$IREN: PG Thermal Imaging Thanks to @AceKicker13 for taking drone photos of IREN’s PG site. Everything is humming along and billing.

Although I think middle case for AI is most likely, the two extremes are not being given equal weight. Alot of AI bears mapping out overbuild bubble but very few are considering implications of ASI. In ASI, asset would become an even more dominant than labor and now knowledge work via robotics and agents. In particular AI assets: semiconductors, power, attention economy.

$IREN x Anthropic: Win-Win Synergy The $IREN community has been expecting an $IREN x Anthropic deal for a while with the most recent cliff hanger being that $IREN has signed with one of the two frontier labs said in a management call. The Information Leak Yesterday, @theinformation had a leak where $IREN is one of the AI Clouds that Anthropic has signed for capacity. This is no surprise because Anthropic is out bidding the rest of the market for capacity whether bare metal or managed services at managed services rates. It's 30m/MW bare metal deal for H100/H200 with SpaceX is at managed services rate. Why NDA The Anthropic deal with IREN cannot be formally announced likely due to the fact that additional capacity is being negotiated. I do not think Anthropic IPO is preventing the information from being made public because Anthropic has made other AI Clouds public. On the otherhands, the other AI Clouds do not have alot of unsold capacity coming up in 2027 unlike $IREN. Additionally $IREN likes to sign contract close to delivery as possible while many other AI Clouds like to show case a massive backlog. These two facts together make it likely that Anthropic signed a small contract for 2026 or very early 2027 capacity and additional negotiations are in flight. Why High Rates for Bare Metal The big question is what are the rates? SpaceX has shown that Anthropic is signing bare metal capacity at managed service rates in it's 30m/MW deal for H100/H200. Even for surge pricing and monthly duration, these are much higher than bare metal rates for H100/H200. Anthropic runs their own optimized inference stack on top of bare metal thus has no use for managed services. However, Anthropic needs to outbid everyone else on the market. Anthropic earns more per MW than startupus so it can outbid everyone and the clearing price is managed services rate. At the end of the day, Anthropic is a full AI Cloud is optimized to serves 1 set of models: Claude + Fable. Why would Anthropic pay managed service rates for bare meta? For Anthropic, they already have the full inference stack software that runs on bare metal GPUs. A org of 100 elite Anthropic engineers can write an optimize inference stack and then a small org of 40 software ops engineers can support 500MW-1GW of GPUs but much more DC technicians aka bare metal software operators are needed. IREN bare metal includes DC technicians and that's all Anthropic needs. In other words, software scales but physical operations do not. This is because much of software ops is routine responses encoded in runbooks with automated script responses. If needed dispatches to DC technicians are needed it can be simultaneously to up to all the GPU racks at once. This is not downtime as there is usually a fallback state while DC technicians get to all the GPUs but it still has to be done in a timely manner.