At Anthropic’s San Francisco headquarters, President and co-founder Daniela Amodei frequently returns to a core principle guiding the company’s strategy: do more with less. This philosophy stands in contrast to the dominant trend in Silicon Valley, where the largest AI labs are betting heavily on scale—racing to secure massive compute resources, lock in chip supply years in advance, and build sprawling data centers in the belief that sheer size will determine who wins the AI race. OpenAI exemplifies this approach, with reported commitments of around $1.4 trillion in compute and infrastructure. The company is constructing massive data center campuses and securing next-generation chips at an unprecedented pace. In this environment, Anthropic is taking a different path—one built on efficiency, smarter algorithms, and strategic resource use. Despite having a fraction of the compute and capital of its competitors, Anthropic has consistently produced some of the most powerful and performant models in recent years. “We’ve always aimed to be as judicious as possible with the resources we have,” Amodei said. “And yet, we’ve managed to stay at the frontier.” The company was co-founded by Daniela and her brother Dario Amodei, who previously helped shape the very scaling paradigm that now drives the industry. That model—where increasing compute, data, and model size leads to predictable performance gains—has become the foundation of the AI arms race, justifying massive investments in hardware, cloud capacity, and long-term infrastructure. But even as Anthropic has benefited from this framework, it is now challenging the idea that scale alone is the key to success. The company is focusing on high-quality training data, advanced post-training techniques to boost reasoning, and product design that reduces runtime costs and improves enterprise adoption. These efforts aim to make models not just more capable, but more practical and sustainable for real-world use. Amodei acknowledges that Anthropic is not operating on a shoestring. The company has about $100 billion in compute commitments and expects those needs to grow as it continues to push the frontier. “The compute requirements for the future are very large,” she said. “We will need more to keep up.” Still, she questions the comparability of the numbers being thrown around. Many deals are structured in ways that don’t allow for direct comparison, and the pressure to commit early to secure hardware can lead to overbuilding. “We’ve been surprised, even as the people who helped create the scaling laws,” she said. “The exponential continues until it doesn’t. And every year, we think it can’t go on—yet it does.” This tension between technological progress and economic reality is central to the current moment. While the technology curve shows no sign of slowing, the real challenge lies in how quickly businesses and individuals can adopt and integrate these tools. “It takes time for technology to be used in real workflows,” Amodei said. “The real question is: How fast can organizations leverage these capabilities?” Anthropic’s enterprise-first strategy has proven effective. Revenue has grown tenfold each year for three consecutive years. Its Claude model is available across major cloud platforms, including AWS, Google Cloud, and Microsoft Azure—reflecting customer demand for flexibility and choice. This multicloud approach allows Anthropic to remain agile, shifting workloads based on cost, availability, and client needs, without being locked into a single infrastructure bet. As 2026 begins, both Anthropic and OpenAI are preparing for public markets, though neither has announced an IPO. They are building out finance, governance, and operational systems to meet the demands of public scrutiny. At the same time, they continue raising capital and securing compute to fund future development. The key question is whether the market will continue to reward scale or begin to value efficiency. If investors start demanding more sustainable spending, Anthropic’s “do more with less” approach could give it a critical edge. The real test isn’t whether scaling works—but whether it’s the only way forward. As Amodei put it: “The exponential continues until it doesn’t.” The next phase of the AI race will depend on what happens when it finally stops.