TAI #194: AI Goes Macro; Job Loss Fears, Military Usage, OpenAI $110B Raise

Author(s): Towards AI Editorial Team Originally published on Towards AI. What happened this week in AI by Louie This week brought a series of developments that signal AI is quickly becoming more than just a technology story: AI’s revenue, its politics, and its labor market consequences are now operating at a scale that reshapes the global economy and the geopolitical order in real, measurable ways. AI, the Pentagon, and the Claude Surge. AI is increasingly critical to US military operations. OpenAI signed a contract with the Department of Defense to deploy its models on classified networks. Hours later, the Trump administration designated Anthropic a “supply chain risk” and directed agencies to stop using Claude, widely interpreted as retaliation for Anthropic’s refusal to lift its safety guardrails for unrestricted military use. Meanwhile, reports emerged that Claude was allegedly used, together with Palantir, during the capture of Venezuela’s then-president Nicolás Maduro in January and again to assist with intelligence assessment during strikes against Iran. I agree with the red lines Anthropic has laid out: no mass surveillance, no autonomous weapons without a human in the loop. Dario Amodei seems more serious about enforcing those boundaries than any other lab CEO, and his willingness to absorb real commercial and political cost to hold that line is notable. That said, the broader question is genuinely complex. Should unelected AI CEOs be drawing the boundaries of how military AI gets used? In principle, that is a job for elected governments. But existing laws were not written with these AI capabilities in mind, and governments have shown little urgency to update them. Until they do, the defaults are being set by a handful of companies in San Francisco. Public backlash against OpenAI’s Pentagon deal appears to have driven a spike in downloads of Claude. Anthropic’s app hit number one on the Apple App Store, and the resulting surge in demand contributed to a major Claude outage on Monday that lasted nearly three hours, following a minor disruption on February 28. GPU and inference capacity are already binding constraints, and we are nowhere near the usage levels many AI economic scenarios assume. OpenAI Raises $110 Billion. OpenAI closed a $110 billion funding round, the largest private financing in history, from Amazon ($50B), Nvidia ($30B), and SoftBank ($30B), at a pre-money valuation of $730 billion. Capital flowing into AI infrastructure is now reaching a scale that shows up in macro aggregates. Between this fundraise, continued $150–200 billion in hyperscaler data center capex per quarter, and SoftBank’s Stargate commitments, AI investment is becoming a material driver of GDP in its own right. The question is whether the productivity gains this infrastructure enables will circulate broadly through the economy, or concentrate in a handful of firms. Citrini’s “2028 Global Intelligence Crisis” and the AI Job Loss Debate. A blog post from CitriniResearch titled “The 2028 Global Intelligence Crisis” went extremely viral recently, reportedly accumulating around 16 million views. The piece is written as a fictional macro memo from June 2028, looking back on how AI-driven white-collar job displacement triggered a cascade of economic and financial consequences: mass layoffs leading to reduced consumer spending, a collapsing SaaS sector, private credit defaults, and eventually stress in the $13 trillion US mortgage market as high-income borrowers lose their jobs. The thesis: AI capabilities improve, companies lay off white-collar workers and reinvest savings into more AI; displaced workers spend less; companies under revenue pressure invest even more in AI to cut costs; and the cycle accelerates. Citrini calls this the “human intelligence displacement spiral.” The piece also describes how agentic commerce erodes the moats of intermediary businesses (DoorDash, Mastercard, insurance brokers, real estate agents) as AI agents are put in charge of your shopping, optimizing for price rather than habit, effectively destroying the “friction premium” that underpins trillions of dollars of enterprise value. Stocks named in the essay, including Uber, DoorDash, American Express, and Mastercard, sold off in the days following the post’s spread. IBM dropped sharply. Reception from economists was mixed, and the piece got plenty of pushback, but the scenario clearly struck a nerve because it stitched together several anxieties investors already had: AI as a margin tailwind in the short run, and AI as a demand and business-model headwind if labor income gets hit hard enough. I think the Citrini thesis is a feasible, low-probability possibility, but with some important caveats. The stock market story and the economic story are two different things. Global labor income is roughly $60 trillion, compared with current S&P 500 profits of $2–2.5 trillion. There is a huge amount of slack in AI-beneficiary names soaking up profit from labor, leading to higher S&P levels, even if GDP falls significantly. The usual intuition that “stocks track the economy” can fail when the economy’s scarce factor shifts from labor to compute. In these scenarios, AI labs will likely have to keep spinning off divisions and vertical platforms to maintain some diversity in the indexes, because you cannot have 5–10 companies making up 90% of market capitalization without structural pressure to break them up. The “technological innovation destroys jobs and then creates even more” line does not hold as a default assumption this time. It has been right for two centuries because every new job required a human to perform it. With general-purpose AI, many of the “new categories” are also automatable, often faster than institutions can train for and professionalize them. There will definitely be human roles that appear or grow significantly for a while, but they may only be a fraction of what gets replaced. One scenario for job growth to offset job losses is if GDP grows multiple times its current level. That seems to be Elon Musk’s primary scenario: one new human job for every nine new AI jobs can still lead to full employment if the total economy is large enough. That is feasible. But the middle ground, where there are neither huge job losses nor an unprecedented economic boom, does not seem […]