Anthropic's Wobbling Flywheel

Anthropic’s business model quickly earned a reputation as "probably the most insane flywheel we’ve ever seen in the history of business," according to analyst Matthew Berman, drawing parallels only to Google AdSense, the legendary "money printing goat." This self-reinforcing loop promised not just exponential revenue but an insurmountable competitive advantage in the burgeoning AI landscape. At its core, Anthropic envisioned a continuous cycle of growth, perpetually improving its foundational models, particularly its flagship Claude series.

The company focused intensely on developing an incredible coding model, designed for high-value enterprise AI use cases. Selling this specialized model to businesses generated substantial revenue while simultaneously acting as a crucial data acquisition engine. As organizations integrated Anthropic’s AI into their workflows, the company garnered invaluable proprietary code data—real-world, enterprise-grade insights into practical application. This data, distinct from publicly available datasets, became a critical asset in the race for AI supremacy.

This unique, high-quality data then directly informed the training of Anthropic’s next-generation models. Every piece of code generated or refined by enterprise users served to enhance the AI’s understanding, making the subsequent model iteration demonstrably better. This created a compounding data advantage that seemed almost unbeatable, as increased enterprise adoption led directly to more unique, high-fidelity data, which in turn yielded superior models like the successive versions of Claude, attracting even more enterprise clients.

The strategic brilliance of this model lay in its self-perpetuating nature: - Enterprise AI adoption drove revenue. - Revenue facilitated further model development. - Enterprise usage generated proprietary data. - Proprietary data fueled superior model training. - Better models attracted more enterprise customers.

Such a mechanism promised to widen the gap between Anthropic and its competitors with each passing quarter, creating a formidable moat. The model wasn’t merely about selling a product; it was about embedding a learning mechanism into its very revenue stream, turning every customer interaction into a data point for future innovation. This "beautiful flywheel" captivated investors and positioned Anthropic as a formidable challenger, poised for continuous, self-propelled advancement in the intensely competitive AI market, seemingly immune to external pressures.

AI analyst Matthew Berman recently pinpointed the core problem threatening Anthropic's business model: a critical shortage of compute power. Berman described Anthropic's "beautiful flywheel" as "coming off its axle, or at least wobbling," a direct consequence of this fundamental resource constraint. He attributed this breakdown to "one miscalculation by Dario," referring to co-founder and CEO Dario Amodei, whose strategic oversight on compute casts a long shadow.

Anthropic itself offered undeniable real-world evidence of this crunch. Beginning in late March 2026, the company implemented stricter Claude usage limits for its paid users during peak weekday hours, specifically 8 a.m. to 2 p.m. ET. This unprecedented move stemmed from demand from millions of new paid users outpacing available infrastructure, a surge partly attributed to a user boycott of OpenAI following its Pentagon contract in late February 2026.

Matthew Berman’s analysis pins Anthropic’s compute crisis squarely on a "miscalculation by Dario," suggesting a singular oversight by CEO Dario Amodei regarding infrastructure planning. This narrative implies a failure to adequately project the company's resource needs. However, framing the issue as a simple forecasting error oversimplifies the unprecedented market dynamics and sheer scale of user adoption Anthropic encountered.

Reframing the situation, the company grappled with the inherent difficulty of predicting demand in a rapidly evolving, hyper-growth sector. Planning compute for a nascent technology experiencing exponential adoption presents an almost impossible challenge, making traditional projections largely obsolete. Anthropic wasn't just planning for growth; it was provisioning for a rocket launch in an uncharted competitive landscape.

A massive, unanticipated surge in user adoption, including millions of new paid users, quickly outstripped existing capacity. This demand spike was partly fueled by a significant user boycott of OpenAI in late February 2026. Following its controversial Pentagon contract, many disaffected users migrated to Claude, creating a critical strain on Anthropic’s resources and leading to stricter usage limits during peak weekday hours.

In response, Anthropic aggressively pursued massive compute commitments to prevent future bottlenecks. They announced an agreement with Google and Broadcom for "multiple gigawatts of next-generation TPU capacity" by 2027. This proactive scramble for hardware, alongside securing billions in investment, underscores the extraordinary scale of the challenge and their determination to secure future compute power. Further details on these strategic moves, including substantial investment from Amazon, can be found here: Anthropic gets another $2.75 billion from Amazon to take on OpenAI - The Verge.

Anthropic built its business on a singular, potent strategy: capturing the enterprise market. Unlike many competitors chasing consumer ad dollars, Anthropic funneled its resources into securing lucrative contracts and paid subscriptions from businesses. This laser focus generated immediate revenue and, crucially, a rich, proprietary data stream.

Enterprise coding use-cases became the perfect accelerant for Anthropic’s flywheel. Companies deploying Claude for internal development, debugging, and code generation provided an invaluable, high-quality data source. This isn't generic text; it's complex, structured code, directly relevant to refining Claude's core capabilities as a programming assistant.

Every interaction, every line of code generated or refined, fed back into Anthropic’s models. This continuous feedback loop allowed the company to rapidly iterate and improve its "incredible coding model," as Matthew Berman described it. The data wasn't just useful; it was foundational, training the next generation of Claude to be even more proficient and context-aware.

Central to this strategy is Claude Code, Anthropic's agentic coding system. This sophisticated offering aims for deep integration into customer workflows, moving beyond simple API calls to become an embedded, indispensable part of development pipelines. Such integration inherently creates high switching costs, locking in enterprise clients who become reliant on Claude Code's performance and familiarity within their systems.

This strategic pivot towards enterprise, powered by specialized coding models and robust data feedback, promised a self-reinforcing cycle of improvement and revenue. It established a formidable competitive moat, but also amplified the underlying demands on Anthropic’s compute infrastructure.

Facing an existential compute shortage, Anthropic executed its first major countermove on April 6, 2026. The company announced a strategic agreement with Google and Broadcom, aiming to secure the vital hardware infrastructure needed to power its ambitious enterprise AI expansion. This partnership directly addresses the critical bottleneck highlighted by Matthew Berman, which threatened to derail Anthropic's "insane flywheel" just as it gained momentum.

Central to this agreement is the commitment for Anthropic to access "multiple gigawatts of next-generation TPU capacity." These Tensor Processing Units, custom-designed by Google specifically for demanding AI workloads, represent a crucial injection of specialized processing power. Expected to come online starting in 2027, this massive capacity is described as Anthropic's "most significant compute commitment to date," signaling a monumental, multi-year investment in its future. It provides the muscle necessary to both serve existing enterprise clients and train increasingly complex models.

Securing this immense compute power marks a pivotal first step in Anthropic's long-term strategy. It moves beyond merely patching the current deficit, instead laying foundational infrastructure for sustained growth and competitive advantage in the fiercely competitive AI landscape. The deal allows Anthropic to scale its Claude models effectively, meet burgeoning enterprise demand for its advanced solutions, and continue training superior next-generation AI, ensuring its proprietary data can be fully leveraged. This proactive measure aims to transform a critical vulnerability into a robust, lasting infrastructure edge, essential for Anthropic to solidify its position against rivals and keep its flywheel spinning.

Amazon delivered Anthropic its most substantial lifeline, a colossal expansion of their partnership with Amazon Web Services (AWS). This multi-faceted agreement directly addresses the compute deficit plaguing Anthropic's ambitious "flywheel" model, injecting unprecedented resources into the struggling AI developer. The scale of this commitment underscores a profound strategic realignment for both companies.

Central to the deal is a commitment to secure up to 5 gigawatts (GW) of compute capacity for Anthropic. This enormous power will fuel the training and inference demands of future Claude models, moving beyond the current constraints. Anthropic gains crucial access to AWS's cutting-edge AI silicon, including next-generation Trainium2 and the upcoming Trainium3 chips, purpose-built for high-performance machine learning.

The financial scope of this collaboration is staggering, representing an estimated commitment exceeding $100 billion over the long term. This figure encompasses cloud credits, infrastructure development, and joint go-to-market strategies, solidifying Anthropic's position as a foundational partner within the AWS ecosystem. Such an investment signals Amazon's deep integration of Anthropic's technology into its future services.

Beyond the compute and cloud credits, Amazon made a direct multi-billion dollar investment in Anthropic, signaling unwavering confidence in the AI startup's trajectory. Amazon initially invested $1.25 billion in September 2023, then followed up with an additional $2.75 billion in March 2024, completing its $4 billion commitment. This cash injection provides Anthropic with vital capital for research, talent acquisition, and further infrastructure build-out, insulating it from immediate financial pressures; for more details on this significant capital infusion, read Amazon invests another $2.75 billion in AI startup Anthropic.

This massive bet positions Anthropic as a cornerstone of Amazon's generative AI strategy, ensuring AWS remains competitive against rivals like Microsoft Azure and Google Cloud. For Anthropic, the Amazon partnership provides not just the compute power it desperately needs but also unparalleled distribution channels and enterprise reach. The deal transforms Anthropic from a compute-starved innovator into a heavily backed industry player, poised for a significant comeback.

Compute crunch, initially a severe handicap for Anthropic, paradoxically transformed into a powerful strategic advantage. This existential threat, forcing Claude AI to implement stricter usage limits during peak hours due to demand outstripping compute, spurred the company to forge deep, multi-faceted alliances with the world’s most formidable cloud and hardware providers. Anthropic did not merely purchase capacity; it engineered profound, long-term partnerships.

Unlike competitors who might tether themselves to a singular cloud provider, like OpenAI with Microsoft Azure, or endeavor to construct their own colossal data centers, Anthropic diversified its compute strategy. This approach minimized single-point-of-failure risks and maximized access to disparate, cutting-edge technologies. The company’s compute deficit became a catalyst for distributed resilience.

These critical agreements extend far beyond simple server access. The April 2026 deal with Google and Broadcom, for example, secures "multiple gigawatts of next-generation TPU capacity" coming online from 2027. This signifies preferential access to custom-designed, purpose-built AI hardware – not just commodity resources. Similarly, the expanded partnership with Amazon Web Services guarantees access to their custom silicon, including Trainium and Inferentia chips.

Such exclusive arrangements provide Anthropic with a crucial strategic moat. They ensure a steady, future-proof supply of specialized hardware essential for both training increasingly complex models and serving millions of users globally. This direct pipeline to advanced silicon, secured years in advance, offers a competitive edge, turning what was once a critical vulnerability into a foundational strength in the intense race for AI dominance. Access to custom AI hardware becomes a differentiating factor, securing Anthropic’s ability to innovate and scale.

Months of intense compute acquisition now manifest in Anthropic’s most formidable model to date: Claude Opus 4.7. This release directly validates the strategic alliances forged with Google, Broadcom, and Amazon Web Services, translating raw processing power into sophisticated AI capabilities. The massive investment in next-generation TPUs and AWS infrastructure underpins Claude Opus 4.7’s remarkable leap forward, proving the compute crunch was a catalyst, not a catastrophe.

Claude Opus 4.7 redefines complex software development and agentic coding, a critical advancement for Anthropic’s enterprise focus. The model demonstrates an unprecedented ability to plan, execute, and iterate on multi-step programming tasks, moving beyond simple code generation to autonomous problem-solving. Opus 4.7 can autonomously break down high-level specifications into actionable code, manage dependencies, and even self-correct errors through iterative refinement. This capability positions it as an invaluable tool for engineering teams grappling with intricate, large-scale projects.

Developers leveraging Opus 4.7 report significant advancements in several key areas, directly leveraging the enhanced compute capacity: - Autonomous feature development from high-level specifications, requiring deep contextual understanding. - Debugging and refactoring large, unfamiliar codebases with minimal human intervention. - Orchestrating multi-agent systems for intricate software architectures, demonstrating superior planning. - Generating production-ready code that adheres to specific style guides and performance metrics.

Internal benchmarks underscore Opus 4.7’s immediate impact. On the rigorous Agentic Coding Benchmark (ACB-v3), Claude Opus 4.7 achieved an impressive 92% success rate, outperforming previous models by over 20 percentage points. This benchmark specifically measures an AI’s capacity for independent problem-solving and strategic planning in complex coding environments, a direct benefit of increased training and inference compute.

Against its top competitors, Opus 4.7 establishes a new performance ceiling. In head-to-head evaluations, the model consistently outperformed GPT-5.4 on 8 out of 10 complex coding challenges involving distributed systems and novel algorithm design, tasks demanding extensive computational resources. Claude Opus 4.7 also surpassed Gemini 3.1 Pro by an average of 15% in multi-file project completion tasks, demonstrating superior contextual understanding and code coherence across extensive repositories. This immediate, measurable return on Anthropic’s compute investment signals a powerful resurgence, transforming a crisis into a strategic advantage.

Despite Matthew Berman’s dire predictions of a wobbling flywheel and a company in trouble, the financial markets delivered a starkly different verdict. Anthropic's strategic maneuvers, particularly its massive compute investments and deep cloud alliances, translated directly into unprecedented financial growth. This trajectory flatly contradicted any narrative of impending collapse.

Anthropic’s run-rate revenue exploded from an estimated $9 billion to over $30 billion in a mere matter of months. This meteoric rise underscored the potent demand for its enterprise-focused AI models, especially following the release of advanced iterations like the Claude 3 family. The market clearly validated Anthropic's aggressive pivot to enterprise contracts and paid subscriptions.

This financial momentum culminated in a stunning private market valuation. On Forge Global, the leading marketplace for private company stock, Anthropic's valuation surged to an astonishing $1 trillion. This monumental figure temporarily eclipsed even OpenAI, signaling an overwhelming investor confidence that silenced critics and redefined the competitive landscape of generative AI.

Investors recognized the long-term value of Anthropic’s compute-intensive strategy. The company transformed its initial compute crisis into a strategic moat, securing exclusive access to next-generation hardware and cloud infrastructure from Google and Amazon Web Services. This move ensured its ability to scale and iterate, a critical advantage in the fiercely competitive AI race.

The market’s enthusiastic response confirmed that Anthropic's calculated risks paid off handsomely. Far from being a company in distress, it emerged as a formidable contender, leveraging its deep tech prowess and strategic partnerships to capture significant market share. For further reading on their product advancements, see Introducing the next generation of AI: Claude 3. The trillion-dollar valuation was not merely a number; it represented a powerful endorsement of their vision and execution in the high-stakes world of artificial intelligence.

Anthropic has not merely weathered its initial compute crunch; it has forged an unshakeable position as a permanent pillar of the AI industry. What Matthew Berman once perceived as a wobbling flywheel now spins with formidable momentum, driven by strategic alliances and an unwavering commitment to its foundational principles. The company now stands as the third titan in a nascent AI triad, alongside OpenAI and Google.

This formidable standing emerged directly from the crucible of its infrastructure scarcity. Forced to innovate and collaborate deeply, Anthropic secured unprecedented deals with Google and Amazon Web Services, guaranteeing a future supply of cutting-edge hardware. These partnerships transformed a critical vulnerability into a strategic moat, enabling the recent launch of Claude 4.7 and validating a market capitalization that approaches a trillion dollars.

Anthropic’s core differentiator remains its deeply ingrained focus on AI safety and governed innovation. Unlike rivals often driven by a "move fast and break things" ethos, Anthropic champions Constitutional AI, a unique methodology that uses AI itself to align other AIs with a set of ethical principles. This iterative self-correction process aims to build models that are robustly helpful, harmless, and honest from the ground up.

This isn’t merely a philosophical stance; it’s a rigorous engineering discipline integrated into every stage of model development. Constitutional AI trains models to evaluate and revise their own responses against a predefined "constitution" of values and rules, minimizing harmful outputs without extensive human supervision. This approach provides a transparent and auditable framework for complex AI systems.

While this principled stance sometimes invites commercial friction—as seen in past disputes regarding sensitive military applications—it represents a calculated, long-term wager. Anthropic understands that as AI becomes increasingly pervasive, trust and safety will transition from desirable features to non-negotiable requirements. This ethical bedrock offers a distinct competitive advantage in a world grappling with AI's profound societal implications.

Company founders Dario and Daniela Amodei built Anthropic on the premise that responsible AI development is not a hindrance to progress but its ultimate enabler. Their bet is that a future demanding explainable, reliable, and fundamentally safe AI systems will overwhelmingly favor those who prioritized these values from day one. Anthropic aims to be the undisputed leader in trustworthy AI, making its "miscalculation" a pivotal moment in its ascent.

What is Anthropic's 'flywheel' model?

It's a self-reinforcing business cycle where Anthropic sells its coding AI to enterprises, uses the code data generated to train an even better model, which in turn attracts more enterprise customers.

Why did Anthropic face a 'compute crisis'?

An explosive surge in new users, partly from users leaving competitors, led to demand for their Claude AI models outpacing their available computing infrastructure, causing usage limits during peak times.

How is Anthropic solving its compute shortage?

Anthropic is securing massive, long-term computing capacity through strategic partnerships, including a $100 billion commitment with Amazon for AWS services and a major deal with Google and Broadcom for next-gen TPUs.

Is Anthropic still a major competitor to OpenAI?

Yes. Despite infrastructure challenges, Anthropic has secured massive funding, is growing revenue exponentially, and its valuation on private secondary markets has even surpassed OpenAI's at times. Its focus on AI safety provides a key differentiator.