The Main Story: This Isn't the Dot-Com Bubble. It's Something Else Entirely.
Every time a new technology attracts irrational-looking levels of capital, someone trots out the dot-com comparison. Pets.com. Webvan. The sock puppet. The narrative is comforting because it's familiar: smart money piles in, dumb money follows, the bubble pops, and everyone pretends they saw it coming.
But the AI investment cycle looks nothing like 1999. The dot-com bubble was driven by companies with no revenue, no business model, and no technical moat. The AI buildout is being driven by the most profitable companies in history — Microsoft, Google, Amazon, Meta — spending their own cash flow on infrastructure that they need for their core businesses. When Microsoft spends $80 billion on AI data centres, it's not a speculative bet. It's a capacity investment for Azure, which is already generating tens of billions in annual revenue from AI workloads.
The more accurate historical parallel is the railroad boom of the 1860s or the fibre optic buildout of the late 1990s. Both involved massive over-investment in infrastructure. Both produced spectacular individual failures. And both created the physical foundation for decades of economic growth. The companies that laid too much fibre went bankrupt. But the fibre itself carried the entire internet economy for the next twenty years.
The same dynamic is playing out with AI infrastructure. Some of the capital being deployed will be wasted. Some companies building AI applications will fail. But the infrastructure — the chips, the data centres, the models, the tooling — will remain, and it will be used. The question is not whether AI infrastructure is valuable. It's whether the current rate of investment overshoots short-term demand. And even if it does, the long-term correction may look more like a slowdown than a crash.
Quick Take: Anthropic's Valuation Leap
Anthropic closed a new funding round valuing the company at $60 billion, roughly tripling its valuation in under a year. The round was led by existing investors doubling down on the company's position as a leading frontier AI lab. Claude, its flagship model, has gained significant market share in enterprise deployments, particularly in regulated industries where safety and reliability matter more than raw capability benchmarks. The AI market is consolidating around a small number of frontier labs, and Anthropic is firmly in that group.
The Stat That Matters
$300 billion — estimated global spending on AI infrastructure in 2026, up from approximately $150 billion in 2024. This is the fastest infrastructure buildout in the history of technology, measured by absolute capital deployment. For context, the entire US interstate highway system cost roughly $530 billion in today's money — and that took 35 years.
What We're Watching
The emerging battle between NVIDIA's CUDA ecosystem and open-source alternatives (particularly AMD's ROCm and the growing Triton compiler community). If an open-source software stack achieves CUDA-level performance and developer adoption, it could break NVIDIA's ecosystem moat and introduce real price competition into the AI chip market for the first time. Early signs suggest this is at least three years away — but the trajectory is worth monitoring.