Cloudflare warns AI agents will kill ad economy

AI & Tech (12)

• Prince explains Cloudflare started as a cloud-based firewall and grew by solving emergent problems for clients, now serving 20% of the internet, 80% of major AI labs, almost 100% of crypto companies, and a rising share of Fortune 500 firms.
• Prince describes Cloudflare's defensive role, handling tens of millions of attacks per second and protecting Ukraine's internet, which led to his personal sanction by the Russian government.
• Prince says internet website creation plateaued from 2011-2012 until recently, but is now growing again at early-2000s rates due to tools enabling more creators and developers.
• Prince contends AI answer engines like ChatGPT are strip-mining the web, pulling content without sending human traffic back, causing a step-function drop in ad revenue for publishers and threatening their survival.
• Prince says AI subscriptions could create a global information divide, shrinking the internet for those in the global south who cannot pay, while trustworthy, unbiased agents become a premium product.
• Prince predicts AI companies will evolve like YouTube, competing by securing unique content niches and compensating creators, potentially leading to a golden age of knowledge-focused content.
• Prince notes Google has vastly more web data than competitors: it sees four times more pages than OpenAI, five times more than Microsoft, six times more than Anthropic, and 22 times more than X.ai.
• Prince clarifies Cloudflare's Content Independence Day gave all customers tools to control AI crawler access, not a blanket block, because many sites want their content in AI systems.
• Prince argues creators must control who accesses their content because if agents consume it without compensation, the creators' business model collapses.
• Prince envisions a future where AI companies identify gaps in human knowledge and compensate creators to fill them, moving from a traffic-based to a knowledge-creation economy.
• Prince observes that local, unique content like his Park Record newspaper is becoming more valuable to AI companies than generic national news, and may earn more from AI licensing than digital ads.
• Prince warns the natural tendency is toward extreme centralization, with few AI companies, content creators, and giant corporations dominating agent commerce, bypassing small businesses.

Big Tech (1)

• Prince argues Google's ad-driven business model, which rewarded traffic, degraded into a rage-bait economy, fueling populism and division by incentivizing incendiary headlines over knowledge creation.

Protocol (2)

• Prince says a micropayment system is essential for the new internet economy, but current crypto rails cannot scale to the needed volume of 5-50 million monetizable transactions per second Cloudflare handles.
• Prince states Cloudflare is working on the 402 payment standard with Coinbase and others, seeking a stablecoin solution that can scale to 100 million TPS, which no existing blockchain supports.

AI & Tech (26)

• Jack Farley notes that Anthropic has surpassed OpenAI as the most popular model for businesses, according to Ramp data, reflecting a dynamic market where newcomers displace incumbents, similar to Cursor replacing GitHub Copilot.
• Businesses increasingly use multiple AI models, become more cost-conscious, and explore AI applications beyond the prevailing narratives of automation and labor replacement.
• Ara Karazian, a lead economist at Ramp, utilizes spend data from 50,000 businesses, representing $100 billion in annual spend, to inform market analysis.
• Seat-based contracts still account for 65-75% of business software spend, and flat platform fees represent 20-30%, indicating minimal shift in purchasing models.
• Token-based pricing for SaaS tools, offered by companies like HubSpot and Adobe, accounts for less than 0.5% of spend on those platforms, showing limited uptake despite increasing availability.
• Figma remains a fast-growing vendor, with businesses continuing purchases despite new competition from AI-native design tools, demonstrating the stickiness of business software spend.
• Perplexity is one of Ramp's fastest-growing vendors, largely due to offering products that core AI model companies have not yet entered.
• An increasing share of firms uses multiple AI models, with early adopters being most likely to integrate additional models and AI vendors over time.
• Businesses are growing more cost-conscious about AI spending; token costs for high-intensity users surged 13x in the last year, now comprising about 2% of their business spend excluding payroll.
• High-intensity AI spenders are increasingly shifting towards platforms like OpenRouter to select between multiple models, including cheaper open-source options, though this represents only 3% of total AI spend on Ramp's platform.
• Core model providers like OpenAI and Anthropic have little incentive to offer auto-routing products that lower AI spend, given that token usage accounts for 80% of their business revenue.
• Max Wheathe notes past speculation that 80% of venture capital-backed tech startups used DeepSeek due to its low cost; however, Ara Karazian refutes this, stating its peak adoption was less than 1% of firms.
• DeepSeek faces challenges in gaining broader adoption due to security perceptions, especially among businesses building customer-facing products, despite its potential cost advantages.
• Ramp's AI adoption index only tracks paid usage, meaning Google Gemini's integration into Google Workspace, which provides free access, causes its overall adoption to be underrated.
• Economists are skeptical that standalone chat subscriptions will drive significant economy-wide productivity gains, suggesting that more comprehensive AI integration is necessary.
• Ara Karazian states that AI adoption trends towards multiple large players, not a single dominant entity. Ramp's future research will measure adoption intensity and define what constitutes successful AI integration.
• Ara Karazian suggests that many layoffs attributed to AI are likely due to prior overhiring, with AI serving as a convenient justification for workforce reductions.
• There is a decoupling of revenue growth from headcount growth in software companies, but firms effectively adopting AI tend to be fast-growing with many opportunities.
• Ara Karazian questions the strategic benefit for AI model companies promoting the narrative of AI destroying all jobs, suggesting it's not helpful to them. Max Wheathe counters that it targets investors to justify high valuations.
• Adio, a rapidly growing London-based CRM, positions itself as an AI-native alternative in a market where Salesforce holds an 80% share, highlighting competition from specialized AI-driven solutions.
• Ramp's aggregated, anonymized business spend data is now accessible via cloud connectors and OpenAI, which Ara Karazian believes improves market transparency and decision-making, despite potentially harming data-selling businesses.
• Legacy software companies face disruption not only from AI but also from inherent market churn if they fail to innovate and respond to changing customer needs.
• Many legacy firms are hesitant to rapidly implement AI, exemplified by Deloitte adopting a conservative "augment but not replace" approach to AI, contrasting with other accounting firms.
• Legacy media companies, such as The New York Times, are actively engaging in licensing deals with AI model companies, positioning themselves to potentially benefit from industry transformations despite internal newsroom controversies.
• Ara Karazian believes AI is generally poor at opinionated or creative writing but excels at highly structured tasks like summarizing and generating bullet points, which could allow reporters to focus on core investigative work.
• XAI, a relatively late entrant, achieved 3% adoption within months of launch but has not sustained the vertical growth seen by Anthropic. Its acquisition of Cursor could be a strategic move to drive model adoption.

Business (2)

• Ara Karazian argues that pronouncements of a "SaaSpocalypse" - either a shift away from traditional SaaS or a change in buying models - are premature and lack support from actual business spending data.
• The prevailing two-year narrative, that AI would eliminate SaaS, collapse seat-based pricing, and centralize models, is difficult to defend when analyzing actual business spending patterns.

AI & Tech (1)

• Google reported a 40% quarter-over-quarter surge in paid enterprise Gemini customers. The company's infrastructure now processes 16 billion tokens per minute, a 60% increase from the previous quarter.

Models (1)

• Sundar Pichai says Google's AI models are at the frontier in overall capabilities, including text, multimodality, voice, audio, reasoning, and intelligence, but acknowledges the company is behind the frontier in agentic coding, tool use, and long-horizon tasks.

Coding (1)

• Pichai identifies coding, particularly for developers working on complex codebases, as a critical frontier and foundational to all Google's work, noting a current gap compared to competitors.

Big Tech (1)

• Pichai asserts Google is the only large company operating at the AI frontier, framing the competitive landscape as a few startups versus Google's long-term, scaled development.

Society (1)

• Pichai views the public's anxiety about AI as natural given its profound and rapid pace of change, arguing the industry must better demonstrate AI's benefits and that democratic dialogue on the topic is healthy.

AI & Tech (7)

• Pichai predicts AI will elevate the baseline capability for everyone, similar to how spreadsheets democratized financial analysis, and enable more people to code, fostering new serendipitous economic outcomes.
• Pichai uses a radiologist example to argue AI demand will be nonlinear, citing that modern digital scans contain 10x more data than older film-based ones, projecting another 10x increase in data volume within 10 years.
• Pichai believes AI's economic value will increase over time, justifying a combination of subscription and advertising models to sustain it, referencing Adam Smith's rules as unchanged.
• Pichai sees a continuum in Google Search evolution, bringing users along methodically from classic results to AI Overviews, but asserts sources and links will always be part of the experience.
• Pichai says recursive self-improvement represents a next level of acceleration with significant implications, but current progress is a continuum and true RSI is not yet achieved.
• Pichai reaffirms Google's foundational pivot to AI ten years ago, stating inevitable progress toward AGI is underway, with recent progress suggesting a timeframe closer to 3-5 years than 5-10.
• Pichai clarifies Demis Hassabis's 'foothills of the singularity' statement refers to the advent of AGI, likely by 2030, and believes it's important for frontier builders to articulate this to help society prepare.