Three thousand megawatts at $11m to $12m apiece is $33bn to $36bn, which makes the note Semafor appended to Amin's non-answer ("more than $30 billion") true and unhelpfully round [4][5][1]. Run his preferred 30-70 split across that and third parties are being asked for $23bn to $25bn, against which Blackstone's committed $3bn is roughly an eighth, while the shareholder side still owes something near $10bn [3][6][2][3][4].
The interesting term is not the total but who holds the chips. In build-to-suit, Amin says, the hyperscaler purchases the GPUs [10], so HUMAIN is selling power, land and buildings against contracted revenue, and the export licence sits on the tenant's books. Neutrality is a far cheaper promise for a landlord than for a lab, and Amin is candid about why he is closer to the former: Allam was trained on about 4,000 GPUs against OpenAI's hundreds of thousands, which is 2% of 200,000 [11][5], and he names the binding constraints as researchers with real depth and large compute [16]. The partnership with Reflection AI buys the first and, in his telling, supplies "the sovereignty that we need" [12].
Debt of $23bn against data centres prices off tenant credit, which pulls HUMAIN toward the American clouds whose government, on Amin's own account, has already cut off access to some models temporarily [13]. The open-weight marketplace and the Mistral announcement hedge the model layer [12][17], while the financing structure deepens the dependency at the hardware layer. The counter is straightforward, and Amin makes it himself, that predictable tariffs and one of the lowest total costs of ownership anywhere are what the buyer is actually pricing [14], and an inference customer paying per token does not audit the flag on the building.
So: leases get signed and the 70% arrives as project debt priced on hyperscaler credit; or access tightens and the Kingdom owns powered shells worth whatever regional inference will pay; or open weights close the gap far enough that the venture fund and the offices in the US, France and the UK turn out to be the business rather than the accessory [7][17]. The first is likeliest, because a $23bn third-party requirement is a requirement for tenants with ratings, not for a thesis about neutrality. What would break that read is the composition of the next tranche: third-party money arriving without an anchor lease, underwritten on Saudi power economics and merchant demand alone, would mean the Switzerland framing is load-bearing rather than decorative.
Semafor credits Amin with a well-timed $3bn investment into xAI before it was folded into SpaceX ahead of the IPO [9], and $3bn is exactly what Blackstone has committed in the other direction [6][6]. Sovereign money buys into an American model lab; American money buys into Saudi kilowatts. Which of those two flows compounds faster determines whether HUMAIN ends up controlling model access or simply renting out the power to those who do.
