How solid is Ed Zitron's 'Case Against Generative AI'? (thenewstack.io)

Tech critic Ed Zitron published an 18,000-word polemic arguing generative AI is a speculative bubble that will collapse — not because the models are intellectually bankrupt, but because the economics are. He claims demand is overstated, profitability is elusive outside a few giants, and runaway infrastructure costs (GPUs, specialized “Neocloud” data centers like CoreWeave, Lambda, Nebius) make the stack unsustainable. Zitron cites deals and signals — NVIDIA’s $1.5B GPU leaseback, reports of massive per-user losses for services like Microsoft Copilot (he estimates Copilot revenue at ~$2.9B/year), and an allegedly circular ecosystem where NVIDIA both funds and sells into neoclouds — to argue investors are funding growth without durable unit economics. The piece matters because it reframes AI debate from capability to cash flow: if training, serving and GPU capacity remain expensive and enterprise pilots don’t convert to profitable scale, investor losses and neocloud failures could ripple through the market. Critics push back: analysts note declining per-customer service costs, elastic API pricing and tiered product strategies can rein in costs; ecosystem investments are common business practice. Technical implications are clear — volatile GPU demand, capital-intensive infrastructure, and uncertain adoption mean 2026 hardware/cloud spend could fall, hurting specialized providers and public investors even if core model R&D and hyperscaler deployments continue.