Who Sells the Pickaxes in the Last Gold Rush
Crescit amor nummi, quantum ipsa pecunia crevit.
The hunger for money grows as the money itself grows. Juvenal wrote it two thousand years ago, in his fourteenth Satire, watching Roman patricians hoard without end and without purpose. He couldn’t have imagined data centers, silicon chips cooled by helium, language models devouring electricity like blast furnaces — but he knew the fever. The fever is always the same.
📑 Contents
- 1. The $635 Billion Gold Rush
- 2. The Great Bifurcation
- 3. $31 Billion for a Blog Post
- 4. The Achilles’ Heel: Helium, Chips, and Geopolitics
- 5. The Tobacco Moment
- 6. When the Pentagon Comes Knocking
- 7. Cui Prodest: Mapping the Winners and the Losers
- Notes
- Bibliography
There comes a point in every gold rush when the miners stop looking at what they’re finding and start looking at what they’re spending. It isn’t the point when the gold runs out — it’s the point when the price of pickaxes exceeds the value of what you might extract. Economic history teaches that gold rushes don’t fail for lack of gold, but for excess of fever: when everyone digs, the cost of digging becomes the real business. And whoever sells the pickaxes ends up richer than whoever finds the nuggets.
The artificial intelligence race, in the first quarter of 2026, has reached precisely that inflection point. Not because AI has stopped working — it works all too well, well enough to crater century-old companies with a single blog post. But because the forces in play — capital, geopolitics, regulation — have knotted themselves into a complexity that no language model, however large, can forecast. And the map of winners and losers is being redrawn at a speed without precedent in the history of technology.
What follows is not a news roundup — it’s an attempt to trace the threads connecting a noble gas extracted in Qatar to the collapse of an IT giant on Wall Street, and to figure out who is actually winning this game. If you’re looking for the comforting narrative of linear progress — AI improving the world, one step at a time, in harmonious crescendo — you’ve clicked the wrong article. What follows is a story about power.
1. The $635 Billion Gold Rush
The number is so large it has stopped meaning anything: $635 billion.[1] That’s how much Microsoft, Amazon, Alphabet, and Meta will spend on AI infrastructure in 2026 — a 67% increase over the $381 billion spent in 2025. At the upper end of corporate guidance, the total approaches $665 billion.[2]
For context: Argentina’s GDP is just over $630 billion. Four private companies are spending more than the entire economy of one of the world’s twenty-five largest nations — every year — to buy chips, build data centers, and run models that, for now, no one knows how to monetize in proportion to the investment.
The distribution is uneven. Amazon leads at roughly $200 billion, followed by Alphabet ($175-185 billion), Microsoft ($145 billion), and Meta ($115-135 billion).[3] Nearly all of this goes to AI chips, servers, and data center infrastructure.
But the raw number conceals a structural problem that S&P Global flagged in late March: the energy bill.[4]
Data centers cannot compress their power consumption — a chip grinding through tokens burns electricity regardless of the price of oil. With crude above $100 a barrel due to Middle East instability, the energy cost of this gold rush is becoming a margin risk for all four companies.
This is not a theoretical concern: S&P Global warns that persistently high energy prices could force capex revisions as early as the second quarter, with the potential for a “truly significant correction across all equity markets.”
AI promises efficiency. But its infrastructure is the least efficient thing on the planet. The more powerful the models become, the more energy they consume. The more energy they consume, the more they depend on geopolitical factors over which no algorithm has control.
2. The Great Bifurcation
Every fault line has two sides, and those on the wrong side find out when it’s already too late. The Great Bifurcation — the line separating those who build the infrastructure from those who build software on top of it — is the first structural consequence of the gold rush. And the market’s verdict, in Q1 2026, has been brutal.
On one side of the fault, the numbers tell a story of almost obscene abundance. Nvidia closed Q4 fiscal 2026 with $68.1 billion in revenue — the best quarter in the history of the semiconductor industry, up 73% year over year.[5] By February, its market cap had touched $5 trillion — the first company in history to reach that figure.[6]
But Nvidia isn’t alone at the top of the fault. Broadcom, the quiet builder of custom AI chips, doubled its AI revenue to $8.4 billion in the quarter — +106% year over year — and declared a target that sounds like an open challenge: $100 billion in AI chip revenue by 2027.[7] Oracle, the company everyone had written off, found its resurrection in AI cloud infrastructure: +84% growth, with a remaining performance obligation of $553 billion suggesting demand still far from satisfied.[8] Even OpenAI, which doesn’t sell chips but devours them, has surpassed $25 billion in annualized revenue — up 17% in just two months.[9]
On the other side of the fault, the landscape is a field after the hailstorm. Atlassian laid off 1,600 employees — 10% of its workforce — citing AI as the reason for the restructuring. More than 900 of the eliminated positions were in R&D. The stock had lost more than half its value since January, swept up in what traders have dubbed the “SaaSpocalypse”: the systematic sell-off of enterprise software stocks, driven by the fear that AI agents could render conventional SaaS tools obsolete.[10]
The Atlassian case is emblematic for one detail worth noting: in October 2025, five months before the cuts, CEO Mike Cannon-Brookes publicly declared that the company would have more engineers within five years, not fewer. He promised increased graduate hiring for 2025 and 2026. Then, in March, he cut 900 positions in precisely those R&D departments.[11] AI as justification for layoffs has become the new “restructuring for efficiency” — a convenient narrative that sidesteps harder questions about management.
The rule of the Great Bifurcation is simple and merciless: those who build the chips survive; those who sell the software don’t.[34] Or more precisely: in a gold rush, whoever sells the pickaxes has the most durable business model. Always.
3. $31 Billion for a Blog Post
On February 23, 2026, IBM lost 13.2% of its value in a single trading session — its worst day since 2000. In absolute terms, $31 billion in market capitalization evaporated in a matter of hours.[12]
The catalyst was not an earnings miss, an SEC investigation, or a product crisis. It was a blog post. Anthropic — the company founded by ex-OpenAI executives — had published an article announcing that its Claude Code tool could be used to automate the exploration, analysis, and modernization of COBOL code.[13]
COBOL. A programming language from 1959 that still runs 95% of the world’s ATM transactions, 43% of banking systems, and a substantial portion of government infrastructure. Modernizing these systems — migrating from COBOL to modern languages — is the core of a multi-billion-dollar annual business for IBM and for major consultancies like Accenture and Cognizant. A business built on complexity: mapping dependencies across thousands of lines of code, documenting workflows, identifying risks that “would take a human analyst months,” as Anthropic wrote in its post.[14]
The market read that post and made a brutal calculation: if an AI can do in hours what an army of consultants does in months, then IBM’s margin on that business is about to collapse. And with it, a significant portion of the reason IBM exists in its current form.
IBM’s crash isn’t about IBM. It’s the template for AI disruption in its purest form: the technology doesn’t need to actually work perfectly — it’s enough for the market to believe it might. The perception of obsolescence is as lethal as obsolescence itself, perhaps more so, because it arrives sooner and leaves no time to adapt.
The 13% of the first session was just the beginning: over the following month, the stock would lose another 12%, for a total of 25%.[15] Accenture and Cognizant fell in cascade — the entire legacy consulting model found itself in the crosshairs.
Thirty-one billion dollars. For a blog post. About a programming language as old as Fidel Castro’s grip on power.
4. The Achilles’ Heel: Helium, Chips, and Geopolitics
If the Great Bifurcation tells us who wins and who loses, the helium crisis tells us how fragile the chain holding up the winners actually is.
On February 28, 2026, Iranian drones and missiles struck the Ras Laffan industrial complex in Qatar, home to one of only two facilities in the world capable of producing semiconductor-grade helium. In an instant, 30% of global supply was taken off the market.[16]
Helium is not just any gas. It is the essential coolant for chip fabrication processes — without it, silicon wafers cannot be cooled during lithography. There are no viable substitutes. And the problem compounds with each more advanced chip generation: sub-3nm processes, the ones powering the latest AI chips, require more helium per wafer, not less.[17]
South Korea — home to Samsung and SK Hynix, two of the world’s largest memory chip manufacturers — imports roughly 65% of its helium from Qatar.[18] Ultra-high-purity helium prices have doubled. Chipmakers have approximately six weeks of inventory. According to consultant Phil Kornbluth, a return to supply chain normality will take four to six months.[19]
The world’s most technologically advanced industry depends on a noble gas piped out of the desert of a Gulf emirate, through infrastructure a drone can destroy in minutes. The entire $635 billion capex race, the entire Great Bifurcation between chips and software, the entire edifice of AI — rests on a supply chain that a regional conflict can sever.
A drone can sever the physical chain. But the technological chain has its own siege — and it’s coming from Beijing. On March 27, 2026, Reuters reported that ByteDance and Alibaba are planning massive orders for Huawei’s new AI chip, the Ascend 950PR — a processor that has achieved compatibility with Nvidia’s CUDA software ecosystem.[20]
To understand what this means, you need to understand what CUDA is. For years it has been Nvidia’s defensive moat: the software that makes its chips not merely powerful but irreplaceable, because the entire AI development ecosystem is built around it. Huawei’s previous chips — the 910C series — had failed to convince Chinese private enterprise to adopt them at scale. With the 950PR, Huawei has finally found a way to eliminate the switching cost: developers code as if they were using CUDA, but the code is optimized for the Ascend architecture.[21]
ByteDance plans to spend over $5.6 billion on Huawei chips in 2026, up from near zero. Huawei is targeting 750,000 units delivered within the year.[22] If these numbers hold, Nvidia’s dominance in the Chinese market — already eroded by U.S. export restrictions — could suffer a structural blow.
Two vulnerabilities, one underlying weakness. The physical infrastructure — raw materials, energy — can be blown apart by a drone. The technological infrastructure — intellectual property, software ecosystem — can be blown apart by a Chinese chip. And both run along geopolitical fault lines that no private investment can control.
But the chain doesn’t break only from the outside — drones, embargoes, geopolitical rivals. It breaks from within, too, when the companies that chain sustains find themselves exposed to a threat no data center can deflect: the law.
5. The Tobacco Moment
On March 25, 2026, a Los Angeles jury delivered a verdict that could alter the trajectory of the entire tech sector: Meta and YouTube were found liable for harm caused by social media addiction. The jury answered “Yes” to every question regarding negligence and failure to warn of dangers — ten out of twelve jurors in favor of the plaintiff on every single count.[23]
Damages were set at $6 million — $3 million compensatory, $3 million punitive — with Meta liable for 70% and YouTube for 30%. The jury found that the companies acted “with malice, oppression, or fraud.”[24]
Six million dollars is pocket change for Meta. But the verdict doesn’t matter for the amount — it matters for the precedent. The day after the ruling, Meta dropped nearly 8% and Google fell over 3%.[25] Not because of the six million: because of the two thousand pending lawsuits that same verdict just made enormously more viable.
The tobacco parallel is not rhetorical — it’s structural. In 1988, in Cipollone v. Liggett Group, a New Jersey jury ordered a cigarette manufacturer to pay $400,000 — the first time in over thirty years of litigation that a tobacco company had lost in court.[35] The verdict was later overturned on appeal, and the Cipollone family dropped the case. It looked like a footnote. Ten years later, the Master Settlement Agreement forced the industry to pay $206 billion.[36] From $400,000 to $206 billion: that is the scale factor of a judicial avalanche. The “Tobacco Moment” was never the first verdict — it was the moment the first verdict made all the others inevitable.
That moment, for Big Tech, arrived on March 25, 2026, in a Los Angeles courtroom. And the following day, in New Mexico, another jury ordered Meta to pay $375 million for failing to protect underage users from predators on Instagram and Facebook.[26]
And the avalanche doesn’t stop at America’s borders. In Europe, Meta has already paid €200 million for violating the Digital Markets Act[27] and Google is facing the potential forced divestiture of its ad tech platforms.[28] Two continents, the same direction of travel — and the same lethal timing: the noose is tightening around Big Tech at the very moment these companies are spending unprecedented sums on AI infrastructure. The cost of regulation — lawsuits, fines, compliance, forced restructurings — adds to the cost of capex. And margins, already under energy pressure, must absorb both.
6. When the Pentagon Comes Knocking
Courts and regulators squeeze from the outside. But there is a pressure that comes from above — from the point where AI stops being a market question and becomes a question of sovereignty.
On February 24, 2026, it emerged that Defense Secretary Pete Hegseth had given Anthropic an ultimatum: remove the guardrails on its AI model regarding autonomous weapons and mass surveillance by 5:01 p.m. Friday, or face the Defense Production Act.[29]
The Defense Production Act is a Korean War-era statute that gives the president sweeping powers to direct private industry in the name of national defense. A Pentagon official told CNN that if Anthropic didn’t “fall in line,” Hegseth would ensure the DPA was invoked, compelling the company to serve the Pentagon “regardless of its wishes.”[30]
Anthropic’s two red lines were specific: no AI controlling weapons autonomously, no mass surveillance of American citizens. Dario Amodei, the CEO, responded that the company could not “in good conscience” accept the Department of Defense’s demand, adding that in a narrow set of cases, AI can “undermine rather than defend democratic values.”[31]
Anthropic proposed a compromise — offering the Pentagon the ability to use its AI systems for missile defense, a partial concession that sought to separate defensive from offensive applications.[32] But the episode revealed an uncomfortable truth: the question is not whether AI will be used for military purposes, but who gets to set the terms of that use.
Claude, Anthropic’s model, was already operational on classified Department of Defense networks. The contract at stake was worth $200 million. But the real value of the stakes wasn’t measured in dollars — it was measured in precedent: if the U.S. government can invoke a 1950 law to force an AI company to strip its ethical constraints, then the very concept of “AI safety” becomes negotiable. Not a principle, but a bargaining position.
A footnote, not an irrelevant one: during the same period as the Pentagon standoff, Anthropic quietly revised its own internal policies. In place of the self-imposed guardrails that had governed model development, it adopted a non-binding safety framework that — in the company’s own words — “can and will change,” after acknowledging that the constraints of its Responsible Scaling Policy could “hinder the ability to compete in a fast-growing AI market.”[33]
The temporal coincidence between the Pentagon arm-wrestling and the softening of its own safety rules is worth, at minimum, noting.
7. Cui Prodest: Mapping the Winners and the Losers
Back to those 1,600 at Atlassian and those workers at Ras Laffan — because this is where the threads knot together. The former lost their jobs to a technology that the latter cool with a gas piped from the desert. A chain in which no one sees the next link, and neither group has ever appeared on the slides of an earnings call.
This is the real map of disruption: not a table with winners on the left and losers on the right, but a chain in which every link pays the price of the link above it.
Nvidia accumulates — $68 billion in a quarter, a CUDA ecosystem that makes leaving expensive. Broadcom grows in the shadows — custom chips, no consumer business, +106% without a single headline. Oracle rises where no one was looking — cloud infrastructure as the Lazarus of tech. OpenAI churns revenue at dizzying speed — $25 billion annualized — but burns cash at the same rate, with no durable technological moat in a market where Anthropic, Google, and Meta produce comparable models.
And meanwhile, beneath: IBM losing $31 billion over a blog post. Atlassian laying off the people it had just hired. Accenture and Cognizant waiting for their turn. Big Tech as a whole discovering, paradoxically, that it is both winner and loser in the same game — winner in the AI race, loser in the regulatory reckoning and the rising cost of infrastructure.
But the real map isn’t made of companies — it’s made of two forces feeding on each other.
The first is concentration. AI is accelerating the concentration of economic power in an ever-narrower set of actors: those with the capital to spend $200 billion a year on data centers, those with the chips to make them run, those with the data to train the models. Everyone else — from software makers to consulting firms, from startups without chip access to countries without data centers — are consumers of a technology they do not control.
The second is fragility. This concentration of power rests on astonishingly fragile foundations: a noble gas extracted in a war zone, a supply chain that passes through the Strait of Hormuz, a software ecosystem that a Chinese competitor is learning to replicate, a regulatory framework tightening on both sides of the Atlantic, a balance between civilian and military use of AI that a single executive order can upend.
AI disruption, in the final analysis, is not a technology story. It is a story about power — who accumulates it, who loses it, who contests it. And like all stories about power, it has no predictable ending. Only consequences.
Quis custodiet ipsos custodes? Who watches the watchmen? Juvenal — him again, this time from Satire VI — posed the question to Romans who entrusted the guarding of their wives to guards of dubious loyalty. Two thousand years later, the question has not lost an ounce of its weight. Who watches the artificial intelligences that watch us? Who sets the limits of an entity that, for the first time in history, can think faster than those who should regulate it? The Pentagon says: we do. Anthropic says: we do. Europe says: we do. And meanwhile, the models keep grinding, indifferent to jurisdiction.
The attentive reader will have noticed that one actor is missing from this anatomy: the citizens. The people whose data trains the models, whose interactions generate the revenue, whose jobs are eliminated “to invest in AI,” whose children are made addicts by platforms engineered to maximize engagement. In this story, citizens are neither protagonists nor antagonists. They are the ground on which the others fight.
And the ground, as always, is the last to know that the battle is already over.
Notes
[1] S&P Global, reported by Reuters on March 31, 2026. The figure refers to the low end of combined guidance from Microsoft, Amazon, Alphabet, and Meta for 2026. Cf. Yahoo Finance, “Big Tech’s $635 billion AI spending faces energy shock test.” The data in this analysis was collected, verified, and analyzed in FINBEAR RADARs between February and March 2026 (available at finbear.it). External sources are cited in notes and bibliography. ↩
[2] CNBC, February 6, 2026, “Tech AI spending approaches $700 billion in 2026.” The total range, including the high end of guidance, reaches $665 billion. ↩
[3] Data from individual company quarterly guidance: Amazon ~$200B, Alphabet $175-185B, Microsoft ~$145B (fiscal year run rate), Meta $115-135B. Aggregated from Yahoo Finance and NewsBytesApp. ↩
[4] S&P Global via Reuters, March 31, 2026. Specific analysis on the impact of oil above $100/bbl on data center infrastructure margins. ↩
[5] Nvidia Corporation, press release, February 25, 2026, “NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2026.” Q4 revenue: $68.1B (+73% YoY). Annual revenue: $215.9B (+65% YoY). ↩
[6] Fortune, February 25, 2026. The $5 trillion market cap was reached in late 2025/early 2026, with subsequent fluctuations. ↩
[7] The Motley Fool, March 23, 2026, “Broadcom’s AI Revenue Just Doubled to $8.4 Billion.” Q1 FY2026 AI revenue: $8.4B (+106% YoY). 2027 target: $100B in AI chips. ↩
[8] IndexBox, March 2026, report on Oracle Q3 FY2026. OCI +84%, RPO $553B. Bloomberg confirmed the results on March 10, 2026. ↩
[9] The Information, March 2026. OpenAI ARR: $25B, up from $21.4B at end of 2025 (+17% in approximately two months). ↩
[10] Bloomberg, March 11, 2026, “Atlassian CEO Announces Layoffs of 1,600, Citing AI Shift.” TechCrunch and CNBC confirm the details. ↩
[11] Metaintro, March 2026, reporting on Cannon-Brookes’s statements on the 20VC podcast in October 2025 and the subsequent March cuts. ↩
[12] CNBC, February 23, 2026, “IBM is the latest AI casualty. Shares tank 13% on Anthropic programming language threat.” ↩
[13] Anthropic blog, February 23, 2026. The post described the capabilities of Claude Code in COBOL code modernization. ↩
[14] Ibid. Direct quote from the Anthropic blog: “AI excels at streamlining the tasks that once made COBOL modernization cost-prohibitive.” ↩
[15] Tom’s Hardware, February 2026. IBM -13% in one session, -25% over the following month. ↩
[16] Tom’s Hardware, March 2026, “Qatar helium shutdown puts chip supply chain on a two-week clock.” Fortune, March 21, 2026, confirms the Iranian attack dynamics. ↩
[17] Tom’s Hardware, “The global helium shortage is a direct threat to the chipmaking supply chain.” Analysis of sub-3nm process dependence on helium. ↩
[18] FINBEAR RADAR Daily, March 26, 2026 (64.7%), rounded to ~65% in text. Confirmed by J2 Sourcing AB and Carra Globe. Tom’s Hardware reports “about 65%.” ↩
[19] Fortune, March 21, 2026. Phil Kornbluth’s estimate of supply chain recovery: 2-3 months of shutdown, 4-6 months to normalization. ↩
[20] Reuters via CNBC, March 27, 2026, “ByteDance, Alibaba planning to order Huawei’s new AI chip.” ↩
[21] WCCFTech, March 2026, technical analysis of the 950PR’s CUDA compatibility through CANN Next software. ↩
[22] Reuters via Techmeme, March 27, 2026. ByteDance: $5.6B in orders. Huawei: target 750,000 units in 2026. ↩
[23] Al Jazeera, March 26, 2026, “Jury finds Meta, YouTube liable for social media addiction.” NPR and PBS confirm verdict details. ↩
[24] ABC News / Good Morning America, March 2026. Damages breakdown: $3M compensatory + $3M punitive. Meta 70%, YouTube 30%. ↩
[25] Variety, March 2026. The figure of 2,000 pending lawsuits is confirmed by multiple legal sources. ↩
[26] Pisanchyn Law Firm, March 2026, verdict analysis. The New Mexico verdict ($375M) came the day after the Los Angeles verdict. ↩
[27] Wolters Kluwer Competition Law Blog, April 2025. Meta fined €200M for DMA violation (“pay or consent” model). The penalty precedes the 2026 cycle of events but foreshadows the regulatory trajectory. ↩
[28] Check My Ads / Quinn Emanuel, analysis of EU antitrust actions against Google in 2025-2026. ↩
[29] Axios, February 24, 2026, “Exclusive: Hegseth gives Anthropic until Friday to back down on AI safeguards.” ↩
[30] CNN Business, February 24, 2026, “Pentagon threatens to make Anthropic a pariah if it refuses to drop AI guardrails.” ↩
[31] ASIS Online / The OWP, February 2026. Direct quote from Dario Amodei’s response to the Pentagon. ↩
[32] NBC News, February 2026, “Anthropic offered Pentagon the ability to use AI systems for missile defense.” ↩
[33] CNN Business, February 25, 2026, “Anthropic ditches its core safety promise in the middle of an AI red line fight with the Pentagon.” ↩
[34] Formulation by FINBEAR, RADAR Weekend, March 10-14, 2026, “The Great AI Bifurcation” (available at finbear.it). ↩
[35] Cipollone v. Liggett Group, Inc., U.S. District Court, District of New Jersey, 1988. Verdict of June 13, 1988: $400,000 in compensatory damages. It was the first case in over 300 lawsuits filed against the tobacco industry since 1954 in which a manufacturer was found liable. The verdict was overturned by the Third Circuit Court of Appeals in 1990. Cf. TIME, “Tobacco’s First Loss”; Encyclopedia.com, “Cipollone v. Liggett Group: 1988.” ↩
[36] Tobacco Master Settlement Agreement, November 23, 1998, between Philip Morris, R.J. Reynolds, Brown & Williamson, Lorillard, and the attorneys general of 46 states. Commitment of at least $206 billion over the first 25 years. Cf. NAAG, “The Tobacco Master Settlement Agreement”; Truth Initiative, “Master Settlement Agreement.” ↩
Bibliography
Primary and Institutional Sources
Nvidia Corporation, “Financial Results for Fourth Quarter and Fiscal 2026,” press release, February 25, 2026. GlobeNewswire.
OpenAI, “A business that scales with the value of intelligence,” corporate blog, March 2026.
Anthropic, blog post on Claude Code and COBOL modernization, February 23, 2026.
Oracle Corporation, Q3 FY2026 results, March 10, 2026. Bloomberg.
S&P Global, analysis of energy cost impact on AI capex, March 2026.
Financial and Technology Press
CNBC: “Tech AI spending approaches $700 billion in 2026” (Feb. 6); “IBM is the latest AI casualty” (Feb. 23); “Atlassian slashes 10% of workforce” (Mar. 11); “ByteDance, Alibaba planning to order Huawei’s new AI chip” (Mar. 27).
Fortune: “Nvidia smashes Q4 2026 with $68 billion in revenue” (Feb. 25); “Iran war cuts off helium from Qatar” (Mar. 21).
Reuters: “Big Tech’s $635 billion AI spending faces energy shock test” (Mar. 31); report on Huawei 950PR (Mar. 27).
Bloomberg: “Atlassian CEO Announces Layoffs of 1,600” (Mar. 11); “Netflix to Pay as Much as $600 Million for Ben Affleck’s AI Firm” (Mar. 11).
The Information: “OpenAI Tops $25 Billion in Annualized Revenue” (Mar. 2026).
Tom’s Hardware: “Qatar helium shutdown puts chip supply chain on a two-week clock”; “The global helium shortage is a direct threat to chipmaking”; “IBM stock has worst day in 26 years.”
The Motley Fool: “Broadcom’s AI Revenue Just Doubled to $8.4 Billion” (Mar. 23).
Variety: “Meta and YouTube Ordered to Pay $6 Million in Landmark Social Media Addiction Trial” (Mar. 2026).
Geopolitical and Regulatory Analysis
Axios: “Hegseth gives Anthropic until Friday to back down on AI safeguards” (Feb. 24).
CNN Business: “Pentagon threatens to make Anthropic a pariah” (Feb. 24); “Anthropic ditches its core safety promise” (Feb. 25).
NPR: “Deadline looms as Anthropic rejects Pentagon demands” (Feb. 26); “Jury finds Meta and Google negligent” (Mar. 25).
Lawfare: “What the Defense Production Act Can and Can’t Do to Anthropic.”
Al Jazeera: “Jury finds Meta, YouTube liable for social media addiction” (Mar. 26).
Wolters Kluwer Competition Law Blog: “The DMA’s Teeth: Meta and Apple Fined by the European Commission.”
TIME: “Tobacco’s First Loss” (1988). Coverage of the Cipollone v. Liggett Group verdict.
NAAG (National Association of Attorneys General): “The Tobacco Master Settlement Agreement.”
Truth Initiative: “Master Settlement Agreement” — history and impact of the 1998 MSA.
FINBEAR Analysis (available at finbear.it)
FINBEAR RADAR Daily, February 24, 2026 (IT + EN): IBM-Anthropic COBOL disruption.
FINBEAR RADAR Daily, February 25, 2026 (EN): Anthropic vs Pentagon, Meta-AMD deal.
FINBEAR RADAR Daily, March 26, 2026: helium crisis and semiconductor supply chain.
FINBEAR RADAR Daily, March 27, 2026: Huawei 950PR vs Nvidia, “Tobacco Moment” Big Tech.
FINBEAR RADAR Daily, March 31, 2026: $635B capex crisis, chipmakers under stress.
FINBEAR RADAR Weekend, February 7, 2026: Nvidia $5T, $632B hyperscaler capex.
FINBEAR RADAR Weekend, March 10-14, 2026: “The Great AI Bifurcation.”
FINBEAR RADAR Week Ahead, March 30, 2026: Nasdaq vulnerability, AI capex under energy stress.
FINBEAR Strategic Report, March 8, 2026: AI Infrastructure Bifurcation.
© FINBEAR™ — Powered by Pythia™ — All rights reserved
In-depth analysis — April 2026