Lambda Seeks $12B on 8x Revenue Before Its IPO • The Weighted Average

Lambda, the Nvidia-backed GPU cloud, is in talks to raise as much as $3 billion at a valuation of $12 billion or more, with multiple term sheets already received and an IPO possible as soon as next year. Bloomberg’s report on the pre-IPO round adds the number that matters: the company is expected to earn more than $1.5 billion in revenue this year. That prices the business at roughly 8.0 times current-year revenue — a multiple worth carrying into your next capacity negotiation, because it tells you what the seller believes about future scarcity.

Eight times revenue on a business that was a tenth this size

Set the trajectory against Lambda’s own record. A GuruFocus account of Lambda’s January pre-IPO financing reports revenue above $520 million for the fiscal year ending September 2025, third-quarter sales up 80% year over year, and a $175 million loss. Divide the expected $1.5 billion by that $520 million and you get 2.9× growth in roughly a year. The same report describes Mubadala Capital in talks to lead a $350 million convertible round at about a 20% discount to the eventual IPO price, with penalties if no listing arrives within a year — a term sheet written by investors who wanted downside protection at a much lower price.

Stitch the two disclosures together and the repricing is stark. That January round was reported around a $5.9 billion post-money mark; today’s discussion starts at $12 billion. Call it a doubling in about eight months, against revenue that roughly tripled. Growth is carrying the mark, not multiple expansion — which is the more defensible version of the story, and also the more fragile one, because it means the valuation is a bet on the next tripling.

There is a second way to read the same figures, and it is less flattering. A neocloud’s revenue is rented depreciation: the fleet that produces this year’s $1.5 billion has a useful life measured in a handful of years, and refreshing it consumes the round. Compare that with a software business at 8× revenue, where the cost of serving the next dollar approaches zero. Lambda’s multiple is closer to an infrastructure operator’s than a platform’s, and infrastructure operators are valued on contract duration and utilization, neither of which is public here.

The capital stack behind it is already unusual. Lambda has priced a $926 million senior secured term loan B at SOFR plus 3% with a Moody’s Baa2 rating and December 2030 maturity, secured on hardware and contracted cash flows, and it raised over $1.5 billion in Series E led by TWG Global in November 2025. Equity, investment-grade debt and now a pre-IPO round: three financing channels for the same GPUs, which is what it takes to keep a neocloud’s fleet current.

What the multiple means when you are the buyer

For anyone renting compute, an 8× revenue mark is a forecast about your future bills. Lambda’s published on-demand rates list B200 SXM6 at $6.69 per GPU-hour and H100 SXM at $3.99, with 1-Click Cluster B200 pricing from $8.87 to $9.86 per GPU-hour depending on size and term. Note the inversion: committed cluster capacity costs more per GPU-hour than first-come on-demand instances, which is what a supply-constrained market looks like from the customer side. Reserved pricing is quoted privately, and that is where the leverage sits.

The counterpoint is the comparable set, and it is not reassuring. Bloomberg’s report notes CoreWeave’s shares are up 20% this year but shed nearly a fifth of their value in the past week, while Nebius remains up about 150% after falling 26% from a June peak. Neocloud equity is volatile even when the demand story is intact, and Lambda’s own loss history — $175 million on $520 million of revenue — shows the model consumes capital while it grows. A listing window that closes would leave the convertible investors’ penalty clauses, not the growth rate, setting the terms.

The practical read for operators is narrow and useful. First, a company raising $3 billion to expand capacity is telling you it expects demand to exceed supply through the round’s deployment horizon; do not plan your 2027 inference budget on falling GPU rates. Second, ask any neocloud for term pricing in writing before its listing, not after — the same dynamic our analysis of CME’s compute futures listing described, where a public price for GPU-months arrives just as private capacity gets scarcer. Third, treat single-vendor capacity concentration as the risk it is: a lender-financed fleet tied to contracted offtakers is efficient right up to the moment your workload is not the contracted one. Today’s lead on Hugging Face fielding $13 billion bids is the same trade one layer up, and our earlier read on SpaceX’s capital-heavy neocloud economics shows how quickly the arithmetic turns when capex outruns revenue.

There is also a structural warning in the timing. Bloomberg counts a queue: Nscale seeking as much as $3 billion in an IPO that could come within weeks, Crusoe in talks for $3 billion at potentially triple its valuation, Blackstone’s digital infrastructure vehicle raising $2 billion in May, Brookfield-backed Csquare $1.21 billion in July, and Switch filed confidentially for a listing as soon as November. When five infrastructure issuers reach the market inside two quarters, the marginal buyer of the fifth is not the enthusiast who bought the first. Pricing power in compute is being converted into public equity at speed, and the conversion rate is set by whoever is still buying in the autumn.

What would change the verdict? A signed round at a materially lower valuation, or an IPO filing that discloses gross margins and contract duration. Until the S-1 lands, the 8× multiple is the most informative number Lambda has published — and it was published by other people.