Meta has an AI product problem (techcrunch.com)

Meta revealed that its aggressive AI buildout — massive hiring, two new hyper-scale data centers and what could be hundreds of billions of dollars in U.S. infrastructure spending — has started to show up on the balance sheet: operating expenses jumped ~$7B year‑over‑year and capex surged by nearly $20B in the quarter. CEO Mark Zuckerberg says the outlays are to secure compute for “frontier” models from the Superintelligence (MSL) team and to power new AI products, but analysts heard few concrete monetization plans. The market reacted harshly: Meta’s stock fell about 12% in days following the call, wiping out some $200B in market value. For the AI/ML community the story underscores a growing tension between build‑first investments in compute and talent versus the need for fast, productized revenue. Technically, Meta is doubling down on large‑scale compute and frontier model development (the implicit aim is novel capabilities beyond existing LLMs), but current outputs — Meta AI (1B+ users largely via existing apps), Vibes video generation and Vanguard smart glasses — are experimental and have limited direct monetization. Compared with peers (Nvidia’s infrastructure engine, Google’s AI products, or OpenAI’s rapid consumer growth), Meta lacks a clear high‑velocity revenue stream to justify this level of spend, raising questions about ROI, product roadmap clarity, and regulatory/data‑use strategies as it races to translate raw compute and models into sustainable products.