The OWID Brief: Beef in… | Our World in Data Summary

1. Key Themes

AI infrastructure spending is accelerating at an extraordinary pace

US data center construction spending has exploded since ChatGPT's release, marking one of the clearest quantifiable signals of the AI boom's physical footprint.

"between 2018 and 2022, spending on building data centers in the US hovered between $500 million and $1 billion per month (when adjusted for inflation). But since late 2022, large investments in AI have led to a rapid growth in data center construction"
"$4.4 billion was spent in June 2026 alone — a 5-fold increase compared to November 2022, when ChatGPT was released by OpenAI"
Notably, this figure understates true AI capex: "These figures cover privately-funded construction... They don't include the cost of IT hardware like servers and storage, which can represent a very large share of total investment."

AI adoption creates a productivity/learning tradeoff that depends on usage discipline, not the tool itself

A large-scale study on AI use among students reveals a nuanced result relevant to any knowledge-work context: AI speeds up task completion but can erode the underlying skill-building if it's used to shortcut effort.

"pupils who used AI saw their average homework score rise by 18%, and the time they spent on each assignment fell from 64 minutes to 45. But when exam time came, the same pupils scored 20% below their peers who had not used AI."
Critically, the effect is behavioral, not inherent: "the drop was concentrated among pupils who rushed their homework. Those who used AI but spent as long on assignments as non-users achieved broadly similar exam scores."

Domestic demand, not export markets, drives the largest environmental externalities of commodity production

Common narratives blame international demand (e.g., export-driven deforestation) for environmental harm, but the data show otherwise for Brazilian beef.

"beef stands out: it has driven 41% of deforestation this century. Two-thirds of this has happened in Brazil"
"around 85% of deforestation in the country was driven by demand in domestic markets. The other 15% was exported and spread across many countries."

Institutional and incentive failures, not scientific limits, explain stalled health progress

Medical R&D shows a bifurcated pattern — some diseases see dramatic progress while others stagnate for non-scientific reasons, a lesson applicable to any innovation-dependent sector.

"five-year survival for childhood leukemia rose from 15% before the 1970s to 85% now"
"The first malaria vaccine was developed in the 1990s but took decades to reach children. R&D spending on trachoma fell to zero in 2023. Ninety-five percent of rare diseases have no approved treatment at all."
"the biggest reasons behind these bottlenecks are funding, incentives, and institutions, rather than the underlying science. She points to advance market commitments and platform trials as fixes that have already worked."

2. Contrarian Perspectives

Beef-driven deforestation is mostly a domestic problem, not a globalization/trade problem. The conventional wrangling over export supply chains and international consumer boycotts misses that most deforestation serves local Brazilian demand: "around 85% of deforestation in the country was driven by demand in domestic markets."
AI's harm to learning isn't about the technology — it's about how people choose to use it. Rather than concluding AI tools are inherently detrimental to education (a common alarmist take), the underlying data suggest the tool is neutral and the behavior (rushing) is the causal factor: "Those who used AI but spent as long on assignments as non-users achieved broadly similar exam scores."
Democratic pessimism is largely a framing problem versus the long arc of history. Despite widespread dissatisfaction with democracy today, the long-run trend is one of major, non-obvious progress: "The world used to be deeply undemocratic, and despite setbacks, it remains much more democratic than it was only half a century ago."

3. Companies Identified

OpenAI — AI research/product company; referenced as the trigger point for the AI infrastructure buildout. Quote: "a 5-fold increase compared to November 2022, when ChatGPT was released by OpenAI."
Doubao and DeepSeek — Chinese AI models; cited as the tools adopted by the student cohort in the education study. Quote: "Around 80% of these pupils started using AI models such as Doubao and DeepSeek during that period."

4. People Identified

Hannah (Ritchie) — OWID researcher; author of the beef/deforestation article. Why mentioned: authored the piece analyzing Brazil's role in global deforestation. Quote: "Read Hannah's article."
Bastian, Esteban, and Lucas — OWID researchers; authors of the electoral democracy history article. Why mentioned: produced the long-run analysis of democratic development. Quote: "Read Bastian, Esteban, and Lucas's article."
David Strömberg, Victor Lei, and Wu Yanhui — Researchers who authored a June 2026 preprint on AI use among Chinese students. Why mentioned: conducted the large-scale study (27,000 pupils) on AI's effect on homework and exam performance. Quote: "tracked about 27,000 Chinese pupils aged 12 to 18 between late 2022 and 2025."
Saloni Dattani — Former OWID researcher, now writing for Works in Progress; gave a related TED talk. Why mentioned: argues health progress bottlenecks stem from institutional/incentive failures, not science. Quote: "writes in Works in Progress about how much medicine has improved but why more of that progress doesn't reach people."

5. Operating Insights

For builders in AI/data infrastructure: Physical capex data (construction spending) is a leading, quantifiable proxy for the scale of AI investment cycles and can be tracked via public sources like the US Census Bureau — useful for gauging where the infrastructure buildout stands relative to hype cycles: "The latest figures released by the US Census Bureau estimate that $4.4 billion was spent in June 2026 alone."
For companies deploying AI tools internally or in products: Speed gains from AI assistance can mask a hidden skill/quality debt; monitoring whether users maintain effort/time investment (not just output volume) is a proxy for whether the tool is enhancing or hollowing out capability.
For health/biotech and public-policy-adjacent operators: Mechanisms like "advance market commitments and platform trials" are cited as proven structural fixes for misaligned R&D incentives — a playbook worth studying for de-risking investment in neglected disease areas.

6. Overlooked Insights

The deforestation data implies that consumer-facing pressure campaigns (e.g., boycotts of Brazilian beef in importing countries) may have limited leverage compared to policy interventions targeting Brazil's domestic market — a distinction easily lost in the more visible "trade/export" narrative: "The other 15% was exported and spread across many countries."
The stalled progress on rare diseases and neglected conditions ("Ninety-five percent of rare diseases have no approved treatment at all") signals a potentially underexplored investment/policy frontier where incentive redesign (not new science) could unlock outsized returns — a whitespace opportunity implicit in Dattani's argument but not the article's main focus.