Anthropic’s Daniela Amodei on Efficiency, Safety, and the Compute Arms Race

The defining characteristic of the current AI landscape is not just exponential growth, but the staggering capital required to fuel it. Amidst rivals announcing trillion-dollar commitments to compute infrastructure, Daniela Amodei, President and co-founder of Anthropic, champions a philosophy of algorithmic efficiency and safety-first development. This disciplined approach suggests that the next phase of the AI race will be won not simply by brute force scale, but by who can deliver the most capability per dollar of compute.

Amodei spoke with CNBC’s MacKenzie Sigalos about Anthropic’s unique origins, the inherent tension between rapid development and robust safety, and the strategic financial roadmap that includes a potential 2026 IPO. The company’s founding narrative, which began in late 2020/early 2021, involved a group of co-founders, many of whom were siblings or long-time colleagues from OpenAI, gathering in the midst of the pandemic to build a technology company focused on transformative AI but with "a very intense sort of focus on making the system safe and reliable." Amodei characterized the departure from their previous organization not as "running away from something," but as "running towards something," specifically a clear vision for a company where safety was central to the mission.

This founding principle, baked into Anthropic’s structure as a Public Benefit Corporation (PBC), has shaped its strategic spending. While Amodei acknowledges the industry’s high capital demands—especially for training frontier large language models (LLMs)—she notes that Anthropic has consistently achieved high performance with a fraction of the resources deployed by its largest competitors. The belief underpinning this strategy is that safety and reliability are not simply regulatory hurdles or ethical overhead; they are core features that provide economic value and competitive advantage. Amodei stated that the company operates under the belief that safety and reliability "actually were correlated and that they went together."

The focus on technical safety is paramount, driven by the rapid, often unpredictable, acceleration of model capabilities. Amodei discussed the urgent necessity for technical safety work, such as mechanistic interpretability and Constitutional AI, which builds guardrails directly into the models. This research, much of which Anthropic publishes publicly, serves as a proactive measure against potential societal risks, including the weaponization of AI systems—a risk the company takes seriously, given the observed misuse of models in cyber espionage campaigns.

Amodei noted that the pace of progress continues to surprise even the pioneers of scaling laws. "The exponential continues until it doesn’t," she remarked, highlighting the persistent challenge of forecasting technological leaps. The company’s recent infrastructure commitments, including a massive $50 billion planned investment, reflect the reality that maintaining a frontier position requires scaling hardware access dramatically. Anthropic aims to be available across all major cloud providers—Amazon, Google, and Microsoft—a strategy that balances competition with the necessity of accessing compute resources and serving a diverse enterprise customer base already reliant on those cloud platforms.

For Anthropic, the strategic prioritization of business-to-business (B2B) enterprise customers has been critical. Unlike competitors who initially chased consumer adoption, Anthropic recognized that enterprises inherently value reliability, security, and controlled deployment—the very characteristics prioritized by their safety-focused development process. This alignment means that an inability to access a safe, high-performing model like Claude is seen by businesses as detrimental to their own operations. Amodei summarized the internal ethos: "We’re really here to do the work... we’re a Public Benefit Corporation, we care about providing value for our customers."