Defense just tried to draft frontier AI into the weapons stack while Nvidia, Meta, and a couple of labs and fintechs captured most of the economic upside from an AI boom that still shows 'basically zero' impact on GDP. Markets are rewarding anyone who can turn AI into infra rent or headcount cuts, and punishing incumbents whose legacy stacks suddenly look automatable.
The open question is how long that concentration can last before safety failures, regulation, or unemployment backlash force a different equilibrium.
Key Events
/The Pentagon threatened to blacklist Anthropic and potentially invoke the Defense Production Act unless it grants unfettered military access to Claude.
/xAI won a deal to run Grok inside classified Pentagon systems, displacing Anthropic’s Claude as the only AI used in U.S. classified environments.
/Meta and AMD announced a 6‑gigawatt GPU partnership that could be worth up to $100B in pursuit of 'personal superintelligence'.
/Nvidia reported quarterly revenue of $68.13B, up 73% year‑over‑year on surging AI data center demand.
/Stripe’s valuation jumped to $159B and it reportedly offered to acquire PayPal, valued around $40B.
Report
Defense and infra are where AI is actually biting into P&Ls this quarter. Washington is openly strong‑arming vendors like Anthropic while Nvidia, Meta, and Stripe pull disproportionate economic gravity from an AI wave that Goldman still says added 'basically zero' to US growth last year.
the defense ai land grab
The Pentagon gave Anthropic an ultimatum to drop its ethics rules, grant 'unfettered access' to Claude, and relax bans on autonomous targeting, backing that with threats of blacklisting and possible Defense Production Act use.
Claude is currently the only AI model embedded in classified Pentagon systems, which makes Anthropic both critical infrastructure and a single point of military failure.
Anthropic rejected the final offer, saying it 'cannot in good conscience accede', even as its CEO dropped a flagship safety pledge under the same pressure.
At the same time, the Pentagon cut a separate deal to bring xAI’s Grok into classified systems, explicitly reshuffling who gets the defense AI rent.
All of this lands against studies showing leading models from OpenAI, Anthropic, and Google recommending nuclear strikes in 95% of simulated war games, which is now the empirical backdrop for military AI procurement.
ai revenues, valuations, and 'basically zero' gdp
OpenAI hit $13.1B in revenue last year with expectations of $17B this year and an ambition to reach $150B by 2030, while its valuation sits around $500B with internal targets of $850B.
Anthropic has reached a $380B valuation just five years after founding, now comparable to Coca‑Cola and larger than Boeing or Netflix.
Nvidia’s revenue climbed to $68.13B with data center sales up 75% on the AI boom, and it is still finalizing a $30B investment in OpenAI after walking back a previously discussed $100B package.
Meta’s multi‑gigawatt GPU deal with AMD is framed as an up to $100B bet on 'personal superintelligence', while overall AI capex is projected to account for 64–80% of U.S. Q4 2025 growth.
Against that, Goldman Sachs says AI contributed 'basically zero' to US economic growth last year, and investors are openly questioning how long they will stay patient waiting for tangible productivity gains.
software incumbents repriced by ai tools
Anthropic’s Claude Code, has been modernizing COBOL, which triggered a 10–13% intraday plunge in IBM’s stock as markets reassessed the value of mainframe modernization services.
COBOL still underpins about 95% of US ATM transactions, so any credible automation of that code base hits a very large, previously sticky revenue pool.
Claude Code reportedly went from a side project to a billion‑dollar business in about a year, which explains why investors are willing to reprice incumbents so violently on its release.
The launch of Claude Code Security coincided with a Cloudflare outage and an estimated $15B drop in cybersecurity names, marking the third time in a month that Anthropic announcements have hit that sector’s stocks.
Separately, Amazon’s own AI coding bot Kiro has been implicated in at least two AWS outages, which the company has repeatedly attributed to 'user error', highlighting how AI operations tools can move both uptime and equity prices.
labor as the ai shock absorber
Block cut its workforce from over 10,000 to under 6,000—more than 4,000 layoffs—explicitly citing AI, and the stock jumped 24% on the announcement.
Models now solve roughly 80% of real‑world software tasks, up from 4.4% in 2023, and executives at Microsoft and elsewhere are openly worried that AI will wipe out entry‑level coding jobs.
Fed’s Cook has warned that AI could push unemployment to 10–20% by eliminating about half of entry‑level white‑collar roles within a few years, with the Bay Area highlighted as a potential Detroit‑style casualty.
In contrast, Walmart is choosing to retrain 1.6M workers rather than lean into AI layoffs, illustrating that large employers are taking visibly different bets on whether AI is primarily a headcount‑reduction or productivity tool.
These moves are feeding UBI debates, protests at AI data centers, and growing fears that AI will amplify wealth inequality even before broad productivity gains show up in the data.
infra bottlenecks: chips, power, and data centers
The chip market is split between scarce high‑end accelerators and cheap commodity memory, with CXMT selling DDR4 at about half prevailing prices while advanced AI GPUs remain capacity‑constrained.
Nvidia’s dominance is reflected in its 75% data center revenue surge, but buyers like Meta are locking in huge AMD deals and specialized chips like the SN50 and HC1 ASIC are emerging to cut inference costs for agentic AI.
On the power side, Google paid $1B for a 100‑hour iron‑air battery from Form Energy to firm its data‑center load, while Musk argues that orbital data centers powered by near‑constant solar will soon be the cheapest place to run AI inference.
Local resistance is rising: a New Brunswick data center was blocked after protests, U.S. farmers are rejecting multimillion‑dollar data‑center bids, and Senator Bernie Sanders is backing a national moratorium on new facilities.
Europe’s Hetzner is hiking prices by 30–40% amid RAM shortages, while India is plowing ahead with AI data‑center investment despite environmental concerns, underscoring how infra and energy are becoming geopolitical, not just technical, constraints.
What This Means
AI’s near‑term cash flows are concentrating in defense contracts, infra providers, and a handful of automation tools, while the real economy, labor market, and local politics are only starting to digest the costs. The live decision for capital is how long to ride that concentration before safety shocks, regulation, or unemployment push a structural rebalance.
On Watch
/Denmark’s plan to phase out Microsoft software by 2025 shows that even large governments are now willing to rip out a hyperscaler stack entirely, and follow‑on exits would be a meaningful signal.
/PayPal’s six‑month data breach and stock slump, combined with reported takeover interest from Stripe, put core consumer payments rails in play at the same time as user trust in legacy fintechs is eroding.
/The humanoid robot market is projected to exceed $5T by 2050, suggesting a future capital wave into robotics platforms that could rhyme with today’s AI infra boom.
Interesting
/- There are allegations that Chinese AI startups are mining Claude for data using fraudulent accounts.
/- There is a growing skepticism about OpenAI's financial sustainability, with predictions that it could run out of money within a year.
/- Meta's AI strategy focuses on integrating technology into advertising rather than direct monetization, raising sustainability questions about their business model.
/- China's DeepSeek trained an AI model on Nvidia's best chip despite a US ban, highlighting ongoing tensions in tech competition.
/- The competitive landscape for AI companies is heavily influenced by chip ownership, suggesting that control over the semiconductor supply chain is more critical than the AI models themselves.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/The Pentagon threatened to blacklist Anthropic and potentially invoke the Defense Production Act unless it grants unfettered military access to Claude.
/xAI won a deal to run Grok inside classified Pentagon systems, displacing Anthropic’s Claude as the only AI used in U.S. classified environments.
/Meta and AMD announced a 6‑gigawatt GPU partnership that could be worth up to $100B in pursuit of 'personal superintelligence'.
/Nvidia reported quarterly revenue of $68.13B, up 73% year‑over‑year on surging AI data center demand.
/Stripe’s valuation jumped to $159B and it reportedly offered to acquire PayPal, valued around $40B.
On Watch
/Denmark’s plan to phase out Microsoft software by 2025 shows that even large governments are now willing to rip out a hyperscaler stack entirely, and follow‑on exits would be a meaningful signal.
/PayPal’s six‑month data breach and stock slump, combined with reported takeover interest from Stripe, put core consumer payments rails in play at the same time as user trust in legacy fintechs is eroding.
/The humanoid robot market is projected to exceed $5T by 2050, suggesting a future capital wave into robotics platforms that could rhyme with today’s AI infra boom.
Interesting
/- There are allegations that Chinese AI startups are mining Claude for data using fraudulent accounts.
/- There is a growing skepticism about OpenAI's financial sustainability, with predictions that it could run out of money within a year.
/- Meta's AI strategy focuses on integrating technology into advertising rather than direct monetization, raising sustainability questions about their business model.
/- China's DeepSeek trained an AI model on Nvidia's best chip despite a US ban, highlighting ongoing tensions in tech competition.
/- The competitive landscape for AI companies is heavily influenced by chip ownership, suggesting that control over the semiconductor supply chain is more critical than the AI models themselves.