Welcome to The Corner. In this issue, we look at how Microsoft is exploiting its control over Bing search data to force adoption of its cloud services and AI systems. ![]() Microsoft Doubles Down on Bundling of Cloud, Gen AI, and Search Karina Montoya For many years, Microsoft presented itself to the public and policymakers as the friendly and open alternative to Google’s tightly integrated set of platform monopolies. For instance, Microsoft for years allowed smaller rivals in search to access its Bing index through application programming interfaces, or more specifically “Bing Search APIs.” No longer. In the wake of two landmark government antitrust victories over Google—in which courts found that both the corporation’s search and advertising technology platforms to be illegal monopolies—Microsoft has moved swiftly to concentrate its own control over search data, as well as over cloud services and AI systems. Following two years of raising prices for access to Bing Search APIs, Microsoft in May announced plans simply to shut down this line of business entirely in August. It will then replace the traditional service with an AI product for developers hosted on Azure, its cloud business. The move appears to open a new frontier of anticompetitive conduct, with a dominant player tying access to cloud, AI systems, and search into a single bundle. It also comes to show just how technology corporations can distort competition in ways that entrench their dominance. And how, amid the scramble to develop AI, how little they fear scrutiny from antitrust regulators. If left unchallenged, the move will likely have large effects over the long-term. Accessing high-quality search indexes is crucial for developing generative AI tools, which produce output from two sources: training data ingested by an AI model, and web content fetched in real time to update the AI model. The latter is a process that happens in a split second, known as retrieval augmented generation (RAG) or ‘grounding.’ Microsoft’s Copilot Search and OpenAI’s ChatGPT ground results on the Bing search index, while Google’s AI Mode and Overviews fetch real-time data from Google’s own search index. For a search query, such as “What’s the latest on the heat dome in the U.S.,” each AI assistant may generate similar summaries. But news sources accompanying such text will likely be different, mostly because of variations in how RAG systems fetch raw search data and present it, based on their developers’ instructions on how to update the AI model. During Google’s longstanding monopoly over search, Microsoft worked hard to make Bing Search APIs a viable resource for small search engines, and later for AI companies. But under the new regime, starting in August these smaller players will have to use ‘Grounding with Bing Search,’ Microsoft’s proprietary RAG system that works on Azure, to continue accessing search data from Bing to create AI-generated output. Prices will also go up to $35 per 1000 queries compared to a range of $10 to $25 for the same query volume using APIs. This tightens Microsoft’s grip over competitors on several fronts. Companies using ‘Grounding with Bing Search’ will have less flexibility to decide when and how to ground a user search query, since the new tool will make that decision. Also, these companies will have to develop their AI systems in Azure, in part or whole. The change will especially affect smaller companies. More established rivals such as DuckDuckGo have long-term licensing deals to access Bing’s raw search data, as reported by Wired. At least for now, other providers of similar APIs to access search indexes, such as Brave, You.com or Mojeek will remain open. But their indexes are far less robust than those of Microsoft or Google. In a way, this follows a broad, longstanding pattern of tying and self-preferencing by the three technology corporations that dominate cloud services—Amazon, Microsoft, and Google — as Open Markets’ recent cloud report detailed. For example, even when Microsoft is no longer OpenAI’s exclusive cloud provider, it has right of first refusal when OpenAI seeks additional cloud capacity. In Amazon’s partnership with Anthropic, the latter has to use AWS as its primary cloud provider and use Amazon’s AI chips for its AI models. The remedies for the Google Search case, which include opening Google’s search index through APIs at minimum costs, may offer these smaller search rivals some relief. But Microsoft is likely to exploit this as well to further its own power, by using Google’s index to improve its own AI products while keeping its own index largely private. A more radical approach may be needed. Subbu Vincent, director of Media Ethics at the Markkula Center for Applied Ethics at Santa Clara University, in reaction to Microsoft’s announcement, proposed one potential fix: “Shouldn’t there be a way for high-quality search indexes to be built as public digital infrastructure, since it’s everybody’s content being indexed?” 📝 WHAT WE'VE BEEN UP TO:
🔊 ANTI-MONOPOLY RISING:
We appreciate your readership. Please consider making a contribution to support the continued publication of this newsletter. 📈 VITAL STAT:$300 millionThe total compensation some new hires at Meta’s superintelligence lab would reportedly be paid over four years. Meta has made at least 10 of these offers to OpenAI staffers. (Wired) 📚 WHAT WE'RE READING:Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Task: This academic research paper from MIT Media Lab examines the neural and behavioral consequences of using large language models to help with writing essays. Using EEGs on participants to assess cognitive load during essay writing, researchers found that those who used only their brain to write essay exhibited the strongest, most distributed neural networks, while search engine users showed moderate engagement and AI users displayed the weakest connectivity. The paper underscores the cognitive and educational costs of reliance on LLMs. ![]() Order Legal Director Sandeep Vaheesan’s new book: Sandeep Vaheesan, the legal director at the Open Markets Institute, published his first book Democracy in Power: A History of Electrification in the United States on December 3, 2024. Vaheesan examines the history—and presents a possible future—of the people of the United States wresting control of the power sector from Wall Street, including through institutions like the Tennessee Valley Authority and rural electric cooperatives. 🔎 TIPS? COMMENTS? SUGGESTIONS? We would love to hear from you—just reply to this e-mail and drop us a line. Give us your feedback, alert us to competition policy news, or let us know your favorite story from this issue. |