Back in 2025, when artificial intelligence startup Spellbook entered into a multi-year agreement with Anthropic to power its legal document review systems, Chief Executive Scott Stevenson felt secure. Representatives from Anthropic assured him directly that they had no designs on entering the legal software market themselves, as Stevenson recalls.

However, within a span of twelve months, he was completely blindsided when Anthropic announced the rollout of Claude for Legal. Designed specifically for law firms, the software featured an array of functions—from regulatory tracking to automated citation analysis and contract examination—that directly intersected with Spellbook's core offerings. This created an undeniably awkward dynamic, according to Stevenson, prompting him to exercise much greater caution regarding the exchange of roadmap data and sensitive metrics with the AI powerhouse.

Anthropic has been a great partner but it made us realize that they are a massive company and two parts of the org can have conflicting mandates. Ultimately this is inevitable.

Despite previously depending on Claude to execute upward of half a million contract inspections every month for its client base, Spellbook has since redirected a significant portion of its financial allocation toward OpenAI.

The Amazon Basics Playbook in the AI Era

Spellbook's dilemma is hardly unique across the tech landscape. A vast contingent of emerging enterprises depends heavily on Anthropic for its industry-leading foundational intelligence models, which rank among the most powerful available. Yet, the relentless expansion and sheer market muscle of the AI titan have left numerous clients feeling as though they are perpetually navigating a precarious precipice, knowing the corporation could pivot at any moment to engineer a competing product.

One tech entrepreneur likened the unfolding scenario to the early days of Amazon's marketplace strategy, where the e-commerce giant eventually began manufacturing rival goods under its Amazon Basics brand. The founder noted the familiar pattern of selling on a platform only to watch the host introduce a competing version of the exact same merchandise. Anthropic opted not to provide comments regarding these comparisons.

This exact friction point manifested quite publicly around design enterprise Figma, which utilizes Anthropic models to support its native AI utilities. In April, merely days after Anthropic Chief Product Officer Mike Krieger abruptly stepped down from Figma's board, Anthropic debuted Claude Design, a direct challenger to Figma's layout editing software. During a user conference shortly following the release, Figma CEO Dylan Field publicly remarked that Anthropic was not consistently transparent in its communications—borrowing phrasing reminiscent of the brief and high-profile ouster of OpenAI leader Sam Altman, as reported by Upstarts Media. When questioned about Claude Design on the All-In Podcast, Field observed that it remains easy to mistake model progress for genuine design progress. Figma chose not to issue an official statement.

Soaring Valuations and Unyielding Margins

These mounting pressures have intensified in tandem with Anthropic's tightening grip on the market. Its technological outputs consistently rival or surpass those of OpenAI at the frontier level, a trajectory clearly reflected in its financial disclosures. According to its filed S-1 documentation, the enterprise currently services 300,000 corporate clients alongside an astonishing 900% net revenue retention rate. More than 11,900 accounts contribute upwards of $100,000 annually. For the initial half of 2026, the firm reported $16.3 billion in revenue alongside operating losses totaling $693 million, while targeting a staggering valuation between $1.5 trillion and $2 trillion ahead of its upcoming initial public offering.

Armed with such outsized leverage, Anthropic frequently dictates the terms during contract discussions with smaller enterprises. Even though recent weeks have seen price reductions, the company's intelligence infrastructure remains extraordinarily costly.

If we build our business on just Anthropic’s models, we won’t be able to make money ever.

Numerous startup operators informed reporters that Anthropic displays very little flexibility regarding pricing structures and volume-based discounts, complicating vendor relationships. One enterprise that previously poured millions into Anthropic systems completely terminated its usage due to prohibitive expenses, transitioning instead to open-weight models developed by China's DeepSeek. To mitigate some friction, Anthropic broadened its Claude Startups initiative in October, introducing incentives such as a complimentary year of team access alongside up to $45,000 in promotional credits for builders utilizing Claude.

In response to these cost barriers, a growing number of organizations are scaling back expenditures on frontier systems in favor of budget-friendly open-weight models that developers can freely download and customize. George Hu, president of Fireworks—a platform enabling startups to construct proprietary AI applications using open-weight frameworks—highlighted a deep-seated industry backlash during an August interview, noting enterprise chief information officers expressing frustration over essentially funding luxury real estate for executives.

Anthropic's uncompromising posture on pricing is largely driven by the immense financial pressures of preparing for what could become the largest stock market flotation in history. The organization must demonstrate profitability to sustain the staggering $130 billion it has amassed and the massive capital burned through computing infrastructure. Even with skyrocketing revenue figures, financial obligations loom large, with projections indicating the company is committed to pouring $1 trillion into AI infrastructure over the next two decades.

Geopolitical Battles and Open-Source Clashes

Beyond commercial tensions, Anthropic has consistently courted controversy on regulatory and geopolitical fronts. In February, the company rejected demands to strip safety safeguards from a Department of War contract, which barred military units from deploying Claude for fully automated lethal combat and widespread domestic surveillance. Consequently, the Pentagon blacklisted the firm as a supply chain risk. Although Anthropic challenged the decision in court—with a judge labeling the designation illegal and baseless—a D.C. circuit court ultimately upheld the government's stance. Today, public sector agreements constitute a mere 1% of the company's overall revenue.

While the broader technology sector praised the firm for pushing back against federal overreach, Anthropic drew intense criticism for its stance on open-weight AI. When federal regulators weighed restrictions on Chinese open-weights in July, Anthropic stood alone among major AI developers in declining to sign an industry coalition letter supporting open-source accessibility. Venture capitalist and administration AI advisor David Sacks criticized the move on social media, accusing the firm of attempting to kneecap the open-source community.

Anthropic countered that it never lobbied for an outright prohibition on open-weights, maintaining instead that such systems lack adequate safety guardrails and risk exploitation for biological or cyber threats. Conversely, industry figures like venture capitalist Bill Gurley have characterized Anthropic's emphasis on safety regulations as a tactic of regulatory capture designed to stifle competing open-source developers.

Additional friction points persist around ecosystem integration. Founders note that Anthropic provides minimal avenues for third parties to embed external applications directly within Claude—unlike competing platforms such as ChatGPT, which permit direct multimedia generation via third-party tools. Instead, Anthropic promotes its Model Context Protocol (MCP) framework, a standard allowing AI assistants to connect with external software. However, critics like Figma's Dylan Field have argued that MCP disproportionately benefits frontier laboratories by compelling companies to surrender proprietary data without receiving reciprocal insights in return.

For businesses willing to lock into lengthy six-month or twelve-month financial commitments, Anthropic offers enhanced access to post-integration engineering teams to troubleshoot operational roadblocks. However, several founders note a distinct institutional push to keep clients firmly within the proprietary ecosystem.

We felt like they really wanted us to be an Anthropic shop in terms of the models that we use.

Mitigating Vendor Lock-In

As the risks of over-reliance become increasingly apparent, tech enterprises are taking defensive measures against sudden policy shifts or service interruptions. In June 2025, Anthropic abruptly terminated the access of AI coding startup Windsurf with less than five days' notice amid reports of acquisition talks with OpenAI. While Anthropic co-founder Jared Kaplan defended the move as inconsistent with utilizing Claude, the fallout prompted other players in the ecosystem to alter their operational blueprints.

For instance, AI coding platform Cursor—historically one of Anthropic's largest clients—trained its own frontier model to insulate itself from volatility following customer churn toward Anthropic's rival Claude Code tool. Ultimately, as Stevenson and fellow founders emphasize, the prevailing strategy across Silicon Valley is ensuring that future product developments remain resilient against direct competition from the industry's largest gatekeepers.