have already been committed, increasing the pressure on both sides. The result is “pilot paralysis.” Many experiments were conducted, but few scaled. The priority now is to focus on a small number of use cases that can truly be used across the business.
If the next stage of the AI race is decided in boardrooms rather than labs, leaders will need to focus less on deploying tools and more on getting people to use them. This means that rather than assuming that access alone will change behavior, you need to embed AI into daily workflows and train employees to apply the tools to their work.
For now, companies are taking a completely different approach. Some have been clear about how employees should leverage AI. Others said little. Without clear rules, staff can only guess. Some companies avoid AI altogether, while others rely on personal ChatGPT or Claude accounts to process company emails, documents, and data, increasing the risk of sensitive information leaving the organization.
As the AI race moves from model quality to everyday use within enterprises, unclear rules are already holding some companies back. As technologies converge, the winners will be those who actually make AI usable for everyday use.
Michael Wade is Tonomus Professor of Strategy and Digital at IMD Business School.
