Signals Inbox·July 28, 2026·Frontier AI

Who is Anthropic’s biggest threat?

OpenAI is Anthropic’s biggest threat today because it can challenge Claude across chat, coding, APIs and enterprise work right now. Google is the slower, broader danger: once Gemini consistently matches Claude on difficult work, its cloud, chips and distribution could make it the harder rival to escape.

Q1What would make a rival Anthropic’s biggest threat?

Anthropic’s biggest threat is the rival most capable of taking Claude’s enterprise and coding customers now, then keeping that pressure up as AI models become cheaper.

A benchmark win that lasts a week is too narrow. We need to look at who can match Claude on difficult work, reach buyers without starting from zero, secure enough computing power and survive aggressive pricing. OpenAI competes directly with Claude, Claude Code and Anthropic’s API. Google brings its own models, cloud, chips and products used by billions. Open-model labs push prices down, while AWS and Microsoft can keep the customer relationship even when Claude does the work.

OpenAI can hurt Anthropic fastest. Google has the better chance of reshaping the market around it. Today, the first danger is larger.

Anthropic’s strongest threats by type

Q2Why is Anthropic so hard to challenge right now?

Anthropic is now too large, too fast-growing and too embedded in valuable work to be knocked aside by one strong model release.

Anthropic says its annualized revenue run rate crossed $47 billion after ending 2025 at roughly $9 billion. That is more than a fivefold increase in less than half a year. The number of business customers spending at least $1 million annually also climbed from more than 500 to more than 1,000 in under two months. Run-rate revenue annualizes the latest pace, so it is not the same as revenue already collected. Even allowing for that, the jump is far too large to explain with a few experiments or pilot contracts.

Claude Code explains much of the acceleration. Anthropic reported more than $2.5 billion in run-rate revenue from the product, with weekly active users doubling from the start of the year. Business subscriptions quadrupled, and enterprise customers now produce more than half of Claude Code revenue. KPMG is rolling Claude out to more than 276,000 employees, while TCS is using it in banking, insurance and software-engineering work. Claude has moved well beyond startup coding teams.

Rivals now have to pull important work away from companies that already use Claude at scale. Removing it from real workflows is much harder than beating Anthropic in a demo.

Q3Has Anthropic really overtaken OpenAI in enterprise AI?

Yes—on the clearest measures of paid enterprise model use, Anthropic has overtaken OpenAI. Many customers still pay both companies.

Menlo Ventures estimated that Anthropic captured 40% of enterprise foundation-model spending in 2025, against 27% for OpenAI and 21% for Google. Anthropic had held only 12% two years earlier. OpenAI lost almost half its share, while Google tripled its position. Menlo built the estimate from roughly 500 US enterprise decision-makers and a bottom-up model of spending on models, infrastructure and applications.

Ramp found the same broad shift by looking at payments from more than 70,000 American businesses. Anthropic appeared in the spending of 34.4% of companies, just ahead of OpenAI at 32.3%. More revealingly, 52% of businesses paying either provider paid both. Anthropic leads, but plenty of buyers still run Claude and OpenAI side by side.

Menlo and Ramp are imperfect in different ways. Menlo covers about 500 US buyers, while Ramp only sees payments from its own customers. Still, they point in the same direction: when American companies pay directly for frontier AI, Anthropic now comes out slightly ahead.

Enterprise model share and paid adoption

Q4Why is OpenAI still Anthropic’s biggest threat today?

OpenAI is still Anthropic’s biggest threat because it attacks every core Claude product and reaches far more users.

ChatGPT has more than 900 million weekly users and over 50 million consumer subscribers. OpenAI also reports more than one million business customers and over nine million paying business users. Enterprise products now generate more than 40% of OpenAI’s revenue, while its APIs process more than 15 billion tokens per minute. Anthropic may lead some paid enterprise measures, but far more people already know OpenAI’s products and can bring them into work.

The overlap is almost complete. ChatGPT competes with Claude, Codex competes with Claude Code, and both companies sell APIs, enterprise workspaces and agents that work across files and software. OpenAI has also brought its models and Codex onto AWS, placing them inside the procurement and security environment that helped Claude spread through large companies.

OpenAI’s latest models have closed most of the quality gap as well. Anthropic can still win individual evaluations, but OpenAI is close enough and starts with a much larger base of users. A strong OpenAI release can hit Claude sales almost immediately.

Q5Can OpenAI’s Codex take Claude Code’s lead?

Codex can now take real share from Claude Code because independent testing no longer shows a meaningful performance gap.

Artificial Analysis gives Claude Code with Opus 5 and Codex with GPT-5.6 Sol the same score of 67 on its Coding Agent Index. A benchmark tie cannot predict every repository or workflow, but it removes the easy argument that Claude Code is clearly better. Developers can run both tools and decide on speed, price, workflow or company policy instead.

Anthropic still leads the paid market. Menlo estimated that Claude models controlled 54% of enterprise coding-model spending, compared with 21% for OpenAI. OpenAI, however, says Codex has passed five million weekly users, more than six times its level when the desktop app launched. Around one fifth of those users now work outside software development, so Codex is already spreading into analysis, finance and operations.

Claude Code leads enterprise coding spending for now. The awkward bit for Anthropic is that teams can test coding agents side by side and switch before a normal software contract would come up for renewal.

Q6Is Google catching Anthropic fast enough to become the bigger threat?

Google is closing on Anthropic quickly, but its weaker coding position keeps it behind OpenAI as the threat most likely to hurt Claude today.

As seen above, Google’s enterprise model share tripled in two years, according to Menlo. In its latest quarterly results, Google said nearly 90% of the Fortune 100 used Gemini Enterprise. Its model APIs processed about 22 billion tokens per minute, up from 16 billion one quarter earlier, and more than nine million developers were building with Google models each month.

Google is producing that usage without currently owning the strongest public model. Sundar Pichai recently acknowledged that coding and agentic coding were areas Google needed to improve. Gemini 3.5 Pro was still in testing, and the company had begun a much larger training run for Gemini 4. Google has enormous demand; some of its hardest technical work is still unfinished.

A model that is good enough, cheaper and connected to a company’s existing cloud can already win plenty of work. Google becomes the bigger threat once Gemini can match Claude consistently in coding and difficult professional tasks. The reach is already there.

Q7Why is Google’s distribution so dangerous for Anthropic?

Google’s distribution gives Gemini a route into daily work that Anthropic cannot copy.

The Gemini app now reaches more than 900 million people each month. Google says AI Mode in Search has passed one billion monthly users, while AI Overviews reaches more than 2.5 billion. The audiences overlap heavily, so adding them would create a fake total. The useful point is how often Google can place Gemini in front of someone who opened Search, Gmail, Docs, Chrome or Android for another reason.

Google also says Gemini powers all 13 of its products with more than one billion users, including five above three billion. Anthropic has to persuade a person or company to choose Claude. Google can add Gemini to something the user already chose, spend less finding each user and bundle the assistant more easily.

ChatGPT is still used more frequently as a standalone AI product, with more than 900 million people opening it each week rather than each month. Anthropic has disclosed no comparable consumer audience. Google owns more of the surrounding workflow, which is why it remains the most dangerous long-term rival even while OpenAI poses the larger threat now.

Q8Can OpenAI and Google simply spend more than Anthropic?

Anthropic can finance the race today, but OpenAI and Google can survive more failed bets and keep spending.

Anthropic raised $65 billion at a $965 billion valuation, enough to support years of model and infrastructure work. OpenAI raised $122 billion soon afterward. Funding and revenue measure different things, but the rounds still show how long each company can keep paying for chips, researchers, products and sales teams before the economics need to improve.

Google operates on another scale. Alphabet spent $44.9 billion on capital expenditure in its latest quarter and raised its full-year guidance to between $195 billion and $205 billion, mostly for AI infrastructure. Alphabet also ended the quarter with $242.5 billion in cash and marketable securities. Anthropic’s entire funding round equals roughly four months of Alphabet’s planned capital spending.

OpenAI is also avoiding dependence on one supplier. It has a $38 billion AWS compute agreement, multigigawatt commitments with Nvidia and AMD, and a Broadcom-designed inference chip called Jalapeño. Google already owns its TPU roadmap and can use the same infrastructure across Gemini, Search and Cloud.

Money cannot guarantee a better model. It does buy more experiments, more chips, wider sales coverage and more chances to recover from a poor release. Anthropic is well funded; its two largest rivals have more room to get things wrong.

Q9Is Anthropic too dependent on Google and Amazon for computing power?

Anthropic has spread its computing supply across several providers, yet Google and Amazon still own essential machinery behind Claude.

AWS remains Anthropic’s primary cloud provider and training partner. More than 100,000 customers run Claude through Amazon Bedrock, according to Anthropic, and the company currently uses over one million AWS Trainium2 chips. Its latest agreement commits more than $100 billion to AWS technologies over ten years and secures up to five gigawatts of new capacity.

Google is another major supplier. Anthropic agreed to use as many as one million Google TPUs in a deal expected to bring more than one gigawatt online, then signed for five gigawatts of next-generation TPU capacity with Google and Broadcom. Anthropic has also committed to Azure compute and secured access to SpaceX’s Colossus GPU clusters. Using Trainium, TPUs and Nvidia GPUs reduces the danger of one supplier failing.

A sudden cutoff looks unlikely because Amazon and Google make money when Claude grows. The pressure comes during contract renewals. Anthropic buys enormous amounts of infrastructure from companies that also sell competing models, cloud platforms or workplace AI, leaving those partners close to its customers and in a strong position to set terms.

Q10Could AWS and Microsoft turn Claude into just another model?

Yes. AWS and Microsoft help Claude reach more companies while quietly making it easier to swap out.

Amazon Bedrock now offers more than 100 foundation models, including Claude, OpenAI models and open-weight alternatives. Its new endpoint supports both OpenAI-compatible APIs and Anthropic’s Messages API, so developers can move existing integrations with relatively little work. Bedrock can also route prompts between models to balance quality and cost. A company may keep Claude for hard reasoning and send simpler requests elsewhere without leaving AWS.

Microsoft is building similar choice into its products. Claude models are available through Microsoft Foundry, and Claude Opus 5 is rolling into Word, Excel, PowerPoint, Copilot Chat and Copilot Studio. Microsoft has also enabled Anthropic models by default for many commercial-cloud customers. When a company configures Claude through Foundry, billing and administration can remain inside Azure.

Selling Claude through AWS and Azure opens doors for Anthropic, but the cloud provider still controls the account and the surrounding tools. A company can change the underlying model while retaining the same contract, identity system, security controls and interface. Claude gets the distribution; AWS and Microsoft keep the harder-to-replace customer relationship.

Q11Are cheap open models becoming a real threat to Claude?

Cheap open models already weaken Claude’s pricing power, even though they have barely displaced it in large enterprises.

Menlo estimated that open models handled 11% of enterprise model usage, down from 19% one year earlier. Chinese open models represented only about 1% of the total. Large-company buying still favors Anthropic, OpenAI and Google, especially when security, reliability and support matter.

The quality and price gap is closing much faster among developers and smaller companies. Artificial Analysis now places Kimi K3 only four points behind Claude Opus 5 on its Intelligence Index, with an estimated benchmark cost of $0.72 per task versus $2.03 for Opus 5. DeepSeek V4 Pro scores lower, but its estimated cost of $0.04 is about fifty times cheaper than Opus 5.

Companies can send classification, extraction, first drafts and routine support work to a cheaper model, then reserve Claude for tasks where extra intelligence changes the result. Routing work this way cuts Anthropic’s volume and makes premium pricing harder to defend even when Claude stays on top.

The damage currently shows up more in margins than customer losses. Kimi, DeepSeek and other labs become much more dangerous if they add hosting, security controls and customer support designed for big companies.

Q12Is model quality enough to protect Anthropic?

Anthropic’s model lead buys breathing room, but today’s benchmark gaps are too small and unstable to protect the business for long.

Claude Opus 5 currently leads the Artificial Analysis Intelligence Index with a score of 61. Claude Fable 5 follows at 60, OpenAI’s best model at 59 and Kimi K3 at 57. A two-point gap between Opus 5 and OpenAI can decide a close evaluation, but it is unlikely to protect Anthropic for years. Artificial Analysis also finds Opus 5 relatively slow and expensive for its class.

Coding is tighter still. Claude Code and Codex share the same current coding-agent score, and Google has publicly acknowledged that it is working to close its own gap. Releases arrive often enough for a clear lead to disappear within weeks. A company choosing an AI platform for several years cannot assume today’s leaderboard will survive the contract.

Anthropic is safer when companies have built Claude into coding workflows, internal tools, stored knowledge and approval processes. Embedded workflows take longer to copy than a model score, although AWS, Microsoft and software that works with several models keep lowering the cost of switching.

Frontier model quality and measured benchmark cost

Q13So, who is Anthropic’s biggest threat?

OpenAI is Anthropic’s biggest threat now. Google is the rival most likely to become more dangerous over time.

OpenAI ranks first because the competition is already direct and close. Its best model sits near Anthropic’s leaders, while the company can introduce its products to a much larger pool of consumers and business users. Codex has also reached Claude Code on an independent coding-agent test. Every important Anthropic product now has a credible OpenAI alternative, so a strong release can affect Claude sales almost immediately.

Google can attack from more directions. It controls the cloud, chips, workplace software, browser, mobile operating system and search surfaces through which Gemini reaches users. Enterprise usage is rising quickly even before Google has restored a clear frontier-model lead. Once Gemini consistently matches Claude in coding and difficult professional work, Anthropic will face a rival that can bundle comparable intelligence almost everywhere.

Cheap open models will keep squeezing API prices. AWS and Microsoft can keep the customer relationship even when the underlying model changes. Those pressures are serious, but they do not match OpenAI’s current ability to take Anthropic’s customers across chat, coding, APIs and enterprise agents.

The ranking flips only if Google closes its coding and frontier-model gap while preserving its enormous distribution advantage.