Skip to content
3 min read

Roundup: Inference Costs, China Robots, Labor Markets, Credit Spreads, etc.

Catching up during a week full of meetings and client calls. I have a longer thing coming on the new Xerox, a role for activists, and more, but that will have to wait.

Some things I'm thinking about ...

1. AI Inference Costs Are Coming Due

What’s happening
AMD is buying Taalas for the latter's inference technology. I have spoken about Toronto-based Taalas a few times, with their innovative idea of embedding small models in silicon to drive down costs. On the cost front, DeepSeek, which was lauded last week for driving prices down, is planning a price increase after training the market to expect very cheap model access.

Why it matters
The cheap-AI story remains fragile, outside sovereign subsidies and land wars from pre-IPO companies. Inference is expensive, scale is capital intensive, and someone has to pay for the power, chips, and data centers.

Sources
AMD to Acquire Taalas for AI Inference Chips in Data Center Push
DeepSeek Plans ‘Significant’ Price Increase for AI Services

2. China Wants Robots to Be the Next EVs

What’s happening
Unitree is raising $900+ million in a Shanghai IPO. It has real shipments, strong margins, and plans to spend heavily on humanoid robots and embodied AI.

Why it matters
This looks like the familiar China playbook: subsidized scale, vertical integration, lower hardware costs, then a fight over software and export controls. The humanoid robot market in China is awash in capital and companies—proving my point that the future will see far more AI-aided competition in every zero-billion-dollar category.

Sources
Unitree Robotics Plans $904 Million IPO as China’s First Humanoid Robot Maker

3. AI Debt Still Clears, But the Easy Part Is Over

What’s happening
Alphabet pulled in huge demand for a bond sale tied to AI spending, even if at higher spreads. At the same time, investors are pushing back on data-center CMBS and demanding better terms and wider spreads.