In April this year, Anthropic introduced its AI tool Claude Design, intended for creative design and software application prototyping. Weeks before the launch, the company invited firms like Figma and Canva to be launch partners, showcasing the product's capabilities together. These design companies, long-time users of Anthropic's large models, saw this collaboration as an opportunity to highlight the complementary value of their own products with Claude Design.
However, just days before the scheduled launch, Figma withdrew from the partnership talks; around the same time, Anthropic's Chief Product Officer Mike Krieger resigned from Figma's board. The root cause of the partnership breakdown was Anthropic's significant last-minute changes to the product plan, with the newly launched design tool now posing direct and intense competition to the core businesses of Figma and Canva.
This partnership dispute is the latest signal of industry-wide anxiety triggered as Anthropic has steadily become a leading large model supplier for enterprises. Many companies have purchased Anthropic's models to build AI applications for white-collar work scenarios like programming, customer service, and legal work. Yet now, by launching its own commercial tools in these same areas, Anthropic is directly impacting its core client base.
Quinn Slack, co-founder and CEO of programming startup Amp, a long-time purchaser of Anthropic models, stated, "If the large model race ends with a single winner, it can freely squeeze customers, ignore partnership rules, and operate without restraint. But if the industry landscape isn't dominated by one player, then it cannot do without partners."
The ultimate winner of the race remains undecided, but Anthropic's actions this week have heightened client fears: the company formally launched its highly anticipated Mythos model, while also indicating that if clients use this model to develop their own advanced AI software or hardware, its backend would secretly degrade the model's computational performance.
Facing public criticism, Anthropic made a minor concession on Wednesday, stating it would proactively notify customers before limiting computing power for such tasks in the future.
While this move may hinder rival AI labs or U.S. adversaries from using Anthropic's models to advance their own technology, developers widely fear the company could intentionally weaken the foundational capabilities needed to build AI applications.
Several investment firms warned developers of this risk months ago: Anthropic would reserve its most powerful technology for developing its own competing applications.
In April, Martin Casado, a general partner at Andreessen Horowitz, posted on X: "Sooner or later, only the model developers will get the strongest version of the large model. Everyone else will get a stripped-down, distilled, low-performance version." (Casado also invests in and serves on the board of AI coding tool Cursor, which is both a major Anthropic client and a direct competitor.)
The current industry situation evokes earlier anxieties caused by Microsoft and Alphabet (Google): leveraging their monopolistic operating systems and search engines, they built vast application ecosystems, steering users toward their own browsers, e-commerce tools, and local business platforms, ultimately competing head-on with core commercial partners. Partners accused such behavior of unfair competition, equating it to "luring into partnership and then changing the terms."
Subsequently, U.S. and European courts ruled these companies engaged in illegal monopolization or abuse of market dominance (the Microsoft ruling was later overturned; Alphabet faced only minor penalties in the U.S. and its market value has continued to soar over the past year).
While competitors have not formally accused Anthropic of monopolistic practices—the company is only three years old—its actions, along with those of OpenAI, are raising similar concerns about excessive concentration of power in the AI industry.
Claude's parent company is experiencing unprecedented revenue growth: its annualized revenue surged fourfold in the first five months, approaching $50 billion, already surpassing OpenAI. OpenAI's revenue primarily comes from consumer-facing chatbots rather than enterprise API usage.
Advantage in Pricing Power
Several tech investors find this surprising: the combined revenue of Anthropic and OpenAI exceeds the sum of the next 32 top AI startups. An OpenAI spokesperson declined to comment.
A key driver of Anthropic's revenue surge is its strong pricing power. Since the start of this year, costs for enterprise clients using AI have risen significantly: Anthropic changed its billing model, moving away from fixed subscription fees to charging based on actual computing power consumption. Additionally, its latest-generation model uses a new tokenizer technology expected to further increase client costs.
Many large clients state that to automate processes for programmers and sales roles and improve efficiency, they must bear the rising computing costs. However, numerous companies have already implemented policies to limit employees' daily usage of Anthropic tools to control expenses.
Both Anthropic and OpenAI, relying on their own industry-leading large models, are developing upper-layer applications, impacting a large number of enterprises that purchase their models via API. Ironically, Microsoft is among the major clients affected.
Amid market concerns, Microsoft's stock has fallen nearly 20% year-to-date, while software firms like Figma have seen declines of up to 50%, as investors widely anticipate these companies will lose control over end-customer relationships.
This year, both Anthropic and OpenAI have positioned themselves as AI service providers deeply focused on enterprise services, claiming they can integrate with various office software (including Microsoft products) to achieve full-process automation. The companies state their proprietary AI agent scheduling systems can be embedded into the core data infrastructure of enterprises—the commercial software developed by Microsoft, Salesforce, and others that stores critical business data. Currently, Anthropic and OpenAI are forming joint ventures with private equity firms to implement AI deployment solutions for hundreds of companies.
A growing number of startups and established tech companies, realizing their core AI suppliers are encroaching on their own businesses, are publicly criticizing this competitive behavior.
Deliberate Concealment of Competitive Nature?
Quinn Slack, founder of programming firm Amp, said, "If the large model race ends with a single winner, it can freely squeeze customers, ignore partnership rules, and operate without restraint."
According to insiders familiar with the negotiations, when collaborating with design firms earlier this year, Anthropic management initially assured partners like Figma that the new design tool's functionality would be limited: it would only add basic preview generation for websites and apps to the existing Claude chatbot, and would not include features like multi-user collaboration or high-precision editing that would directly compete with partners.
However, just days before the product launch, Anthropic informed partners that the tool would fully include collaboration and advanced editing modules.
A week after the launch, Figma CEO Dylan Field stated at a private event hosted by Sequoia Capital that Anthropic's "communication was inconsistent," subtly accusing the company of misrepresentation. (Industry media Upstarts previously reported these remarks.)
A Figma spokesperson declined to comment; Danny Wu, Head of AI Product at Canva, issued a statement saying the launch of Claude Design followed over two years of deep collaboration, and Canva is one of the most used third-party applications within the Claude chatbot ecosystem, adding, "We greatly value our partnership with the Anthropic team."
Figma's experience is not an isolated case. A year ago, Anthropic launched its coding tool Claude Code, directly impacting numerous programming service providers that heavily rely on its models. Ultimately, Claude Code's revenue surpassed industry benchmarks like Cursor and Microsoft's GitHub Copilot.
Multiple sources reveal that prior to this year, Anthropic and OpenAI would typically inform partner clients in advance before launching products with competitive attributes. Recently, however, the two large model providers have stopped proactively disclosing such plans.
Several clients report that Anthropic, leveraging its strong market position, is now forcing customers to sign more stringent partnership terms. An executive at a leading programming startup disclosed that Anthropic notified them this year that to continue being featured in official promotional materials for new model launches, they must increase their Claude model procurement budget by 50%. Previously, the company had long been featured in official marketing, boosting its brand exposure and industry influence.
Beyond the programming sector, Anthropic is actively competing with clients in more vertical fields. Just last month, the company launched specialized AI functions for legal and finance, supporting tasks like quickly creating fundraising pitch materials, building financial models, tracking contract version changes, and due diligence data room analysis. This directly targets fast-growing startups like Harvey, Rogo, and Legora that specialize in financial and legal AI.
Max Junestrand, CEO of legal AI firm Legora, appeared unfazed: "Anthropic's legal tool is just a marketing gimmick for attention. It's full of flaws when handling real-world problems, not a mature standalone product—just the generic Claude chatbot with a configuration shell."
However, Junestrand also acknowledged that the design sector faces a greater impact: websites and prototypes are essentially code, and code generation is Anthropic's strength, making the launch of a design tool a natural business extension.
Fundamental Disregard for Client Interests
Ongoing enterprise client resistance to Anthropic's high model usage costs is providing an opportunity for pressured software service providers to fight back. Clients like Uber struggle to quantify the actual benefits of their AI purchases, allowing traditional enterprise software vendors like Palantir to publicly criticize large model companies.
Palantir CEO Alex Karp recently stated at an industry event, "When you partner with a large model company, you find they don't care about your needs at all. After partnering, costs rise, data security risks increase, computing token expenses remain high, yet it's hard to articulate what value this AI actually creates."
A significant portion of market resistance stems from Anthropic's new usage-based billing model, which also gives competitors an opening to persuade clients: rivals claim they can connect to multiple large models, helping companies select more cost-effective solutions.
Jeff Rehl, CTO of legal information service provider LexisNexis, which is facing pressure from Anthropic's legal tools, said, "With Anthropic and OpenAI fully shifting to usage-based billing, corporate costs become completely unpredictable. Law firms need to lock in stable budgets in advance."
Even with unpredictable cost fluctuations, some clients are willing to compromise to simplify their AI vendor list and access superior tools. An executive at a major investment bank, a client of financial AI service provider Rogo, revealed that after gaining approval to use Claude for processing confidential client materials in recent months, their team has almost completely switched to the tool.
The investment bank executive noted that Claude performs better in presentation formatting and financial model building.
A Rogo spokesperson stated that underlying large model developers like Anthropic and vertical-scenario AI service providers like Rogo address two different needs: Rogo adapts general large models for the scaled business processes of financial institutions, with built-in data integration, operational monitoring, and cost-benefit control capabilities, allowing enterprises to enjoy the capabilities of multiple top-tier large models and customized industry functions without being locked into a single vendor.
Comments