Lambda weighs $3B at $12B-plus, more than every prior round it has disclosed combined

Lambda's disclosed funding history before this comes to roughly $2.3 billion: a $320 million Series C in February 2024, a $480 million Series D later that year, and more than $1.5 billion after that ^[7]. The proposed round on its own would be larger than all of it put together ^[14]. It would also be about 1.2 times the entire company as Forbes valued it in September 2025 ^[17]. That ratio, not the headline valuation, is the thing worth staring at: the cheque needed to keep buying capacity has outgrown the equity base it is being bought against.

The 4.8x step-up ^[13] is a comparison of unlike quantities. Forbes was estimating from outside, on a business with about 15,000 paying customers ^[4]. Bloomberg is reporting a price still under negotiation, with no lead investor named ^[1][3]. The publisher says as much: the timing and terms of the two valuations differ ^[5]. So the multiple describes the gap between an outside guess and a term sheet, and the term sheet is what has to be paid for.

What it buys, on Lambda's own account, is 3 gigawatts of AI compute under management by 2030 ^[10]. That is a target, and Lambda has not put an independently verified figure for capacity under management beside it ^[10]. Between a reported round and a 2027 listing, the number that converts the target into a valuation input is contracted, delivered capacity, and it is the one number absent from the record.

The hiring tells you which problem the founders think is binding. On May 5, 2026, Stephen Balaban moved from chief executive to CTO, Michel Combes became CEO, and John Donovan took the chair ^[8]. Combes ran Sprint, Alcatel-Lucent and SoftBank Group International; Donovan ran AT&T Communications ^[9]. Those are careers spent financing construction against long-lived assets, not selling compute hours. The source frames the CEO's job in the same terms: financing construction, expanding financial controls, and convincing investors that GPU fleets earn a durable return after depreciation, replacement and borrowing costs ^[12].

There is a distance here worth measuring. Lambda started because two engineers found AWS too expensive, worked out that $60,000 of owned equipment could displace most of the recurring bill, and cut that bill by 95% ^[16]. The company now describes itself as serving AI developers, enterprises, hyperscalers and frontier-model labs, at scales from a single GPU to hundreds of thousands ^[15]. The arbitrage that started it was between owning hardware and renting it. At 3 gigawatts, the arbitrage is between the cost of capital and the useful life of a GPU, which is a telecom problem wearing a compute label, and it is why the telecom operators are now running the company ^[9][10].