Should Big Tech be broken up?
A federal judge says Google is a monopolist. Mark Zuckerberg sat through weeks of trial in 2025 over whether Meta has to spin out Instagram. The EU just fined Apple half a billion euros. The dispute isn’t whether these companies are big — it’s whether “big” is the problem.
Voir comme graphe de débat“Google is a monopolist”
“After having carefully considered and weighed the witness testimony and evidence, the court reaches the following conclusion: Google is a monopolist, and it has acted as one to maintain its monopoly.”
— Judge Amit Mehta, US District Court for D.C., United States v. Google, opinion, August 5, 2024
Twenty-six years after the Microsoft case, the US government has won a monopolization ruling against a dominant platform. Two sentences from a federal opinion, and the line on whether antitrust law could still bite a modern tech firm just moved.
The bar Mehta applied is the same bar that controlled the Microsoft case: monopoly power in a properly defined market, plus exclusionary conduct used to maintain it. The numbers in the opinion are the giveaway — Google held roughly 90 percent of general search queries on desktop and 95 percent on mobile, and paid Apple and the handset makers billions a year to keep itself the default. The conduct piece is what made this winnable. Market share alone doesn’t lose you a Sherman Act case; the exclusive-dealing arrangements did.
This is the standard industrial-organization toolkit: define the relevant market, measure power inside it, identify conduct that forecloses rivals from achieving the scale needed to compete. Economics Ch.6 (Market Structures and Game Theory) covers the modeling apparatus — cross-side network effects, two-sided platforms, contestable-market theory. The doctrine the court applied isn’t obscure; what’s new is that a court used it against a firm whose product is free at the point of use.
Formally, the test runs through three filters: (1) market definition — the SSNIP test asks whether a hypothetical monopolist could raise price profitably by 5–10%; for a zero-price product, courts substitute quality- or attention-based variants. (2) Market power — Lerner index, share thresholds, durable barriers to entry. (3) Exclusionary conduct — foreclosure of rivals from minimum efficient scale, evaluated under the rule of reason. Google failed at filter (3): the default-search payments to Apple and Android OEMs were ruled exclusionary because they denied rivals the query volume needed to train competitive search algorithms.
Search engines get better the more searches they see. Google paid the only two companies that distribute phones in the United States to make sure rival engines couldn’t accumulate the search volume needed to catch up. That payment is what tipped “you won fair” into “you bought the moat.”
For most of the past three decades the standard view among economists has been that the antitrust machinery doesn’t really apply to platforms like Google, because the product is free and quality keeps improving. The 2024 opinion is the first major rebuttal of that view inside US doctrine. It says: the standard machinery applies, you just have to be willing to read “harm to competition” as something other than a checkout-price increase. That reading isn’t radical — it’s straight out of the dynamic-competition literature in Ch.11 (Advanced Microeconomics). It just hadn’t been deployed.
Antitrust isn’t over. The Sherman Act, as written in 1890, just convicted a search engine in 2024 using merger-review and exclusion doctrines that were already on the books — the same family of tools that broke Standard Oil into thirty-four successor companies in 1911 and split AT&T into the seven regional Bells under a 1984 consent decree. Whether the remedy that follows is good policy is a different question — one the case itself doesn’t answer.
But the case used the mainstream IO apparatus to convict. There’s a parallel argument that says the apparatus itself is the problem — that the standard the courts have been applying for forty years systematically can’t see most of what makes Big Tech actually powerful. That argument has a name, and a journal article, and a former chair of the FTC.
Market power, two ways
“The whole task of antitrust can be summed up as the effort to improve allocative efficiency without impairing productive efficiency so greatly as to produce either no gain or a net loss in consumer welfare.”
— Robert H. Bork, The Antitrust Paradox: A Policy at War with Itself, 1978
One sentence from a 1978 book is the closest thing American antitrust law has to a constitution. The Supreme Court adopted it in the Reiter line the following year. Forty-five years later it’s still the standard every merger gets evaluated against. It tells you what to measure.
Bork’s move was deceptively narrow. The Sherman Act of 1890 was written in the language of trusts, restraints of trade, monopolization: capacious terms that earlier courts had read to cover political and social harms, not just price effects. Bork argued the statute’s only coherent goal was economic efficiency, and that efficiency reduced cleanly to one measurable thing: did consumers pay more, or less? Mergers that lowered prices were good; mergers that raised them were bad; everything else was outside antitrust’s scope. Ch.11 (Advanced Microeconomics) works through how this gets operationalized — the merger-simulation models, the unilateral-effects tests, the upward pricing pressure formula — and they all share one feature: they measure the price effect on the consumer at checkout.
The intellectual provenance matters here. Bork was a creature of the Chicago Law School antitrust seminar — Aaron Director, Henry Manne, Richard Posner. The displacement of the prior “structure-conduct-performance” Harvard school was a self-conscious counter-revolution. History of Economic Thought Ch.10 (Counter-revolution) covers the broader Chicago shift; the antitrust piece traveled into doctrine through the Reagan-era courts. And the policy regime followed: Economic History Ch.16 (Stagflation and the Neoliberal Turn) places the consumer-welfare turn alongside the deregulation wave that hit transport, telecom, finance, and energy in the same window.
Here is the argument the Bork frame can’t see, made at full strength. When Amazon takes a 30 percent cut from third-party sellers, the consumer at checkout doesn’t pay more — she may pay less, because Amazon’s scale economies subsidize the price. The harm is borne upstream, by the merchant who has no realistic option to sell elsewhere, whose margins compress year after year, whose product gets cloned into a private-label competitor, whose listings get demoted when the algorithm changes. None of this shows up as a price increase. All of it shows up as foreclosure: small competitors can’t reach the scale needed to discipline the platform, and so the next generation of price competition never arrives.
This is Lina Khan’s 2017 argument in compressed form, and it is not a complaint about the math. It is a claim about what gets measured. The consumer-welfare standard isn’t a neutral analytical default; it is a 1978 modeling choice keyed to a 1950s manufacturing economy — one where producers and consumers were distinct populations, prices were the binding contract, and platform two-sidedness wasn’t a structural feature of anything. In an economy where the dominant firm sits between two sides of a market and extracts rents from the side without an exit, the price-to-consumer metric returns the wrong number not because the math is wrong but because it is pointing at the wrong outcome.
Khan and Wu didn’t arrive out of nowhere. The 2017 paper is the founding document of what now goes by “neo-Brandeisian” antitrust — a post-2008 institutional-economics revival that sits alongside Piketty on inequality, the Minsky revival on finance, MMT as monetary heterodoxy, and complexity economics as the broader pluralist wave. History of Economic Thought Ch.17 (Modern Pluralism) places this whole cluster as a deliberate departure from the post-1970 mainstream consensus: not a return to Marx but a return to the question of whether the framework the discipline inherited from the 1970s was the right one. The Khan-Wu position is one strand inside that turn.
“Amazon has positioned itself at the center of e-commerce and now serves as essential infrastructure for a host of other businesses that depend upon it… The current framework in antitrust — specifically its pegging competition to ‘consumer welfare,’ defined as short-term price effects — is unequipped to capture the architecture of market power in the modern economy.”
— Lina Khan, “Amazon’s Antitrust Paradox,” Yale Law Journal, 2017
Is “low prices” the right thing to measure?
The Bork standard treats checkout price as a sufficient statistic for competitive harm. That works in a one-sided market. Platforms aren’t one-sided markets. The right read of Khan isn’t “throw out price” — it’s “price is one of several outputs of a competitive process, and you can’t infer the process from the output you happen to measure.”
The honest read: the consumer-welfare standard is partial. It measures a specific slice of competitive harm well and other slices badly. Stage 1’s Google case showed the standard can still find platform-era violations if a court is willing to read “harm to competition” as something a step away from checkout price. But Khan is right that the metric was designed for a different economy, and that designing-around takes work the doctrine hasn’t fully done. The apparatus is partial, and partial apparatus is what we’ve got.
All of which would be a theoretical fight about modeling choices, except that in 2025 a contemporaneous documentary record entered evidence in federal court. The record was internal emails. And the emails were embarrassingly clear.
The killer-acquisition record
“If they were to grow on a large scale, they could be very disruptive for us.” … “There is a non-trivial chance that we will be forced to spin out Instagram and perhaps WhatsApp in the next 5–10 years anyway.”
— Mark Zuckerberg, internal emails (2012 and 2018), introduced as evidence in FTC v. Meta, April 2025
The first email was written six weeks before Facebook bought Instagram for $1 billion. The second was written six years later, by which point Instagram alone was worth roughly $100 billion. The motive for the acquisition is on the page, in the founder’s own handwriting.
The 2012 Instagram acquisition passed merger review without incident. The reason is mechanical and worth understanding: standard merger-review tools, as set up in Ch.11 (Advanced Microeconomics), evaluate the post-merger market against the current market. Instagram had thirteen employees and zero revenue. There was no current overlap with Facebook on any priced product. The Hart-Scott-Rodino review found nothing to flag, because the tools weren’t built to flag potential competition that the acquisition was specifically designed to eliminate.
The literature has a name for this: the killer acquisition. The buyer pays a premium not because the target is profitable today, but because it is the most likely future competitor — or because, in a network-effects market, it’s the only firm that could plausibly reach the scale needed to break the incumbent’s lock-in. The static merger-review test scores this as neutral. The dynamic-competition test scores it as foreclosure of the most important kind: not foreclosure of a current rival but pre-emption of the next one. Ch.6 (Market Structures and Game Theory) covers the network-effects modeling that makes this distinction load-bearing in platform markets specifically.
The dynamic vs. static contrast can be stated cleanly. Static merger analysis evaluates $\Delta CS$ at fixed market structure: the consumer-surplus change holding everything but the merger constant. Dynamic analysis evaluates the expected loss in future consumer surplus across counterfactual trajectories — including the trajectory where the acquired firm matures into the incumbent’s replacement. When the latter trajectory is the very thing being purchased, the static test returns zero harm precisely when the dynamic test returns the largest harm.
Imagine a chess game where the merger review only looks at the current board position. The 2012 Instagram acquisition was a move that took a piece worth thirteen employees and changed the next twenty moves of the game. The review tools said “you took one pawn,” because that’s what they were designed to count.
Tim Wu’s 2018 framing pushes one step further than Khan’s and is worth taking seriously on its own terms. The argument is not that bigness is inefficient. It’s that bigness is politically dangerous even when efficient:
“We have managed to recreate both the economics and politics of a century ago — the first Gilded Age — and remain in deep denial about it… The ‘curse of bigness’ can no longer remain confined to specialist lawyers and economists.”
— Tim Wu, The Curse of Bigness: Antitrust in the New Gilded Age, 2018
The Brandeisian tradition Wu is reviving treats antitrust as a tool of democratic self-government, not just consumer protection. The worry isn’t that Meta will overcharge you for Instagram — you don’t pay for Instagram. The worry is that a private firm whose product mediates the political speech of three billion people, whose acquisitions are not reviewable on the dimensions that matter, and whose CEO writes contemporaneous emails describing the suppression of rivals, is the kind of private power the Sherman Act was originally written to constrain. The efficiency argument and the political-power argument can both be true. Wu’s point is that the consumer-welfare frame retired the second one without a vote.
The tradition has a lineage. Louis Brandeis — Supreme Court justice from 1916, anti-trust theorist from a decade earlier — read the corporate enterprise as a legal-institutional object whose form was a political choice, not an economic inevitability. That reading became the spine of old institutionalism: Veblen on the corporation as conspicuous-production apparatus (Theory of the Leisure Class, 1899), John R. Commons on the legal substrate of capitalist exchange, Wesley Mitchell on the quantitative business-cycle work the NBER was built around. History of Economic Thought Ch.15 (Institutional tradition: Veblen to Acemoglu) traces this line forward through Coase’s transaction costs and North’s institutions-as-rules-of-the-game into the modern NIE / Acemoglu-Robinson program. Wu’s “curse of bigness” is the old-institutionalist reading, retrieved.
Put the two pieces together. There is now a documentary record — Zuckerberg in 2012, Zuckerberg in 2018, and what the dynamic-competition literature already predicted — and the record shows the standard merger-review test missed an acquisition that was, on the buyer’s own evidence, an act of foreclosure. The Chicago school’s claim that the price-to-consumer metric captures everything an antitrust law could reasonably want to know is not a defensible position after this evidentiary record. The metric is incomplete; the gap is non-trivial; reasonable people can disagree about the magnitude but the direction of the gap is no longer in dispute.
Diagnosing the problem is one thing. The remedy question is harder, and the place where the most thoughtful pro-Big-Tech argument lives. The other governments in the world are running a different experiment.
The remedy question
“The Commission found that Apple breaches its anti-steering obligation under the Digital Markets Act… app developers distributing apps via Apple’s App Store should be able, free of charge, to inform their customers of alternative offers outside the App Store.”
— European Commission press release announcing €500 million fine on Apple, April 23, 2025
Five hundred million euros, levied without a Sherman-Act-style trial, on the basis of an ex ante designation that Apple is a “gatekeeper” subject to structural obligations. This is what antitrust looks like when you don’t require proving consumer harm in court.
The EU’s Digital Markets Act is the philosophical opposite of the post-Bork US regime. Where the Sherman Act asks a court to find harm in a specific case after the fact, the DMA designates certain platforms as “gatekeepers” based on structural criteria — user counts, revenue, durable position — and imposes obligations on them by rule. No prosecutor has to prove the consumer is worse off. The presumption runs the other way: if you cross the threshold, certain behaviors are forbidden, and a fine follows non-compliance.
This is a regulatory-design question more than a market-failure question, and it lands inside Ch.18 (Institutional Economics): when do you want bright-line ex-ante rules and when do you want case-by-case ex-post adjudication? The trade-off is standard. Rules are predictable, cheap to enforce, and easy to game around the edges. Standards are flexible, expensive to litigate, and produce inconsistent outcomes across courts. The DMA is the rules choice. The Sherman Act, as the courts have read it since Bork, is the standards choice. Neither is obviously correct, and which one you prefer probably depends on whether you trust regulators or judges more.
“Trust regulators or judges more” sits on top of a whole tradition. The public-choice school spent half a century arguing that regulators are not neutral referees but self-interested actors subject to capture, rent-seeking, and the same incentive problems as the firms they police. History of Economic Thought Ch.14 (Public Choice) (Buchanan, Tullock, Olson) is the lineage backing the worry that a rules-based gatekeeper regime hands too much discretion to an agency that can be lobbied. It’s the strongest theoretical case against the DMA approach, and it cuts in exactly the opposite direction from the neo-Brandeisian argument that ran through the measurement and killer-acquisition stages.
Here is the strongest case against breaking up the dominant US tech firms, made in the voice of the people who actually make it. Marc Andreessen, who co-founded Netscape and now runs the largest venture firm in Silicon Valley, has been blunt about it:
“Innovation doesn’t come from the big company. It never has and never will.”
— Marc Andreessen, “The Little Tech Agenda,” a16z, 2024
Read this argument carefully, because it is more sophisticated than “leave Big Tech alone.” Andreessen agrees that incumbents are stagnant. He agrees that the Instagram and WhatsApp acquisitions were defensive. What he objects to is the remedy: an aggressive antitrust posture that chills the acquisition market is a posture that breaks the Silicon Valley financing model, in which most startup exits happen through acquisition by an incumbent. A founder weighing whether to start a company in 2026 is weighing a probability-weighted set of exits. Remove acquisition from that set — or load it with regulatory uncertainty — and the founder doesn’t start the company. Innovation suffers not because the incumbent suppresses it inside the firm, but because the new firm never gets capitalized.
This is a serious argument and it deserves serious treatment. The counter-argument has the burden: it has to show that the chill on acquisitions is either small relative to the foreclosure harm, or that the chill is itself a feature rather than a bug (because a world where Instagram-circa-2012 has to live or die on its own merits is, on balance, a world with more competition, not less). The empirical record is honestly mixed. The studies on European-style antitrust regimes show modest reductions in VC investment levels but also higher rates of new platform entry. Neither finding is dispositive at the magnitudes that matter for this dispute.
“The current framework in antitrust… is unequipped to capture the architecture of market power in the modern economy.”
— Lina Khan, 2017
Diagnosis isn’t prescription
Khan is right that the measurement gap is real. It doesn’t follow that breakup is the right remedy. The remedy question depends on things the diagnostic argument doesn’t address: enforcement cost, error rate, dynamic-investment chill, and the geopolitics of AI competition with firms that face no comparable constraints.
So here is where this lands. Khan is correct about the diagnosis: the consumer-welfare standard is a partial metric, the killer-acquisition pattern is a real foreclosure mechanism, and the post-Bork doctrine systematically underweights forms of harm that aren’t priced at checkout. The Google verdict and the Meta evidentiary record together establish that as a matter of documented fact, not theoretical inference.
The remedy question is harder than the diagnosis and deserves harder reasoning. The structural-breakup path has US precedent and it has run far enough to be evaluated. Standard Oil in 1911 was broken into thirty-four successor companies under the Sherman Act and the descendants (Exxon, Mobil, Chevron, others) became more valuable than the original combined: the canonical structural-breakup case. AT&T in 1984 was split into seven regional operating companies under a consent decree, and the divestiture is generally read as having opened the long-distance and equipment markets to the competition that produced the 1990s telecom build-out. United States v. Microsoft, decided 2000 and settled 2001, pursued a structural breakup of the operating-system business from the applications business at the district-court level, then settled at appeal for conduct remedies (interoperability disclosures, OEM restrictions) once the political will for structural relief had eroded. Three precedents, three different shapes, three different verdicts in retrospect. Structural breakups — spinning Instagram and WhatsApp back out of Meta, separating Chrome from Google search, forcing Apple to allow alternative app stores — have the virtue of being clean: they restore the counterfactual the killer acquisitions extinguished. They have the cost of being slow, contested, and possibly retrograde in markets where network effects make a single dominant platform the equilibrium regardless of who owns it. Conduct remedies — the kind the EU is pursuing under the DMA, the kind the Mehta court appears to be weighing in US v. Google — trade cleanness for tractability: they don’t break the firm but they constrain the behaviors that turn dominance into permanent dominance. Andreessen’s chill-on-acquisitions worry applies most sharply to structural remedies and much less sharply to conduct remedies.
The right answer for the US in 2026, given that Khan is right about diagnosis and Andreessen is right about innovation chill: aggressive conduct remedies on the established dominant firms (Google, Apple, Meta), preserved as ongoing supervised obligations rather than one-time breakups; serious revival of merger review to block the next round of killer acquisitions before they happen; and explicit institutional acknowledgment that the consumer-welfare standard alone is insufficient to evaluate platform-era cases. The Sherman Act has the tools. What it needed was permission to use them on something that doesn’t look like a 1950s manufacturer. The 2024 Mehta opinion granted that permission. The work of figuring out what to do with it is now the active question.
The pattern here — a metric optimized for a different economy, returning increasingly misleading numbers as the actual economy drifts away from the metric’s assumptions — is not unique to antitrust. It shows up in how the official housing statistics measure supply (Walkthrough 12), in how healthcare cost-per-patient indices measure efficiency (Walkthrough 13), and in how labor-market tightness gets read off of unemployment numbers that no longer mean what they used to. Measurement choices, made decades ago for specific reasons, persist as defaults long after the reasons have changed. Antitrust is one instance of a wider phenomenon.