Artificial Intelligence and Competitive Dynamics in Downstream Markets
OECD, Paris Dec 1, 2025
Julian Wright
National University of Singapore
AI as a general-purpose technology (GPT)
• As with past GPTs (computers, Internet, mobile …)
• raises productivity, lowers business costs, leads to new services and markets
• shakes up industries, as some firms work out how to deploy GPT better than others
• can reinforce incumbency advantages
• downstream dominance mostly arises via entrants in new markets (SAP, Booking.com, Uber)
• overall effect on competition in downstream markets is mixed
Lessons from past GPTs?
• With each GPT, impossible to predict effects on competition at the time:
• Internet in the late 1990s
• increased choice for consumers
• lower costs of doing business
• but the creation of new markets and the rise of new gatekeepers was largely unforeseen
• Amazon marketplace, Booking.com, Facebook, Spotify, Uber …
Impact on downstream market competition
• One view is that there are strong network effects in AI benefiting incumbents
• Data network effects in applications of AI to downstream markets?
• unique and proprietary data from customers
• improve products and get more data
• e.g. Google search and maps
• but in many applications, data is not really unique, feedback signal is limited, value is limited
• generic models trained on general data can often do surprisingly well
Impact on downstream market competition
• Instead expect other sources of incumbency advantage
• lock-in from increased personalized services with individual usage (chatbots)
• leveraging strong upstream position into new downstream markets (shopping)
• But as with other GPTs, also expect a lot of disruption in downstream markets
• Adobe Photoshop with Midjourney, DALL·E 3, Stability AI, Runway, Pika, Google Nano Banana
• Upwork, Fiverr with ChatGPT, Claude, Gemini, Runway, Jasper, ElevenLabs, Synthesia
• Google search with ChatGPT search, Perplexity
• Google browser with Open AI Atlas, Perplexity Comet
• Uber with autonomous vehicles Waymo and Tesla
• Virtual assistants Alexa, Siri, Google home, Cortana with ChatGPT, Grok, …
• Online marketplaces like Amazon and Booking with AI agent providers: ChatGPT, Perplexity
Booking.com and AI
• Has vast data on hotel search and bookings
• Could use data and AI to improve its dominant position
• better match with user preferences and requests
• Faces threats from AI agent providers
• can directly search and transact with hotels knowing user preferences
• can search across all channels (including the platforms and metasearch providers)
• As AI agent providers aggregate user preferences across many users
• become the new gatekeepers for hotels wanting to attract guests
• and if operate over many downstream markets, could be even more powerful
Competition policy implications
• Vigorous use of existing tools to prevent:
• foreclosing downstream competition via exclusive access to data or vertical leverage
• Consider data-portability regulation for individuals
• so that users can move their data across providers to prevent lock-in
• AI algorithms can become a black-box for anticompetitive effects
• price-fixing algorithms, self-preferencing, price-parity clauses, price discrimination, predation
• can we make firms ensure their use of AI is not having these effects (compliance programs)?