The AI party keeps roaring: Why it won’t end anytime soon
SiliconANGLE Robert Hof
AI spending is still roaring, and Dave Vellante says the bubble may not burst soon. The bottleneck is parts and power, not demand — so the party keeps going.
Based on reporting by SiliconANGLE, Robert Hof — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Yes, the AI bubble is real. But the bigger question isn’t whether it exists; it’s whether it can actually pop before the machinery underneath it gives way. Dave Vellante’s latest Breaking Analysis argues that the answer is probably no, because the AI supply chain is still jammed up on high-bandwidth memory, advanced packaging, network fabric, power and site readiness.
That matters because scarcity changes the timing. If the parts, power and facilities needed to build AI systems stay tight through at least 2028 — and maybe longer — then buyers don’t get the luxury of much choice, and the market doesn’t get a clean moment to discover whether it overbuilt. In other words, the usual bubble reckoning gets pushed out.
Vellante also says Nvidia’s Jensen Huang may be reshaping how the whole buildout gets financed. The idea is blunt: make the AI infrastructure stack so large and so interconnected that it becomes hard to unwind quietly. That’s not a comforting phrase, but it does fit the scale of the money flowing into model makers, chip makers and the companies supplying the equipment behind them.
The latest earnings season offered a split-screen view of that bet. CoreWeave, Nebius and Supermicro were rewarded after strong results. Cisco, Cerebras and Applied Materials also beat expectations, but their stocks did not get the same love. Next week is lighter on earnings, then Nvidia, Salesforce and others show up the week after.
Beyond the chips and servers, the strategic bets are starting to diverge. SemiAnalysis argues Google has effectively ceded the frontier-model race after leadership changes around Demis Hassabis and Jeff Dean, while Tim O’Reilly says the bigger prize may be AI infrastructure through Google Cloud. Meta, meanwhile, keeps pushing open-weights models, even as Google, SpaceXAI, OpenAI and Anthropic keep shipping proprietary ones. And Databricks just raised $5 billion without going public, which tells you the money still likes private companies that look like they can grow forever.
My take — AI-written commentary, not fact-checked reporting
The market keeps pretending AI is one neat race for the best model, but the real money is clearly going into pipes, power and packaging. That’s a very 2026 kind of reveal: less glossy demo, more electrical grid. The open-versus-closed drama still sells headlines, but infrastructure is where the grown-ups are hiding the bill.
Read more about this at: SiliconANGLE