The Great Infrastructure Pivot: Why AI Titans are Rethinking Their Financial Plumbing

By PYMNTS | September 8, 2026

The meteoric rise of generative artificial intelligence has brought unprecedented scale to the tech industry, but it has also created a unique set of "growing pains" for its primary architects. Anthropic, one of the leading forces in the large language model (LLM) space, finds itself at a strategic crossroads. With quarterly revenue ballooning to $11.5 billion—up from a mere $787 million just a year prior—the company is no longer just managing a product; it is managing a massive, high-velocity financial ecosystem.

This explosive growth has triggered a fundamental re-evaluation of how Anthropic handles its "financial plumbing." As the company scales, it is weighing the merits of the "build-versus-buy" dilemma, specifically regarding its billing, payments, fraud detection, and tax infrastructure. While long-term partner Stripe has been instrumental in the company’s early success, Anthropic is increasingly looking to bring critical financial operations in-house to gain better control over its unit economics.

The Financial Mechanics of AI at Scale

To understand why a company like Anthropic would consider moving away from an industry giant like Stripe, one must first appreciate the complexity of AI-driven revenue models. Unlike traditional software-as-a-service (SaaS) companies, which often rely on flat subscription fees, AI firms operate on highly dynamic, usage-based consumption.

Every API call, token generated, and query processed represents a micro-transaction that must be metered, billed, reconciled, and taxed. When revenue grows by orders of magnitude in a single year, the margins associated with third-party processing fees begin to loom large.

Anthropic’s recent job postings for staff software engineers reveal a clear internal mandate: the company is looking for talent capable of architecting real-time risk assessment systems, managing complex contract entitlements, and curbing promotional code abuse. These are not merely IT support roles; they are roles focused on the firm’s core economic engine. By bringing these systems in-house, Anthropic aims to tailor its infrastructure to the specific volatility and high-volume nature of AI token consumption, potentially insulating itself from the standardized cost structures of external payment processors.

A Chronology of Partnership and Pivot

The relationship between Anthropic and Stripe has been characterized by deep technical integration. In the early stages of its growth, Anthropic leaned heavily on Stripe’s "full-stack" approach. This included:

• Payment Processing: Leveraging Stripe’s global network to handle international transactions.
• Billing: Utilizing Stripe’s advanced usage-based billing tools to manage AI token consumption.
• Fraud Detection: Implementing Stripe Radar, which the companies claim helped Anthropic reduce incorrectly blocked transactions by 83%.
• Reconciliation: Using Data Pipeline to streamline month-end accounting, which shortened reconciliation cycles by six days.

However, the current phase of the partnership is markedly different from the one described in early case studies. While an Anthropic spokesperson confirmed that "Stripe has been a strong partner to Anthropic for years" and that they "continue to work with them across our business," the shift in focus is undeniable. The company is now evaluating which modules of this stack—specifically those closest to their proprietary pricing logic—should be unbundled from the Stripe ecosystem.

Comparative Strategies: Anthropic vs. OpenAI

The industry is currently witnessing a divergence in strategy between the two biggest names in generative AI. While Anthropic is exploring the path of internal development, its primary rival, OpenAI, has opted for a "multi-processor" strategy designed to prevent vendor lock-in.

OpenAI recently moved its stored card data to an independent intermediary, a move that provides the flexibility to route transactions across multiple payment gateways. By adding Adyen alongside Stripe, OpenAI is effectively playing the two processors against one another. This strategy is not about building the infrastructure from scratch, but about ensuring that the company maintains leverage over its payment providers.

The battleground for these providers is shifting toward specialized billing technology. Stripe’s acquisition of Metronome—a deal valued at roughly $1 billion—was a clear play to dominate the usage-based billing market. Similarly, Adyen’s acquisition of Orb for $335 million in July signaled that the payment processor is keen to offer similar "usage-first" billing capabilities. Both companies are fighting to prove that their off-the-shelf infrastructure is more cost-effective and reliable than anything a high-growth AI firm could build on its own.

The Technical Burden of In-House Payments

Building payment infrastructure in-house is a daunting proposition, even for companies with multi-billion dollar revenues. The "plumbing" of global payments involves far more than just connecting to a bank API. It requires:

1. Jurisdictional Complexity: Managing tax compliance across hundreds of global jurisdictions is a massive, ever-changing legal burden.
2. Fraud and Risk: Developing proprietary fraud detection that can keep up with sophisticated bad actors—especially those attempting to exploit AI API keys—is a constant arms race.
3. Latency and Throughput: For an AI company, billing must be as fast as the model’s inference. If the billing system adds latency to an API request, it directly degrades the user experience.
4. Invoicing and Reconciliation: As enterprise customer bases grow, so does the complexity of managing custom contract terms, volume discounts, and multi-entity invoicing.

Anthropic is currently evaluating whether the cost of maintaining a specialized internal team to solve these problems is lower than the long-term fees paid to external providers. The trade-off is clear: by building internally, Anthropic gains the ability to customize its economics in ways a standardized third-party platform cannot. However, they also lose the benefit of the massive scale and shared intelligence that companies like Stripe and Adyen provide across their entire client base.

The Implications for the Fintech Landscape

The decision by companies like Anthropic to "insource" their financial infrastructure could have profound implications for the future of fintech.

For the payment processors, the risk is clear: the most high-growth, high-revenue customers are the ones most likely to eventually outgrow the standard product offering. If the "AI 50" (of which 86% currently monetize through Stripe) all decide to build their own billing systems, the fintech giants face a significant existential threat.

Conversely, this trend highlights the massive opportunity for "modular" fintech. Instead of relying on a "one-size-fits-all" platform, companies are increasingly demanding "pick-and-choose" infrastructure. They may want Stripe’s core payment processing, but they may prefer to build their own usage-metering and tax-calculation layers.

For investors and industry observers, the takeaway is that AI firms are reaching a level of maturity where they are no longer satisfied being "customers" of the fintech ecosystem—they are becoming "peers." As these companies continue to scale, the distinction between a software company and a financial services firm will continue to blur. Whether Anthropic succeeds in building a more efficient, proprietary financial backbone remains to be seen, but their move signals that in the era of $11.5 billion revenues, even the best third-party infrastructure may eventually feel like a constraint on innovation.

Ultimately, the experiment is a litmus test for the entire tech sector: at what point does a service become so integral to your core operations that it must be brought in-house? For Anthropic, that point arrived with the realization that their financial systems are not just a service, but a competitive advantage.