BLS chief fired after weak jobs report as AI capex hits record $102.5B quarterly

Summary

The head of the Bureau of Labor Statistics was fired after the agency released weak jobs data on Friday. The move sparked debate over whether political pressure drove the decision or legitimate data problems warranted leadership change.

The employment report showed job additions at their slowest pace since 2010 over the past three months. The previous two months were revised downward. Most job growth is concentrated in healthcare. Unemployment for college graduates over 25 is rising fast and now sits higher than any year since 2014. Spending on services, especially travel and tourism, has declined for three straight months, the first such streak since 2008. The economic momentum that propelled tech stocks higher on Thursday evaporated as these labor market cracks surfaced.

Ray Dalio framed the firing through a data-quality lens rather than political theater. The BLS's process for making employment estimates is obsolete and error-prone with no credible plan to fix it. The huge revisions on Friday proved the point, as government numbers swung toward private estimates that were materially better. Dalio says that if Trump fired the BLS chief purely because he disliked the numbers, that would signal institutional decay. Leaders manipulating economic data to suit political goals is a classic sign of the loss of a functioning system with the rule of law and checks and balances. But he also argues the BLS genuinely needs overhaul because the data infrastructure is backward.

Private estimates already outperform government releases, which raises a structural question: why not modernize labor statistics using the vast datasets held by tech platforms? Google, Meta, and other large networks have granular information on location, employment status, and economic activity that could improve accuracy. The friction is privacy; some people resist government visibility into their employment. But the gap between what private-sector analysts can infer and what BLS publishes suggests the current survey-based system is obsolete.

AI capex hits record quarterly run rate

Amid economic weakness, the Magnificent Seven tech firms spent $102.5 billion on capital expenditure in their most recent quarter, according to Wall Street Journal reporting. Meta, Alphabet, Microsoft, and Amazon accounted for nearly all of it. Apple, Nvidia, and Tesla combined contributed only $6.7 billion.

The scale is staggering relative to revenue. For Microsoft and Meta, capex now exceeds one-third of total sales. Inference compute—running trained models on user inputs—now represents the largest operational cost for advanced AI systems and is driving most ongoing performance gains. Yet the math does not yet close. The AI industry is generating roughly $10 billion in quarterly revenue (OpenAI at $3 billion, Anthropic at $1 billion, plus core AI workloads at hyperscalers), while investing $100 billion per quarter. There is no sign that GPT-4 or similar models will generate less than $1 billion monthly, which provides some comfort on the demand side. But the revenue-to-capex ratio is wildly misaligned.

AI infrastructure spending as a percentage of US GDP has already surpassed the dotcom boom's telecom investment and is still climbing, according to Paul Kadrski's calculations. Capex on AI contributed more to US economic growth in the past two quarters than all consumer spending combined, according to Neil Duda, head of economic research at Renaissance Macro Research. In other words, AI infrastructure investment is propping up growth even as the broader labor market softens.

The comparison to prior infrastructure booms is instructive. Railroads in the 1870s and telecom fiber in the 1990s both led to booms followed by devastating busts: the 1873 panic and the 2000 dot-com crash. In both cases, companies overbuild infrastructure that outpaces demand, expectations reset, and debt cannot be repaid. But the outcome was not permanent ruin. Google found dark fiber left behind by the telecom crash and built the internet infrastructure of the 2010s on the cheap. Future companies may similarly benefit from surplus capacity if the AI buildout crashes.

The risk is financial rather than mere shareholder loss. Stock and bond crashes cause dispersed losses; debt-driven bubbles cause systemic crises. A wave of private credit funding the AI capex boom introduces leverage. If hyperscalers slow investment because returns disappoint, the lenders—banks and private credit funds—face losses they cannot absorb. That could trigger a financial crisis even if the underlying tech remains valuable.