OpenAI is gaining on Anthropic with business users, new data indicates

The Myth of the Loyal Enterprise

For the last year, venture capitalists and analysts have been trying to figure out which AI lab will become the next Microsoft. We want to know who is going to own the desk. Recently, new usage data suggests that OpenAI is clawing back market share from Anthropic among business users. But if you look closer at the numbers, the real story isn't about who is winning—it is about how fickle the customers actually are.

In the tech world, we talk a lot about "stickiness." It is the holy grail for any founder. You want a product that is so integrated into a company's workflow that ripping it out would be a disaster. Salesforce is sticky. AWS is sticky. But right now, LLM providers are about as sticky as a post-it note in a hurricane.

What we are seeing is a massive game of musical chairs. Every time a new model drops—whether it is GPT-4o or Claude 3.5 Sonnet—the enterprise crowd shifts their weight. They aren't loyal to the brand; they are loyal to the benchmarks. And for builders, that creates a very strange environment to work in.

The Performance Seesaw

Earlier this year, Anthropic had a massive moment. Developers and business teams started moving toward Claude because it felt more human, had a larger context window, and lacked some of the "lazy" tendencies people were reporting with GPT-4. It looked like OpenAI might be losing its grip on the professional tier.

Then the updates started rolling out again. OpenAI tightened up their latency, lowered their pricing, and improved the reliability of their API. Suddenly, the data shows the pendulum swinging back. Business users who were experimenting with Anthropic are shifting workloads back to OpenAI's infrastructure.

This isn't just about personal preference. For a founder, this volatility is a double-edged sword. On one hand, competition drives prices down and performance up. On the other hand, it makes it incredibly difficult to commit to a long-term roadmap. If you build your entire stack around one provider's specific quirks, you might find yourself lagging behind in six months when the competitor releases a superior version.

Why Builders Should Care

If you are building an AI-first product right now, the takeaway from this data is clear: Model agnosticism is a requirement, not a feature.

The moment you lock yourself into a single provider is the moment you lose your leverage. We are seeing major enterprises set up their internal systems to switch between models at the press of a button. They are using routing layers to send specific tasks to whichever model is currently the cheapest or fastest for that specific use case.

• Coding and logic? They might route to Claude.
• General reasoning and speed? They head back to GPT.
• Bulk processing? Maybe a fine-tuned open-source model.

If the giants like OpenAI and Anthropic can't keep their users from wandering, you certainly can't expect your users to stay if your performance dips because you're tied to the wrong engine. The founder perspective here is about risk management. You have to build a layer of abstraction between your user experience and the underlying model.

The Investor Dilemma

There is a bigger conversation happening in the boardroom. Investors are pouring billions into these labs at astronomical valuations, assuming that enterprise contracts will eventually provide a stable, recurring revenue stream. But these new reports show that enterprise AI spending is currently very volatile.

Companies are willing to flop back and forth between labs as each new update is released. That doesn't look like a stable SaaS business; it looks like a commodities market. If the underlying intelligence becomes a commodity, the margins will eventually collapse. The labs are in a race to the bottom on price while trying to stay at the top on performance.

"If your business model relies on one specific LLM being the best forever, you don't have a business; you have a temporary advantage."

We need to be honest about the fact that "winning" right now is temporary. OpenAI gaining on Anthropic this month is just a snapshot in time. Next month, a new release from the Claude team or a surprise from Google could flip the chart again.

Building for the Long Haul

So, where does that leave the founders and the builders? It means we need to focus on the parts of the stack that are sticky. The data, the UI, the workflow integration, and the specific domain knowledge you bring to the table. The LLM is just the engine. You don't see car buyers switching their loyalty every time a different oil company claims to have a better additive.

Right now, the enterprise is still in the "trial and error" phase. They are testing everything, committing to nothing, and moving wherever the friction is lowest. OpenAI's recent gains are a testament to their ability to iterate fast, but it is not a moat.

As an industry, we need to stop looking at these usage shifts as definitive victories. Instead, see them for what they are: a signal that the market is still wide open and that no one has truly captured the enterprise heart yet. The real winner won't be the one with the highest benchmark today, but the one who can eventually make their platform so essential that businesses stop looking at the benchmarks altogether.

The Takeaway

Don't get distracted by the horse race between Sam Altman and Dario Amodei. Whether OpenAI is up or Anthropic is down doesn't change your job as a builder. Your job is to create value that survives the volatility of the model layer. Assume the models will change, assume the prices will drop, and build your architecture so that you can profit no matter who is currently leading the pack.

Read the original at TechCrunch AI →