Big Tech's M&A Didn't Shrink. It Went Underground.

Based on the research ofYangyang Cheng, D. Daniel Sokol, and Carmelo Cennamo, "Antitrust Enforcement and the Redirection of Platform M&A," Platform Strategy Symposium, 2026

That is the finding at the center of a new working paper by Yangyang Cheng, D. Daniel Sokol, and Carmelo Cennamo, presented at the 2026 Platform Strategy Symposium. Using 6,901 software-sector M&A deals between 2000 and 2022, the authors track what happened after antitrust scrutiny of large technology platforms intensified around 2017. The headline result is not that platforms bought less. It's that they bought differently, and the parts of the deal that regulators can actually see shrank while the parts they can't stayed remarkably stable.

The Regulatory Acquisition Premium

Corporate boundary theory and dynamic capabilities theory have long explained why firms acquire external capabilities instead of building them in-house, but both traditions treat the regulatory backdrop as scenery rather than a variable. Cheng, Sokol, and Cennamo put the backdrop in the foreground. Their core idea is what they call a regulatory acquisition premium: an enforcement-induced cost that depends on the functional relationship between what the acquirer already does and what the target does. A substitutive acquisition, where the target replicates operations the acquirer already runs, looks like a straightforward horizontal competition problem and draws the sharpest scrutiny. A complementary acquisition, where the target fills a gap in the acquirer's portfolio rather than removing a competitor, draws less.

Once enforcement tightens, that gap between the two premiums widens, and the paper's expectation is mechanical: platforms should shift their buying toward the cheaper category. The data bear this out. The probability that a given deal is substitutive falls after the 2017 enforcement inflection, with the pre-period baseline sitting around 42 percent of deals and the post-period decline running roughly 10 to 15 percentage points, concentrated overwhelmingly among young targets, the startups whose competitive trajectory is hardest for a regulator to rule out and therefore the ones an incumbent has the strongest incentive to acquire before they mature into a rival. A deal-type breakdown sharpens the picture further: acquisitions of ecosystem targets fall while horizontal acquisitions within the same product market rise by a nearly offsetting amount, and vertical deals barely move at all. The authors verify this isn't just quiet exits from the M&A market broadly, an event-study check shows flat pre-trends through 2016, a sharp break starting in 2017, and a deepening effect through 2022, with acquirer fixed effects confirming the pattern shows up within the same company over time rather than reflecting a change in which companies happen to be doing the acquiring.

The part of the finding that should worry anyone hoping enforcement had actually shrunk platform power sits in the deal-substance measure. Using a standard text-similarity metric of business descriptions to proxy for real economic overlap between acquirer and target, the authors find that overlap does not decline even as the substitutive label does. Platforms are substituting toward targets that are less regulatorily salient but functionally similar to the ones they used to buy outright. And the effect is concentrated exactly where you'd expect: only Google, Amazon, Facebook, Apple, and Microsoft, the five firms that drew the most enforcement attention in this period, show the simultaneous decline in both substitutive share and textual similarity. Serial acquirers operating under less scrutiny, like Cisco, Intel, and Oracle, show no comparable shift. Between 2001 and 2022, those same five platform giants had acquired more than 800 software startups; the paper's contribution is showing that enforcement changed the shape of that buying spree without visibly slowing it down.

Where the Deals Went

The years just past the paper's 2022 data window read like a field test of its own hypothesis. Adobe's $20 billion agreement to buy Figma is the cleanest recent example of a textbook substitutive deal running straight into the regulatory acquisition premium. The UK's Competition and Markets Authority provisionally found that, absent the deal, Adobe would have kept competing directly with Figma in product design software, and the European Commission's Margrethe Vestager said the acquisition would have "terminated all current and prevented all future competition" between the two firms. Facing no workable path to clearance, Adobe and Figma abandoned the merger in December 2023 and Adobe paid a $1 billion termination fee. Weeks later, Amazon's $1.4 billion bid for Roomba maker iRobot collapsed the same way: the European Commission signaled it intended to block the deal over fears Amazon could foreclose rival vacuum makers on its own marketplace, and Amazon and iRobot walked away rather than fight a losing battle.

Both are exactly the kind of deal the paper's substitutive category is built to describe, an incumbent buying a company that does something close to what it already does or controls, in a market that already features prominently in a platform's core business. What's more interesting is what these firms did next. In March 2024, rather than acquiring Inflection AI outright, Microsoft paid roughly $650 million to license Inflection's models and hired nearly the company's entire 70-person staff, including its founders, in a structure widely reported to avoid triggering the formal merger review that a stock-and-cash acquisition would require. The FTC opened a probe into whether the arrangement was designed to dodge Hart-Scott-Rodino notification. Around the same time, the agency issued formal information demands to Alphabet, Amazon, Anthropic, Microsoft, and OpenAI to examine whether the wave of multibillion-dollar AI investments and cloud partnerships those companies had struck, deals structured as minority stakes and commercial agreements rather than acquisitions, were reshaping competition in ways ordinary merger review would never see. None of this shows up as a blocked deal in any antitrust docket. All of it looks, functionally, like the same capability-sourcing decision a straightforward acquisition would have made.

The receipts changed shape. The shopping didn't stop.

Same Overlap, a Different Paper Trail

What makes the Cheng, Sokol, and Cennamo finding more than a collection of anecdotes is that it locates the mechanism precisely: enforcement operates on regulatory salience, not on economic substance. A deal that would have looked like eliminating a competitor in 2016 gets restructured in 2024 as a talent-and-license agreement, a minority investment, or an acquisition of a company one product category removed from the acquirer's own, different enough to clear review, similar enough to deliver nearly the same capability. The paper's own framing suggests this isn't a platform-specific quirk. Its authors note the logic should generalize to any setting where regulation prices governance modes differently: pharmaceutical firms redirected toward licensing when horizontal consolidation draws scrutiny, banks pushed toward fintech partnerships when banking mergers face resistance, energy firms steered toward joint ventures when vertical integration is penalized. Antitrust enforcement, in this reading, is less a wall than a set of relative prices, and firms with capital and legal sophistication reliably find the cheapest way to buy what they need.

The Uncomfortable Bottom Line

Read narrowly, the 2017 enforcement wave worked: it measurably shrank the share of deals that most resemble a platform simply removing a competitor, and it did so hardest among the youngest, most uncertain targets, which is exactly where a "kill the future rival" story is most plausible. Read at the level the paper's overlap measure captures, the same enforcement wave accomplished far less. The economic footprint of platform acquisitions, how much the businesses being bought actually resemble the businesses doing the buying, held steady through 2022, and the events of 2023 and 2024 suggest platforms have only gotten more fluent since. Adobe and Amazon show what happens when a deal is transparent enough to get flagged. Microsoft's Inflection deal and the FTC's subsequent AI-partnership inquiry show what a five-company industry does next: it doesn't stop buying capability, it buys the same capability through a door regulators haven't finished building a rule for. Enforcement redirected the paper trail. It has not yet redirected the outcome the paper trail was supposed to prevent.