Amazon's Ad Business Soars + AI APIs — With Benedict Evans

Episode Summary

Executive Summary: Benedict Evans argues that big tech’s recent “drawback” is mostly a normalization after pandemic-era distortions, not a collapse. His core thesis is that retail, advertising, media, and even software are increasingly merging into a few “new gatekeeper” platforms, while generative AI is powerful but still too uncertain to replace specialized, vertical products or general search.

Main Topics: Big tech’s post-pandemic normalization (Priority: 5/5): Evans says the last few years were a spike of overenthusiasm and valuation distortion, not a permanent break in tech’s long-term growth. He argues companies like Apple, Meta, Alphabet, and Amazon are reverting toward their pre-pandemic trajectories rather than entering structural decline. Retail media and the blurring of commerce and advertising (Priority: 5/5): A central theme is that retail platforms now function as media platforms. Amazon, Walmart, Target, Uber, and others are turning website inventory, first-party data, and search intent into ad products, collapsing distinctions between advertising, marketing, rent, shipping, and pricing. Amazon as the archetypal new gatekeeper (Priority: 5/5): Evans uses Amazon to show how marketplace design, third-party sellers, and advertising combine into a scalable system that captures more of the commerce stack. He frames Amazon’s marketplace as a wholesale-access model that lets the company scale without owning every inventory relationship. Marketplace unbundling and the rise of direct-to-consumer (Priority: 4/5): The discussion highlights how Shopify, Shein, Temu, and Chinese manufacturers are reshaping distribution by going direct to consumers or through platforms. This creates a fragmented, infinite-product environment where discovery and logistics filters that once constrained retail are disappearing. Big Tech growth slows, but the business model remains intact (Priority: 4/5): Evans says slower growth at Google, Meta, and Amazon is driven by macro conditions and staffing corrections more than a broken model. He emphasizes that the underlying shift to digital commerce, mobile, and online advertising is still in place. Generative AI: transformative, but product uncertainty remains (Priority: 5/5): He is enthusiastic about LLMs but skeptical that chat is the right interface for everything. He expects the strongest use cases to be specialized vertical applications and APIs rather than open-ended general search, because errors are manageable in-domain but dangerous in open-ended settings. Cloud and infrastructure as the likely AI battleground (Priority: 4/5): Evans suggests foundational AI primitives may become part of major cloud platforms, while industry-specific tools will be built by software companies that understand their domains. The commercial opportunity depends on which layer of abstraction proves durable: generic platform or vertical software.

Key Arguments: The recent selloff in tech reflects a correction from pandemic-era exuberance, not an end to digital transformation; internet usage, smartphones, and e-commerce penetration continue to rise. Retail, advertising, and media are converging because online platforms now own both discovery and conversion, making ad spend, rent, shipping, and margin decisions parts of one optimization problem. Amazon’s ad business is so large because it monetizes high-intent traffic, first-party data, and marketplace control, effectively turning commerce pages into media inventory. Marketplace models allow platforms to scale infinitely by letting third parties sell through them, but that also creates infinite product choice and worse usability for consumers. Shopify represents an unbundling of Amazon for merchants, but its long-term power depends on whether it can create consumer-side network effects without undermining merchant brands. Generative AI will be most valuable where users can evaluate output quality and errors are bounded; general search is riskier because users cannot easily verify correctness. LLMs may become invisible infrastructure embedded in software and workflows rather than a standalone consumer product centered on chat. Cloud providers are likely to host core AI primitives, but vertical applications will capture much of the value because industry knowledge matters more than raw model capability.

Data Points: Amazon advertising revenue: $38 billion - Evans cites Amazon’s ad revenue as the size of a major media business and compares it to other categories. Amazon ads vs. Prime: Advertising is roughly the same size as Prime - Used to illustrate how large Amazon’s ad business has become. Amazon ads vs. newspapers: Bigger than the global newspaper industry - Shows that retail media has surpassed a traditional media category. Amazon as media owner: Fifth or sixth biggest media owner on earth - Depending on how media ownership is measured, Amazon now ranks among the largest. Walmart ad business: $2.5 to $3 billion in 2022 - Evans says Walmart is rapidly building a retail media business. Uber ad business: $500 million run rate - Example of non-traditional platforms monetizing traffic via ads. Amazon marketplace share: About 60% of what gets sold on Amazon is sold by third parties - Illustrates how marketplace scaling depends on outside sellers. Shopify scale: 40–45% as big as the Amazon marketplace - Shows Shopify’s significant footprint in e-commerce infrastructure. Chinese marketplace sellers going direct: Between one-third and one-half - Marketplace Pulse estimate cited for Chinese manufacturers selling directly. E-commerce penetration: 20%+ - Evans says e-commerce penetration remains well above pre-pandemic levels. Smartphone adoption: Five billion of six billion adults - Used to argue tech adoption remains structurally deep. Digital advertising share of global ad market: Probably three-quarters - Evans uses this to explain why Google and Meta remain dominant. Cloud adoption in big enterprises: 20–25% - He cites Goldman Sachs CIO survey as evidence that cloud still has a long runway.

Pivotal Quotes: "We had a sort of certain loss of discipline, a certain overenthusiasm." — Benedict Evans: Describing the pandemic-era tech boom and why valuations later corrected. "All of the definitions break apart when it's all kind of happening on the same website and doing the same thing." — Benedict Evans: Explaining why the boundaries between advertising, marketing, and retail no longer hold online. "The magic and the disaster is two in one." — Benedict Evans: Summarizing why LLMs are powerful but hard to deploy safely for general search.

Implications: Retail media, AI, and cloud are converging into new platform gatekeeping layers. Companies that control traffic, data, and workflows will keep gaining leverage, but the biggest near-term AI value may go to specialized software, not a single chatbot winner.