Anthropic Is Beating OpenAI in Corporate AI, and Google Is Closing In With Cheaper Gemini Models

In the race to sell AI to big business, OpenAI is no longer the obvious front-runner.

Industry estimates now show Anthropic, maker of the Claude chatbot, pulling ahead in enterprise adoption and spending, powered by a surge in AI-assisted coding. At the same time, Google is running a different play: flooding the market with Gemini models that score well on benchmarks and undercut rivals on price.

The result isn’t a clean two-company showdown. It’s a fast-moving market where corporate buyers are switching vendors, developers are voting with their credit cards, and procurement teams are obsessing over performance per dollar.

Anthropic takes the lead in enterprise AI

The biggest shift is happening inside companies, not on consumer apps. In 2023, OpenAI was widely pegged at roughly 50% of the enterprise AI market, while Anthropic sat around 12%. Two years later, multiple industry reads put Anthropic ahead, around 32% versus 25% for OpenAI, and some estimates suggest Anthropic may account for as much as 40% of enterprise spending on large language models (LLMs).

That’s not statistical noise. It signals real vendor churn inside large organizations, exactly the kind of change that’s hard to reverse once tools get embedded into workflows.

Where Anthropic is reportedly strongest: software development. Estimates credit the company with about 42% of the “enterprise coding” segment. For executives, the pitch isn’t that the model is clever, it’s that it ships code faster. If a team cuts code review time from two days to half a day, the savings are easy for a CFO to understand.

Anthropic has also cultivated a reputation for being more conservative on safety and reliability, an important selling point when AI touches HR files, legal documents, or sensitive sales data. In many companies, winning the risk committee matters as much as winning the benchmark chart.

Claude Code becomes the wedge product

At the center of Anthropic’s momentum is Claude Code, a developer-focused product that’s being talked about as a “ChatGPT moment” for programmers, when adoption becomes self-sustaining across teams.

The article’s cited figures claim Claude Code crossed $1 billion in revenue in six months after its public launch. That’s about $1 billion in the U.S. as well, no currency conversion needed, but the implication is what matters: companies aren’t just experimenting. They’re paying at scale.

In a large enterprise, AI coding tools don’t spread casually. They come with licenses, API access, security policies, logging, and data-retention rules. If a tool clears those hurdles, it’s usually because it’s delivering concrete value: faster coding, better documentation, automated tests, and fewer painful back-and-forth cycles between product, engineering, and QA.

Anthropic is also pushing beyond engineering, promising more “first-draft ready” outputs for business deliverables, documents, spreadsheets, and slide decks with less cleanup required. That’s aimed at marketing teams, finance groups, and legal departments, where time saved can quickly translate into budget justification.

But product success doesn’t erase the industry’s biggest problem: compute costs. Running these models is expensive, and both Anthropic and OpenAI are widely described as far from profitability. That makes pricing and efficiency as important as flashy demos.

OpenAI still owns the public mindshare, while losing ground at work

ChatGPT remains the household name. For most Americans, “generative AI” still means OpenAI.

Inside corporate budgets, though, the story looks rougher. One estimate cited in the article puts OpenAI’s share of enterprise LLM spending dropping from about 50% in 2023 to roughly 27% in 2025, nearly cut in half in two years.

That doesn’t necessarily mean companies are abandoning OpenAI. It often means they’re diversifying. A data team might keep OpenAI for certain use cases while shifting coding workloads to Anthropic. Procurement departments are increasingly comparing vendors line by line: price per token, latency, contractual guarantees, compliance posture, and support.

OpenAI’s advantage is breadth. Analysts often describe its models as more flexible across a wider range of tasks, even if some competitors are viewed as more consistent in certain workflows. For companies trying to roll out AI across dozens of departments, a broad toolkit can beat a specialized scalpel.

OpenAI has also publicly discussed monthly revenue around $2 billion, about $24 billion annualized. That signals a powerful sales engine. The question hanging over the entire sector is whether that revenue can outpace the infrastructure bill required to generate it.

Google’s Gemini strategy: benchmark scores and aggressive pricing

While Anthropic and OpenAI grab headlines, Google remains the heavyweight, especially because it can bundle AI models with its cloud business and enterprise software footprint.

Google’s approach is less about being beloved and more about being everywhere: inside Google Cloud, inside integrations, and inside the default toolchains companies already use.

The article points to a recent Gemini 3.1 Pro model scoring 77.1% on ArcGi2, a reasoning-focused benchmark. Comparable cited scores were 68.8% for Claude Opus 4.6 and 52.9% for ChatGPT 5.2 on the same test. Benchmarks are always debated, but gaps like that are enough to reset the marketing narrative, and influence buyer shortlists.

Then there’s price. The cited rates for Gemini 3.1 Pro run about $2 per million input tokens and $12 per million output tokens. The article compares that to roughly $5 and $25 for Claude Opus 4.6, and about $1.75 and $14 for ChatGPT 5.2. For companies pushing millions of tokens a day, those differences show up fast on the monthly invoice.

Google is also leaning on distribution, getting models into customers’ hands quickly, before competitors can lock in accounts. It’s a classic Google move: compress the time between research breakthroughs and mass-market product, forcing rivals to respond.

A $37 billion boom, and growing anxiety about a bubble

The broader backdrop is a corporate spending surge. One estimate cited puts U.S. enterprise spending on generative AI at $37 billion over a year, up from $11.5 billion the year before, more than tripling.

Those numbers are framed as production API spending, weighted by customer size, in other words, not small pilots tucked into a corner, but real workloads running in the business. And once AI is in production, the problems get practical fast: prompt governance, data controls, hallucination monitoring, and strict rules about what systems can connect to what.

Whenever tens of billions of dollars start moving this quickly, “bubble” talk follows. The counterargument is simple: companies are seeing real productivity gains and doubling down with bigger budgets. Both can be true at once, strong demand, but fragile momentum if budgets tighten.

Behind the vendor switching is a quieter fight over standards: how AI “agents” should behave, which integration protocols become default, and who controls the tooling layer. If one vendor becomes the standard in developer workflows, it can effectively set the formats and habits that make switching later painful, even if a competitor releases a slightly better model.

The real ceiling: compute costs, profitability, and lock-in

Press releases talk about smarter models. The hard limit is still the cost of computation. If vendors cut prices to win market share but can’t reduce inference costs fast enough, they risk selling growth at a loss.

The article also highlights the capital intensity of the business: Anthropic has reportedly raised about $60 billion since its founding, backed by investors including Amazon and Google. For customers, that raises a practical question: which provider will still be stable, and strategically consistent, five or ten years from now?

There’s also lock-in, both technical and cultural. Google can simplify procurement by offering cloud, models, and tools in one stack, but that can trap customers in a single ecosystem. And when OpenAI or Anthropic becomes the default tool for developers, teams build workflows and habits around that model. Migrating later isn’t just swapping an API key; it’s retraining people and rewriting processes.

That’s why this “AI war” may not end with one winner. It’s increasingly looking like the cloud market: companies will pick models by use case, risk tolerance, and cost, and often stick with what’s already embedded, even when something shinier comes along.