Microsoft Keeps AI Capex Unchanged at $190B — First Hyperscaler to Hold the Line
On Wednesday, Microsoft kept its AI capex forecast unchanged at $190B for the calendar year (adjusted to $175B due to an accounting change), making it the first major hyperscaler to hold the line on spending. The stock jumped 8% on the news.
Context: The Industry's Spending Spree
Google, Amazon, Microsoft, and Meta had already planned to spend over $700B this year on AI data centers. Most competitors are still increasing their forecasts:
- Alphabet increased capex by $15B (stock fell sharply).
- Tesla bumped up its projection (stock fell).
- Meta narrowed its range, raising the midpoint by $2.5B.
What's Driving Costs?
Soaring memory chip prices are a major factor. Memory is a large component of data center costs, and higher prices have made it more expensive to build AI capacity. One research finding: higher memory prices explain about 45% of the growth in cloud companies' capex this year.
Microsoft's decision to keep spending flat may imply it's actually pulling back on building new capacity, since rising costs would otherwise push the dollar figure higher. For developers, this could signal a stabilization in AI compute supply — or at least a pause in the relentless expansion.
📖 Read the full source: HN AI Agents
👀 See Also
Why AI Is Still Hard to Fully Deploy Across Enterprise Domains
A Reddit discussion highlights that probabilistic AI models struggle in high-accuracy fields like scientific research and report generation, where basic errors are unacceptable.
Fine-tuned Qwen2.5-7B to 96% of Claude Haiku with $3 and Zero Human Labelers
A developer fine-tuned Qwen2.5-7B to 96% of Claude Haiku's performance on a domain-specific decision-reasoning task using only $3 in API calls and zero human labelers, using a novel DV-DPO method.
Why OpenClaw's Open Source Architecture Matters
UW Researchers Plan to Use Teacher-Worn Cameras for AI Training, Parents Opt-Out
University of Washington researchers planned to have preschool teachers wear first-person cameras to record children for AI model training, with an opt-out consent model.