Fluidstack becomes AI infrastructure originator
Fluidstack
These TeraWulf contracts mark Fluidstack’s shift from selling short GPU rentals to locking in utility-like infrastructure revenue for a decade. The important change is not just size, but credit quality and duration. Instead of depending on many smaller marketplace deals, Fluidstack now has 360 MW of pre-leased AI data center capacity at Lake Mariner with about $6.7B of base term revenue beginning in 2026, and Google’s support makes those payments financeable like infrastructure cash flows rather than startup demand forecasts.
-
The Lake Mariner relationship was built in steps. TeraWulf first announced 200 plus MW of 10 year leases worth about $3.7B, then expanded with an additional 160 MW CB-5 lease, bringing Fluidstack’s total contracted capacity there to roughly 360 MW and total base term revenue to about $6.7B.
-
Google’s role matters because it is not the end tenant at Lake Mariner, it is the credit enhancer. TeraWulf disclosed about $3.2B of Google credit support across the three Lake Mariner leases, plus equity warrants for Google, which lowers lender risk and helps turn a young GPU cloud’s demand into debt financeable projects.
-
This is also a clue to how Fluidstack competes with CoreWeave and Crusoe. The company is no longer only brokering spare GPUs. It is signing multi hundred megawatt capacity blocks, pairing them with outside credit support, and using hosts like TeraWulf to stand up dedicated private cloud clusters for very large AI customers.
From here, Fluidstack looks increasingly like an AI infrastructure originator, not just a GPU marketplace. If Lake Mariner deliveries start on schedule in 2026, these contracts become the template for more long dated, pre financed capacity deals, with Google backed commitments helping Fluidstack scale from hundreds of millions in revenue toward a much larger infrastructure base.