# After Microsoft’s $357B Loss: Why Content Is the Next Bottleneck
Microsoft lost $357 billion in market value on January 29, 2026—the [second-largest single-day loss](https://www.cnbc.com/2026/01/28/microsoft-msft-q2-earnings-report-2026.html) in stock market history. Azure grew 39%, just below the 39.4% analysts expected. A fraction of a percentage point cost shareholders more than the entire market cap of 90% of S&P 500 companies.
The miss wasn’t about demand. Ben Thompson’s analysis “[Microsoft and Software Survival](https://stratechery.com/2026/microsoft-and-software-survival/)” explains: Microsoft prioritized internal AI products—M365 Copilot, GitHub Copilot—over Azure customer capacity. CFO Amy Hood confirmed that allocating those GPUs to Azure would have pushed growth “over 40.”
Thompson’s thesis: “The last decade was about growing the pie; the next decade will be about fighting for it, and the model makers will be the arms dealers.”
He focuses on compute allocation. The bigger story is what comes next. If model makers are the arms dealers, someone needs to supply the ammunition. Every AI company fighting these adjacency wars needs one resource that cannot be synthesized: licensed content to train on.
## The Adjacency Wars
Thompson describes a fundamental shift in SaaS economics. Per-seat licensing—the model that grew as companies hired—is dying. AI agents reduce headcount while expanding capabilities. SaaS companies are responding by using AI to attack adjacent markets.
M365 Copilot competes with specialized productivity tools. GitHub Copilot threatens code generation startups. Every major AI product launch doubles as a market invasion.
Each new capability requires training data. Each adjacent market needs domain-specific content to make the model credible. Enterprise deals with major publishers cover news archives, stock photography, academic papers. They miss the long tail.
## The Licensing Landscape in 2026
2025 changed the calculus for AI content licensing. OpenAI, Google, Microsoft, and Meta committed [nearly $3 billion](https://www.cbinsights.com/research/ai-content-licensing-deals/) in licensing deals. OpenAI alone has [18 publisher agreements](https://digiday.com/media/publishers-scorecard-for-big-techs-ai-licensing-deals/) globally, including News Corp, The Atlantic, Vox Media, Axel Springer, and the Financial Times.
Meta entered the market in December 2025, [signing deals with CNN, Fox News, People Inc., USA Today, and Le Monde](https://www.axios.com/2025/12/05/meta-ai-deals-news-publishers) to feed real-time news into its AI assistant.
Then came the settlement that reset expectations. In September 2025, Anthropic agreed to [pay $1.5 billion to authors](https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai) whose works were downloaded from pirated databases for Claude’s training—the largest copyright settlement in U.S. history. The settlement established a baseline of approximately [$3,000 per copyrighted work](https://authorsguild.org/advocacy/artificial-intelligence/what-authors-need-to-know-about-the-anthropic-settlement/).
The message: unlicensed training creates billion-dollar liabilities.
## The Content Bottleneck
In 2023, the bottleneck was GPUs. AI companies couldn’t train models fast enough because compute was scarce. CoreWeave and similar providers solved this by offering neutral infrastructure—GPU capacity without competing against customers.
In 2026, the bottleneck is content access.
Enterprise deals cover mainstream media. AI systems need something broader: niche technical blogs, specialized trade publications, community forums, independent Substacks, documentation sites. The billions of web pages that constitute the internet’s actual knowledge base.
[W3Techs reports](https://w3techs.com/technologies/details/cm-wordpress) that WordPress powers 43.5% of all websites. Individual creators, small publishers, domain experts sharing knowledge without corporate backing.
## Neutral Infrastructure
Copyright.sh operates as licensing infrastructure—the layer that lets any AI company access any content under clear, standardized terms. No AI models, no end users, no competing products.
The adjacency wars require speed. A SaaS company attacking a new market needs domain-specific training data immediately, not after six months of enterprise negotiations. They need to discover, license, and pay for content at scale—across millions of sources.
Meta tag-based licensing enables this:
“`html
“`
One line of HTML. One API call. Automatic license discovery, transparent pricing, clear terms. The creator gets paid, the AI company gets legal certainty.
## The Long Tail
Enterprise content deals are table stakes. Every major AI company has them. Competitive advantage comes from accessing content others cannot reach: the independent blogger who wrote the definitive guide to industrial valve systems, the technical writer who explained distributed systems better than anyone, the niche community with a decade of accumulated expertise.
This content has no licensing team. One person, or a small team, who needs licensing to be as simple as adding a meta tag and connecting a payment account.
When licensing infrastructure lives in WordPress, Ghost, and Substack, millions of content creators can participate without technical expertise or legal resources.
## Usage-Based Economics
Thompson highlights a critical problem: per-seat licensing breaks when AI reduces headcount. Pricing models that depend on companies hiring more people bet against their own product’s value.
Usage-based licensing aligns incentives. Charge per AI interaction instead of per employee. Revenue scales with AI success, in proportion to actual usage. A company testing AI in a new market pays only for what they use.
## The Choice
As adjacency wars intensify, AI companies face a decision: build on licensed content with clear legal standing, or build on unlicensed content and hope the lawsuits never catch up.
After Anthropic’s $1.5 billion settlement, the economics favor proactive licensing. Enterprise deals alone won’t provide the content diversity AI systems need to compete in adjacent markets.
The companies that win need the best content access. In the fight Thompson describes, ammunition matters as much as weapons.
—
*Copyright.sh provides licensing infrastructure for AI companies and content creators. Learn more at [copyright.sh/ai-companies](https://copyright.sh/ai-companies).*
Leave a Reply