Economics FAQ

President Trump ordered federal agencies to cease using Anthropic's technology on February 27, 2026, after CEO Dario Amodei refused to strip safety constraints from Claude's Pentagon deployment. Anthropic maintained prohibitions on mass domestic surveillance and fully autonomous weapons. Defense Secretary Pete Hegseth then designated Anthropic a 'Supply-Chain Risk to National Security,' a label previously reserved for foreign adversaries like Huawei and never before applied to an American company.

The FY2026 DoD budget request earmarks $13.4 billion for AI and autonomous systems, a roughly 7x increase from the $1.8 billion allocated in FY2025. This marks the first time AI has its own standalone line item inside the defense budget. The total defense request is $892.6 billion. For context, Anthropic's entire annualized revenue as of February 2026 was approximately $14 billion.

Both companies stated similar red lines: no mass domestic surveillance and no fully autonomous weapons. The key difference was framing. Anthropic refused to let the military use its models across 'all lawful use cases' without explicit restrictions. OpenAI agreed to deploy on the Pentagon's classified network while retaining control over technical safeguards, model selection, and deployment environments, limiting deployment to cloud rather than edge systems.

In July 1993, Secretary of Defense Les Aspin and Deputy Secretary William Perry invited defense CEOs to dinner at the Pentagon and told them Cold War budget cuts meant most would not survive. Norman Augustine of Martin Marietta named it the Last Supper. Within four years, 51 prime defense contractors consolidated into five: Lockheed Martin, Boeing, Raytheon, General Dynamics, and Northrop Grumman. Between 2011 and 2015, an additional 17,000 U.S. companies exited the defense industry.

Both, and increasingly the former. Palantir posted $4.48 billion in FY2025 revenue with 53.7% from government contracts, down from a peak of 58.2% in 2021 as its commercial AI platform gained traction. Its $10 billion U.S. Army Enterprise Agreement consolidated 75 existing software contracts. At a $320 billion market cap, Palantir is now worth nearly twice Boeing, making it the clearest example of a tech company operating as a defense prime.

IDIQ (Indefinite Delivery, Indefinite Quantity) contracts account for roughly 56% of DoD contract award dollars and run five years with extension options. They matter because once an AI company is embedded in classified systems with a security-cleared workforce, switching costs become close to prohibitive. Palantir's Maven Smart System contract, for example, started at $480 million and expanded to nearly $1.3 billion through 2029. A competitor cannot simply offer better inference speed to displace an incumbent with IDIQ access.

The sticker price is $8 million for a 30-second spot, with premium positions reaching $10 million. But the fully loaded cost is $16–23 million once you add production ($1–4M), celebrity talent ($1–5M), and mandatory companion buys on the same network ($7–10M). Some estimates reach $40–50 million when including agency fees, music licensing, and digital activation.

The evidence is genuinely mixed. Stanford researchers Hartmann and Klapper found Budweiser earned a 172% ROI from its Super Bowl ads. But Bridgewise found that a portfolio of Super Bowl advertisers underperformed the S&P 500 by 9.2% after six months. Kantar reports $4.60–$5.20 return per dollar invested. The answer depends heavily on category exclusivity: when two competing brands both advertise, neither gains incremental profit.

Stanford research showed that when two competing brands both advertise in the same Super Bowl, neither gains incremental profit because the effects cancel out. Yet both rationally choose to spend because opting out concedes the benefit to a competitor. This creates a collectively suboptimal but individually rational equilibrium that the NFL exploits to command rising prices.

At $8 million reaching roughly 125 million viewers, the Super Bowl's effective CPM is around $63–65. Standard primetime TV runs $20–30, streaming TV runs $15–35, and TikTok runs $5–10. The premium buys the engagement factor: EDO estimates a single Super Bowl ad generates the same brand-search engagement as 1,056 typical primetime ads.

The NFL controls a structural scarcity: the Super Bowl is the last true monoculture event in American media, reaching 125+ million simultaneous viewers in an era of fragmented attention. Inventory sells out months in advance. The prisoner's dilemma among advertisers prevents collective price resistance. And the price itself signals legitimacy, creating a Veblen good dynamic where high cost is part of the value proposition.

In 2025, NFL games accounted for 83–84 of the top 100 most-watched U.S. telecasts. The Super Bowl drew 127.7 million average viewers, roughly 2x the most-watched presidential debate and 6x the Oscars. No other single moment reaches such a broad cross-section of America simultaneously, making it the last true mass-reach event in an increasingly fragmented media environment.

Cornell research shows households with a GLP-1 user cut grocery spending by 5.3% within six months, with high-income households dropping 8.2%. Fast food spending falls 8.0%. These users aren't switching brands; they're simply eating less.

Savory snacks see the largest decline at 10.1%, followed by sweets, baked goods, and cookies. Even staples like meat, eggs, and bread decline. Yogurt is the only category showing a statistically significant increase, with fresh fruit and nutrition bars trending up slightly.

About 34% of users discontinue within the sample period. When they stop, spending doesn't just return to baseline; it becomes less healthy. Candy and chocolate purchases rise 11.4% above pre-adoption levels, suggesting the drugs suppress appetite biologically without teaching new habits.

Higher-income households show even steeper spending declines (8.2% vs 5.3% average) and are more likely to use GLP-1 medications for weight loss rather than diabetes. They're also the most profitable customers for fast food chains, creating a "double whammy" where companies lose their highest-margin customers.

The AI productivity paradox describes how tools that make individual tasks faster often increase total workload rather than freeing up time. Research shows 77% of employees say AI tools have added to their workload, and workers in AI-exposed occupations now work roughly 3 extra hours per week while leisure time has dropped by the same amount.

The Jevons paradox, observed in 1865 when more efficient steam engines increased coal consumption rather than reducing it, applies to AI in that efficiency expands what we're expected to do. When you can build an app in a weekend that used to take months, you don't build one—you build six. The friction that once protected us from infinite expectations evaporates.

Keynes predicted a 15-hour work week by now, and we got the productivity gains he anticipated—yet we work longer hours than ever. Only 21% of employees use time saved by AI for personal life; the rest reinvest it into work. When capability expands, so does the definition of "enough," and the bar rises accordingly.

Parkinson's Law states that work expands to fill the time available. The AI corollary is that work expands to fill capabilities available. More capability means more possibility—and more obligation. In competitive environments, someone who uses that expanded capability while you rest will outrun you.