#AI horizons 25-07 – Market trends

July delivered key structural shifts across the AI ecosystem. Cloudflare’s crackdown on unauthorized scraping marks a turning point in how content owners assert control, while Amazon’s launch of its AI Agent Store expands commercial access to autonomous software. Apple is reportedly licensing Anthropic’s Claude after losing confidence in its in-house models. OpenAI is building a native shopping checkout system, signaling a revenue diversification push. Meanwhile, Meta is quietly stepping away from its open-source strategy, and AI agent traffic continues to reshape internet dynamics. These developments underscore the platformization of AI and its march into monetized infrastructure.

Key Points

• Cloudflare blocks AI crawlers by default and introduces a Pay Per Crawl monetization system
• Apple is licensing Anthropic’s Claude for Siri and investing heavily in cloud-based model deployment
• Amazon unveils its AI Agent Store, competing with Microsoft and Google’s marketplaces
• Adobe reports a 3,200% increase in AI-generated Prime Day shopping traffic
• OpenAI begins testing in-chat checkout, pushing ChatGPT into commerce
• Meta signals future superintelligence models may no longer be open-source
• Google licenses Windsurf AI for $2.4B after OpenAI acquisition fails
• Microsoft, OpenAI, and Anthropic fund a $22.5M AI training initiative for U.S. teachers
• DeepMind launches AlphaEarth Foundations to turn planetary data into AI-ready maps
• Skild AI unveils Skild Brain, a foundational model for general-purpose robotics

In-Depth Analysis

Cloudflare Blocks AI Crawlers, Introduces “Pay Per Crawl”

Cloudflare has announced it will now block AI crawlers by default for new customers, marking a shift in web infrastructure policy. It also unveiled a “Pay Per Crawl” program, allowing publishers to monetize AI access to their content. The system uses the long-dormant HTTP 402 status code (“Payment Required”) to give site owners three choices per AI crawler: allow, block, or charge.

This move is backed by publishers including The Atlantic, Fortune, Quora, and Stack Overflow. As CEO Matthew Prince told The Verge, “People trust AI more over the last six months, which means they’re not reading original content.” With OpenAI’s GPTBot traffic up 305% year-over-year, the initiative lays the groundwork for standardizing compensation across AI content pipelines.

Cloudflare’s roadmap anticipates a future where AI agents—not humans—make most internet queries. A user might assign a budget to their AI assistant to retrieve legal advice or academic studies, and Cloudflare’s infrastructure will enforce paywalls or charge on the fly. This turns the open web into a transactable platform for intelligent agents.

Apple Bets on Claude as Internal Siri Models Stall

Apple has reversed course on its internal AI model strategy after tests revealed its homegrown LLMs underperformed. According to The Information, executives found Anthropic’s Claude the most reliable among tested third-party models. As a result, Apple is negotiating to license Claude while also approving a multibillion-dollar budget to run its own models via the cloud in 2026.

A roughly 100-person internal team continues development, but Apple’s decision echoes past moves where it leaned on external tech (e.g., Intel chips, then M1). This pivot comes amid a broader shift across Big Tech toward licensing or co-developing with foundation model providers instead of going fully solo.

Amazon Enters the AI Agent Market

At its AWS Summit in New York, Amazon announced the launch of an AI Agent Store, a marketplace for prebuilt and customizable AI agents. The platform enables businesses to browse, compare, and deploy agents directly from their AWS consoles. Early partners include Anthropic and other open agent vendors.

Following Microsoft’s Copilot Runtime and Google’s Vertex Agent Builder, Amazon’s offering adds competition in a market racing to platformize agent-based computing. Usage is expected to grow among developers, startups, and enterprise IT teams seeking automation without building agents from scratch. Each agent will be monetized through usage- or subscription-based pricing, with Amazon taking a standard revenue cut.

AI Dominates Prime Day Shopping Behavior

AI tools were central to consumer behavior during Amazon’s Summer Prime Week. According to Adobe Analytics, AI-generated visits spiked 3,200% compared to 2024. Fifty-five percent of surveyed users said they used AI to research products, while 47% leaned on AI for recommendations and 33% to generate shopping lists.

Although email and paid search still drove the majority of traffic, the findings suggest that AI-powered browsers and chatbots are reshaping digital commerce. This behavior shift gives platforms like OpenAI, Microsoft, and Perplexity a more strategic role in the buying funnel.

OpenAI Prepares Native Checkout Experience

OpenAI is reportedly developing a shopping checkout system within ChatGPT, according to a report from The Financial Times and confirmed by Reuters. This will allow users to make purchases directly in the chat interface, without being redirected to external retailers.

The feature is being tested with early Shopify partners and could expand to a broader merchant base. The business model is simple: OpenAI takes a commission on every completed transaction, giving the company a revenue stream from non-paying ChatGPT users. This move places OpenAI in competition with e-commerce ecosystems like Google Shopping and even Amazon.

Meta Signals End of Open-Sourcing Flagship Models

In a recent company letter, Mark Zuckerberg stated that Meta’s most advanced AI systems may not be released as open source due to safety concerns. This shift contrasts with earlier positioning that openness was Meta’s competitive advantage over rivals like OpenAI and Google. As reported by TechCrunch and Business Insider, Meta has paused testing on its next-gen LLaMA model and is investing in a private superintelligence program backed by $14.3 billion in funding.

This represents a broader industry trend toward hybrid release strategies—public weights for older models, guarded access for frontier models.

Google Licenses Windsurf AI for $2.4B

After OpenAI’s $3B acquisition of AI coding startup Windsurf fell apart, Google stepped in with a non-exclusive license deal worth $2.4B. The agreement gives Google access to Windsurf’s technology without acquiring the company. Windsurf’s top leadership, including CEO Varun Mohan, joined DeepMind, while the company retains most of its 250 employees and clients.

This “reverse acquihire” reflects a pattern where Big Tech gains talent and IP while avoiding antitrust complications. It also highlights a growing gap between model performance and deployability—Google’s Gemini agent line is under pressure to deliver developer-grade tools rapidly.

Microsoft, OpenAI, Anthropic Launch AI Teacher Training Program

A new coalition led by Microsoft, OpenAI, Anthropic, and the American Federation of Teachers has created the National Academy for AI Instruction, backed by $22.5 million in funding. The goal is to train 400,000 U.S. teachers over five years in integrating AI into the classroom.

The curriculum includes workshops on prompt writing, student use policies, and tool-specific training for Copilot, ChatGPT, and Claude. Based in New York, the program offers professional development credits and aims to become the national standard for AI literacy in education.

DeepMind Launches AlphaEarth Foundations

Google DeepMind released AlphaEarth Foundations, an AI system that compresses, processes, and maps massive planetary datasets. It uses radar, optical, and climate data to produce unified Earth maps at 10-meter resolution, reducing storage needs by 16x and error rates by 24%. Over 50 research groups—including MapBiomas and the Global Ecosystems Atlas—are using it for climate modeling and ecosystem tracking.

The model is now available via Google Earth Engine and supports backfilled data from 2017 to 2024. It simplifies geospatial research and lowers the barrier for integrating Earth data into AI tools.

Skild AI Brings Foundation Models to Robotics

Skild AI, backed by Amazon and SoftBank, unveiled Skild Brain, a foundational model trained across multiple robot types, simulations, and human-video demonstrations. Unlike traditional robot models limited to narrow tasks, Skild Brain learns continuously from both synthetic and real-world interactions. Cliets include LG CNS and multiple logistics and industrial partners.

The company raised $300 million last year and is now valued at $1.5 billion. Its bet: that shared learning across machines can close the generalization gap in robotics faster than bespoke training.

Business Implications

The monetization of AI content access—led by Cloudflare and publisher alliances—suggests a fragmentation of the training web. AI companies must now navigate a patchwork of licenses, fees, and consent mechanisms. For cloud vendors, the move toward agent marketplaces is a bid to lock developers and enterprises into ecosystem-specific agents and workflows. Apple’s pivot from builder to buyer reflects the rising costs and diminishing returns of independent foundation model development, a trend mirrored by Google’s Windsurf license.

OpenAI’s venture into commerce shows how AI interfaces are becoming not just tools but channels—first for discovery, now for transactions. Meta’s reversal on open source suggests that safety narratives will increasingly be tied to strategic control over model deployment. And Microsoft’s teacher training initiative represents a push to own the educational narrative of AI adoption in classrooms.

Why It Matters

The AI economy is moving from free exploration to structured monetization. What once flowed freely across the web—text, images, knowledge—is becoming priced, gated, and governed by agents. The infrastructure built by companies like Cloudflare, Amazon, and Microsoft is now being tuned for AI-first interactions. Meanwhile, the largest consumer platforms are converging on AI not just as a backend capability, but as the interface itself—whether in Siri, ChatGPT, or shopping flows.

Strategic advantage in AI now hinges not only on model quality, but on access, distribution, and revenue capture. The winners will be those who understand not just how AI thinks, but how it gets paid, licensed, trusted, and embedded.