Paris has decided that 2025 is the year to crack down on “autopréférence,” with the Autorité de la Concurrence opening a public consultation in June under France’s new law “to secure and regulate the digital space.” The inquiry asks interested parties to identify cases where a cloud-computing service provider treats its own software better than its rivals, and to suggest fresh legal tools to stop it. A final report is scheduled to be delivered to Parliament in November.
As usual, before lawmakers jump to conclusions, they should pause on the economics. One of us had a 2020 Concurrences piece that laid out the basic point: the so-called “vertical-discrimination presumption” has little empirical or theoretical backing. Platforms that integrate downstream are just as likely to expand markets as to foreclose them. Self-preferencing is ultimately just a mild form of vertical integration. In digital markets, it often enables platforms to optimize how their own services work together, creating operational efficiencies that benefit from coordinated design and shared infrastructure.
In one much-discussed example, Facebook’s integration of Instagram boosted demand for the entire photography-app category, helping independents as much as it helped Facebook itself. Google’s own entry into Android photo apps with Google Photos expanded total innovation, rather than crowding it out. Blockbuster first-party video games enlarge the console user base, lifting all developers. Even Amazon’s forays into third-party product lines have produced ambiguous—not clearly negative—effects on rival sellers. The evidence reveals no consistent pattern of harm. Vertical integration can create demand spillovers, signal quality, and fund platform investment.
Since that 2020 piece was written, we have even more related evidence.
Take Amazon. In a 2023 paper, Chiara Farronato, Andrey Fradkin, and Alexander MacKay tracked 184 heavy Amazon shoppers through a custom browser plug-in. Their findings showed that Amazon-brand items vault roughly four search-rank slots closer to the top—about half the bump that a paid “sponsored” listing buys—even after controlling for price, reviews, Prime eligibility, delivery speed, and keyword match.
That may seem unfair. If you really could compare goods that are identical (suppose identical packages of toilet paper), it would seem arbitrary to put the Amazon product higher up, and would do nothing to help consumers. But here, we don’t actually know the items are identical. We need to control for measurable product attributes—price, reviews, delivery speed, and feature sets. The authors stress that this mix of merit-based and “house” advantage doesn’t automatically translate into consumer harm; it merely flags the tradeoff that policymakers must measure, not assume.
Thus, the authors did a follow-up study. This time, they hid Amazon’s own brands to see what happens and how people respond when those labels vanish. They then estimated a full model and ran a simulation of what would happen if Amazon products disappeared. In the model, when those labels disappeared, prices on the remaining items ticked up a hair (about 0.6%) while overall consumer surplus fell 3.8%, with the heaviest losses (more than 10%) in categories like batteries and pain relievers.
Search effort didn’t budge; shoppers simply swapped to near-identical products that carried fewer reviews. In a few low-stakes categories (socks, toilet paper), welfare did actually rise when the Amazon labels vanished. In other words, the private-label effect is heterogeneous but net-positive more often than not. This is hardly the smoking gun many presume.
Even enforcement actions concede efficiencies from self-preferencing. The Federal Trade Commission’s (FTC) 2024 amended complaint against Amazon attacks how the platform puts sellers (often itself) into the Buy Box, but also concedes that the nationwide Fulfilment-by-Amazon network “can ship products faster and cheaper… benefits that are shared with shoppers via faster deliveries and cheaper products,” and that U.S. fulfilment facilities are needed “to timely and reliably serve shoppers.”
Internal emails about Seller-Fulfilled Prime likewise fret that shutting the program would strip Prime customers of “faster speeds” on more than 115 million items. That’s the operational tradeoff: Amazon’s logistics integration enables faster delivery speeds that benefit consumers. But restricting access to these capabilities would eliminate those speed advantages for millions of products, directly harming the customer experience the FTC claims to protect.
The FTC spins these efficiencies as merely benefits of scale that are denied to other retailers—therefore deeming it a harm. But in so doing, the agency also admits that these are benefits provided by the vertical integration of shopping with logistics. These scale benefits are inherently tied to vertical integration because they require coordinated investment in logistics infrastructure, standardized processes, and unified quality-control systems that independent third-party logistics providers could not easily replicate at comparable scale and efficiency. Courts ultimately must weigh those scale economies against any exclusionary harms before finding antitrust liability.
None of this means that self-preferencing is always and everywhere benign. Rather, it means that you cannot identify harm by the self-preferencing label alone. Any French rule that starts with a presumption of illegality will either ban pro-competitive integration outright, or end up riddled with exceptions that swallow the rule.
The Autorité’s questionnaire itself gives the game away. It asks stakeholders only two things:
- Have you observed practices of self-preferencing; and
- What procedural or legislative improvements would you suggest to fight such self-preferencing?
Notice what is missing. Nowhere does the consultation invite evidence that a cloud provider favoring its own software lowered prices, sped up processing, or pushed rivals to innovate. Instead, it asks only how to “fight” self-preferencing, skipping entirely the question of whether it is something that must always be “fought” in the first place. By shining its flashlight solely where complaints are likely to lurk, the Autorité risks the classic drunk-under-the-lamp-post mistake: you see only the grievances that happen to be illuminated and miss the benefits that lie just beyond the cone of light.
Cloud services—the consultation’s focus—do not magically support a default assumption that self-preferencing causes consumer harm. In fact, the market’s technical realities mean that enterprise customers routinely multi-home across Amazon Web Services, Microsoft’s Azure, and Google Cloud. This creates competitive pressure; marginal costs are near zero, while fixed costs are lumpy, encouraging efficient bundling. Latency drops when software is co-located with the data, making integration a performance necessity, rather than a mere preference. Forbidding a provider from bundling its own database with its compute nodes might sound like “neutrality,” but it could just as easily raise prices, slow workloads, and shrink the very market upon which French firms rely.
So here’s our law & economics plea: keep the standard squarely on consumer welfare. Ask first whether the practice raises quality-adjusted prices, or blocks innovation that would otherwise reach users. Demand data: prices before and after, product availability, entry patterns, etc. If the numbers point to foreclosure, antitrust already has tools to address it. If they point to more software, faster downloads, and bigger markets, celebrate the competition and move on.
