Reading the Roadmap
The market leader now ships on a one-year rhythm, and AMD has matched it. Hyperscalers run their own chip programmes in parallel: Google's TPU since 2016, Amazon's Trainium, Microsoft's Maia and Meta's MTIA. The result is more announcements than any buyer can evaluate on specifications alone. Faraday's read is that the discipline lies in the stage, not the figure. Announced, in production, available and deployed are four different claims, and only the last one draws power. Read every roadmap for its memory, its scale-up domain and what it asks of the building, and treat its far end as intent rather than schedule.
Knowledge check
Why is a statement that a chip is in full production a weak indicator of when it will be running in data centers?
Show the answer
Full production describes the factory: manufacturing has started at volume. The product must still be built into systems by partners, offered to customers and installed in facilities with suitable power and cooling, and each of those stages can take months. Only deployment at scale affects power demand and revenue in the way investors and operators care about.