Concrete Capital

Construction spending on data centers just hit $50B annualized. In 2018 it was $6B. That curve is not AI. The first half of it — 2018 through 2022 — is cloud. AWS, Azure, Google Cloud fighting for enterprise workload migration. The line moves but it is controlled, deliberate, cyclical infrastructure investment. Then 2023 happens and the line goes vertical. What changed in 2023 was not demand — it was the visibility of demand. Every hyperscaler suddenly had a 10 year reason to build as fast as physically possible. The $1.5T combined cloud backlog did not exist in 2022. The race to 10 gigawatts of AI compute did not exist in 2021. The construction spending curve is just the physical manifestation of commitments that were made in boardrooms starting in late 2022 and are now showing up in concrete, steel, and power infrastructure. $50B annualized is not even the top yet. Every hyperscaler has guided higher capex in 2026 than 2025…maybe the sky is the limit as the old saying goes…

🇺🇸🇨🇳 As Trump heads to China, it's time for a reality check on the economy. For a decade, the narrative was that China would overtake America. Every bank had a chart. Every strategist had a date. The date kept getting pushed back. Now it's gone entirely. $28.3T vs $17.7T. The gap is $10.6 trillion — and it is getting bigger every day. - United States — $28.3T. growing - China— $17.7T. stalling - Gap— $10.6T. wider than 2015 China's GDP peaked relative to the US in 2021 at roughly 77 cents on the dollar. It has fallen every year since. The property crisis wiped out $18 trillion in household wealth. The population is shrinking. Youth unemployment hit 21%. Capital is leaving. The yuan is weak. The convergence thesis that Wall Street sold for twenty years is dead. America innovated its way out of every crisis China's boosters said would end US dominance. The 2008 financial crisis. COVID. The debt ceiling. Each time the obituary was written. Each time the US economy came back larger. The AI boom — the biggest productivity revolution since electricity — is happening in California, not Shenzhen. The semiconductor ecosystem that powers the next century is in Arizona, Texas, and Ohio. The capital markets that fund it all are in New York. Trump is in Beijing to negotiate. That's the right move — you negotiate from strength, not fear. And the chart shows exactly what kind of strength he's negotiating from. America is not in decline. America is pulling away. The data has been saying this for three years. The narrative just hasn't caught up yet.

86.1 million downloads in one week!!! From March 15 through April 26, Anthropic's Claude Code was the dominant AI coding agent by weekly install volume—ranging between 9.9M and 13.1M downloads per week while OpenAI Codex sat in the 3.1M–4.1M range. That was the steady state for seven weeks. Then the week of May 3 happened. Codex downloads hit 86.1M — a 21× spike in a single week. The spike is a product launch effect — OpenAI released an upgraded Codex with expanded agentic capabilities—multistep autonomous coding, full pull request workflows, direct GitHub integration. Every engineering team that had been waiting for a credible autonomous coding agent from OpenAI installed it in the same week. What you're seeing in that bar is seven weeks of accumulated latent demand, released in one day. The question is what week two looks like. A 21× spike that reverts to 10–13M is a launch. One that holds above 30–40M is a category shift. Claude Code's 7.2M in the same week Codex spiked tells you the market isn't zero sum. Both products are growing at crazy rates. The developers who downloaded Codex this week are not the same people who stopped using Claude Code. The agentic coding market is expanding faster than either company can capture it. Neither side is losing yet. The war just started… We are not bullish enough folks. Jensen saying inference gonna go by a billion X, anthropic brings in billions in ARR in the first few months of the year…massive markets to say the least.