📈 Data to start your week
$1 trillion boom ↑ GLP-1s eat pharma ↑ China carbon ↓
Hi all,
Here’s your Monday round-up of data driving conversations this week in less than 250 words.
Let’s go!
AI boom ↑ US data center and software investment now exceeds $1 trillion annualized (3.5% of GDP).1 See our analysis.
Prompt assist ↑ On a realistic business task, AI cut the performance gap between less‑ and more-educated adults by three-quarters.
Appetite for profit ↑ The global GLP-1 drug market is set to make up almost 16% of the global pharmaceutical market by 2030, at ~$268 billion.
Grid hunger ↑ IEA forecasts global electricity demand through 2030 to grow 50% faster than the past decade, as consumption from industry, electric vehicles, air conditioning and data centers increases.
Peak China carbon ↓ China’s CO2 emissions fell 0.3% in 2025 – flat or falling for 21 straight months.
Can’t beat the sun ↑ More than 99% of all new US electricity generating capacity in 2026 will be solar, wind and storage, equivalent to adding 70 nuclear plants.
The reasoning jump ↑ Gemini 3 Deep Think scored a record 84.6% on ARC-AGI-2.2 ARC-AGI-3 drops in March.
Europe’s AI alternative ↑ Mistral shot over $400 million annualized revenue, up more than 20x in the last year.
Generation AI ↑ People between 16 and 24 years old in the EU are twice as likely to use generative AI (64%) as the general population (32.7%).
Today’s edition is brought to you by Polymarket
Who has the best AI model by the end of February?
*Disclaimer: Polymarket is a real‑money prediction market. Trading involves risk and may not be available in all jurisdictions. Nothing here is investment, tax, or legal advice. Please check your local regulations and only stake what you can afford to lose.
Thanks for reading!
The multi‑hundred‑billion capital expenditure plans reported by big tech firms are global capex numbers that also include other assets (such as warehouses, logistics infrastructure, operating costs and non‑data‑center facilities that aren't included in the buckets above), and so they cannot be directly compared to the US‑only fixed‑investment values in the graph here.
ARC-AGI benchmarks test how well AI can solve novel reasoning puzzles that humans generally have no issue with.
"The task is a general (not domain specific) exercise..."
Is this study of a task or tasks? There are few recent articles about cogitative or decision overload as the number of tasks a AI agent completes increases.
Is tasking via large scale agent pool then pushing the human productivity bottleneck to management skill (either via experience or ability)?
The $1T annual AI investment number caught me - that's staggering scale for infrastructure that mostly doesn't exist yet.
I've been tracking who's actually winning in the agent space vs who's getting funded. The gap is massive. Most investment goes to infrastructure plays (models, chips) while the operational value compounds at the application layer.
Wrote about the Feb 2026 data: https://thoughts.jock.pl/p/ai-agent-landscape-feb-2026-data
The educational performance gap stat (75% reduction via AI) is interesting but feels optimistic. What's the implementation complexity? Most gains require operational changes, not just AI access.