Newsletter

Chips & Infrastructure

Deep Dive — Alibaba wants to own the whole stack. Memory says no.

The Arena

Deep Dive — Alibaba wants to own the whole stack. Memory says no.

The most important number Alibaba put on stage in Hangzhou this week was not ten trillion. Alibaba opened its annual Apsara Conference on Tuesday with a plan that is unusually easy to restate, because the company said it in one sentence: it will train Qwen models of five trillion to ten trillion parameters, run them on its own accelerators wired into clusters of up to 500,000 cards, and operate more than 20 gigawatts of data centre capacity worldwide by 2032.

AI 101 — What is mechanistic interpretability?

The Guardrails

AI 101 — What is mechanistic interpretability?

Mechanistic interpretability is the work of reverse-engineering what actually happens inside a neural network — finding the specific internal parts responsible for a specific answer, rather than inferring everything from the model's behaviour. "Mechanistic" is the load-bearing word: the deliverable is a mechanism, a step-by-step account of the computation, not a description of what the model tends to do. Why it matters right now The cheapest way anyone audits a frontier model is by reading…

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

The Arena

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

Alibaba's cloud conference in Hangzhou opened this morning with a three-part commitment that is harder to wave off than the usual keynote arithmetic: a model bigger than anything trained so far, the chip to run it, and the power to feed both. Huawei spent the same week putting a coding model inside its own operating system, and a small project on the local-AI forums quietly showed that a model's refusals can be switched off without touching a single weight. Alibaba says its next Qwen models…

Amazon pledges $20M to the Colorado River without its own water data

The Everyday

Amazon pledges $20M to the Colorado River without its own water data

Amazon put $20 million into the Colorado River this week — and still won't say how much water its data centers take out of it. Amazon announced Monday that it will spend $20 million on water conservation across the Colorado River basin, as part of a new corporate coalition targeting $100 million over the next two years. The basin supplies 40 million people across the western United States and is in what federal forecasters call an unprecedented drought, with Lake Mead still declining.

Europe puts a data centre energy label on the grid, California sends the bill

The Guardrails

Europe puts a data centre energy label on the grid, California sends the bill

The European Commission proposed a mandatory A-to-G sustainability label for data centres, with the first labels expected to appear in 2027. The delegated regulation, published Monday, requires operators of facilities of 500 kW or more to disclose how efficiently they use energy and water, how clean their electricity is, and whether they can hand waste heat back to the local grid — plus information on how their water use sits against water stress in the area.

Local opposition blocked $68B of US data centers in one quarter

The Guardrails

Local opposition blocked $68B of US data centers in one quarter

Two hard numbers landed today on the two most-hyped physical bets in AI — the buildings that train the models and the robots meant to work in them. Both came in far smaller than the money already committed to them. Local opposition blocked or delayed 45 US data center projects worth $68 billion between April and June — more than half of every large-scale development the research group Data Center Watch tracked in the quarter.

How to — cut your LLM bill without switching models

The Stack

How to — cut your LLM bill without switching models

Your token spend keeps climbing, and the obvious fix — swap in a cheaper model — is the one you should try last. In a live AI feature, most of the waste isn't the price per token. It's paying full price for tokens the provider has already seen, and generating output tokens nobody asked for. Work the list in this order. The first three moves usually move the number more than a model swap would, and none of them cost you quality. 1.

See all