Code: The Gemstone of AI

Throughout history, civilizations have competed not merely for land, metals, or trade routes but for symbols of power. In ancient China, jade occupied a singular position: not simply as a stone, but as an object laden with moral, political, and cultural significance. During the Han Dynasty, exceptionally fine varieties, such as "mutton-fat jade," were reserved for elites. To lack it signified weakness.

It is remarkable how history echoes in this era. Watching the artificial intelligence race unfold in recent years, I have asked myself which asset plays the role of jade: the scarce material whose possession separates the powerful from the rest. The answer, I believe, is code.

Anthropic, founded in 2021 by the Amodei siblings, offers the clearest evidence. On May 28, 2026, the company closed a US$65 billion Series H round at a US$965 billion post-money valuation (surpassing OpenAI to become the most valuable private AI company in the world) and reported that its run-rate revenue had crossed $47 billion, up from $30 billion just weeks earlier and roughly $10 billion the previous year. What distinguishes it from OpenAI, Google, Meta, and the Chinese tech giants? The answer is deceptively straightforward: Anthropic concentrated its efforts on code and largely eschewed the consumer entertainment market (videos, memes, images), which, while popular, has proven far harder to monetize.

Why Code? The Strategic Logic

The market context explains the bet. Grand View Research valued the global AI code tools market at US$4.86 billion in 2023, projecting US$26.03 billion by 2030 at a 27.1% compound annual growth rate, while SNS Insider extends the horizon to US$37.34 billion by 2032. The methodologies differ, but the trajectory is consistent: this is one of the fastest-expanding categories in enterprise software.

Within that landscape, Claude Code  (Anthropic's agentic coding tool, launched publicly in May 2025) has compressed a decade of typical product growth into months. Its run-rate revenue passed US$500 million by September 2025, US$1 billion by November 2025, and US$2.5 billion by February 2026. For comparison, GitHub Copilot, the category's pioneer, took roughly three years to reach its first billion. Anthropic now serves more than 300,000 business customers, over 500 of whom spend in excess of US$1 million annually, and enterprise use represents more than half of Claude Code's revenue.

The demand is not coming. It is already here.

The survey data reveals who is driving it. According to the JetBrains 2026 Developer Ecosystem Survey, 46% of developers with ten or more years of experience prefer Claude Code, versus 9% for Copilot; the Pragmatic Engineer survey of February 2026 found that 71% of developers who regularly deploy AI agents use it as their primary tool and that adoption at small companies reaches 75%. Meanwhile, the incumbent's position eroded: GitHub Copilot's share among professional developers fell from 67% to 51% in the Stack Overflow Developer Survey, while Claude Code's workplace adoption jumped from 3% in April 2025 to 18% by January 2026.

The Moat: Intelligence at the Right Altitude

Anthropic's competitive moat is not derived from being first. Copilot launched in 2021 with a multi-year head start. It is a product of architectural depth. Claude Code's agentic capabilities, anchored by the extended context and multistep reasoning of the Claude Opus family, allow it to hold the scope of a software project in working memory and make coherent, long-horizon decisions. This matters enormously at the enterprise level, where the unit of work is rarely a single function but a feature shipped across dozens of interdependent files.

The economics complete the argument. There were 28.7 million software developers worldwide in 2024, a population projected to reach 45 million by 2030. Stack Overflow's 2025 survey found that 84% of professional developers now use or plan to use AI coding tools, with average self-reported savings of approximately 3.6 hours per week. Priced against the U.S. median software developer salary of roughly US$130,000 (BLS), that saving implies a productivity gain near US$11,000 per developer per year against subscriptions costing US$1,000–US$2,000 annually. The willingness to pay is structurally embedded in the customer's own cost savings. Code is the singular domain where AI value is immediately quantifiable and tied to tangible business output.

Whoever accelerates the builders collects a toll on everything that gets built.

Two caveats deserve honest recognition. First, the headline figures are vendor-stated: neither Anthropic nor OpenAI has published audited accounts, so the revenue comparisons circulating in the business press are indicative rather than like-for-like measures. Second, the competitive response is underway: Copilot still leads adoption in enterprises above 10,000 employees, at 56%, defended by Microsoft's compliance infrastructure and procurement inertia, while Cursor and OpenAI's Codex compete fiercely for the same developers (https://blog.getbind.co/claude-code-vs-github-copilot-in-2026-an-honest-data-driven-comparison/).

Dominance in this market is rented, not owned.

What This Means South of the Data Centers

For Latin America, the strategic reading begins with an uncomfortable ratio. According to the Latin American Artificial Intelligence Index (ILIA 2025, CENIA/ECLAC), the region accounts for 6.6% of global GDP but captures only 1.12% of global AI investment. Against a world adding developers by the millions, that gap measures how far the region remains a consumer, rather than a creator, of the new productivity.

The regional map is instructive. Chile leads the ILIA 2025 for the third consecutive year with 70.5 points, followed by Brazil (67.3) and Uruguay (62.3), the region's three "pioneers" while Mexico sits in the "adopter" group alongside Colombia and Argentina. For Mexico specifically, the index carries a double signal: it is one of only four countries in the region (along with Brazil, Chile, and Colombia) with a robust data center industry, yet its position in specialized talent and research remains mid-table. In other words, Mexico is building the physical layer of the AI economy faster than the human layer that would let it capture the value running on top.

That is precisely where the code economy opens a door. As with jade, discovering the raw material is insufficient. One requires master craftsmen (developers) who, through patience, sculpt the stone into optimal form: algorithms, models, agents, and other value-generating structures. Unlike frontier model training, which demands billions in compute, participating in the software value chain demands developers, and developers are made in classrooms. Each of the millions of engineers the world will add by 2030 represents a recurring unit of economic value for whoever employs them, and AI tools are raising, not lowering, the return on that talent. For Mexican and Latin American students weighing their options, the signal is manifest: pursuing STEM education is no longer an abstract academic wager but a concrete pathway into the sector where new productivity is being defined. It is not enough to consume the tools; the region must also form the people who wield them.

Code as Strategy, Not Feature

Anthropic's trajectory illuminates a principle that will define the AI economy for the rest of this decade: the highest-value application of general-purpose AI is the one that sits closest to the production of other software. Code is not simply a use case; it is the economic substrate through which AI compresses the cost of building everything else. Every application built faster and cheaper because of an AI coding agent represents an invisible tax collected on the entire technology economy.

Looking ahead, the question for Latin America is not whether to buy the gemstone, but whether to train the craftsmen. Whoever controls the tools with which software is written (and whoever educates the people who use them) controls the next chapter of the digital economy.