Lambda

At a Glance

Four numbers, four different epistemic statuses — and the article keeps them apart rather than blending them into one story. The $9.0B is a derived secondary mark, not a round. The ~$760M is annualized, not audited. The absent S-1 is a verified fact. The 3GW is a stated ambition with a 2030 date on it. Treating the fourth like the third is how pre-IPO research goes wrong.

1 · What Lambda Actually Is

A GPU workstation company that became a power company

Stephen and Michael Balaban — twin brothers — founded Lambda in 2012 to build machine-learning products, including a face-recognition and photo-editing tool. The pivot came from their own cost line: AWS compute for training neural networks was expensive enough that building the machines themselves was cheaper. In 2017 Lambda began selling AI hardware — laptops, workstations, and servers with GPUs built in — mostly to researchers. In 2018 it began renting compute instead.

That second business ate the first. By August 2025 Lambda had shut down its on-premise hardware line entirely to focus on cloud and on building the data centers it calls AI factories. Revenue from the two segments had already crossed over in mid-2024, with cloud at roughly $250M annualized against roughly $150M for hardware. The company today describes itself as "The Superintelligence Cloud" and serves, by its own account, tens of thousands of customers spanning individual researchers, enterprises, frontier AI labs and hyperscalers.

The strategic read: Lambda is one of the few neoclouds that came to capital-intensive infrastructure from the developer side rather than from the real-estate or bitcoin-mining side. IREN and Cipher converted mining sites; CoreWeave started as an Ethereum miner. Lambda started with researchers who wanted a machine that worked. That heritage is worth something in enterprise credibility — and nothing at all against a $35B capital-expenditure budget.

2 · The Funding Ladder

$4.5M to $9B in nine years, with the last 3.6× in seven months

Lambda spent its first decade as an unremarkable hardware business and its last three as one of the fastest-marking private companies in infrastructure. The ladder below is the single most useful artifact in the story, because the slope changes exactly when NVIDIA supply became the binding constraint on AI.

The investor list is the tell. Alongside conventional venture names — 1517 Fund, Gradient Ventures, B Capital, G Squared — sit NVIDIA (supplier and customer as well as shareholder), Mubadala Capital, T. Rowe Price, ARK Invest, In-Q-Tel, and the supply chain itself: Supermicro, Pegatron, Wistron and Wiwynn. When your server ODMs are on your cap table, you have solved allocation. You have also concentrated your fate.

3 · Financials & Unit Economics

38× revenue growth in three years — funded by other people's balance sheets

Lambda's estimated annual revenue was roughly $20 million in 2022. Reported figures put it at approximately $425 million in 2024, a $505 million annualized run rate by May 2025, and more than $520 million over the trailing twelve months from October 2024 through September 2025, with Q3 2025 sales up about 80% year over year. QuantLogix's roster carries approximately $760 million annualized as of Q4 2025.

The margin picture is the part that will surprise people. Third-party estimates put Lambda's gross margin at roughly 50% blended, or about 61% excluding the non-cloud lines — respectable for infrastructure, far below software, and structurally capped by the price of the GPUs underneath. Against the $520M+ trailing revenue, Lambda reportedly carried approximately $175 million in losses. That is not a distress signal; it is what a company looks like when it buys assets today for revenue it recognises over the following three to five years.

• The capital is borrowed, not just raised. Lambda closed a $1 billion credit facility in August 2025 on top of ~$2.3–2.4B of equity — the standard neocloud pattern of financing GPUs against the contracts they serve.
• Depreciation is the whole argument. Whether a GPU is a six-year asset or a three-year one decides if this business is profitable. That schedule is an S-1 disclosure, and it is not currently public for Lambda.
• Revenue quality is unknown. Committed multi-year capacity and hourly self-serve compute are worth very different multiples. Lambda has never published the split.

The honest framing: on the public evidence, Lambda is a fast-growing, gross-margin-positive, net-loss-making infrastructure company of roughly $760M annualized scale, with a debt-financed asset base whose useful life has not been disclosed. Every one of those clauses is load-bearing.

4 · IPO Status & Timeline

Everything is ready except the filing

The preparation is unusually complete. In September 2025 Lambda hired Morgan Stanley, J.P. Morgan and Citi for a listing then expected as early as H1 2026. In January 2026 reporting moved the target to H2 2026 and disclosed the pre-IPO convertible. Through May 2026 Lambda assembled a public-company leadership bench: a CFO from Turo and Charter, a Chief Legal Officer from Altice, a Vice Chairman of Compute Delivery who was COO of Snap and an early AWS infrastructure leader — and a CEO who has run Sprint, SoftBank International and Alcatel-Lucent.

And yet: a search of the SEC's EDGAR full-text system on August 15, 2026 returns no S-1 or S-1/A from Lambda. The only hits for the name are unrelated filers referencing the company. This is a checkable fact, and it is the most important one in this brief.

That absence is not necessarily bearish. Under the JOBS Act an emerging growth company files a confidential draft registration statement that only becomes visible when publicly flipped — conventionally about 15 days before a roadshow. Lambda could be six weeks from pricing and look exactly like this. The point is that nobody outside the deal can currently distinguish "quietly ready" from "quietly delayed", and anyone who tells you otherwise is inferring, not reading.

5 · The Microsoft & NVIDIA Triangle

Your supplier is your investor is your customer

In November 2025 Lambda announced a multibillion-dollar, multi-year agreement with Microsoft to deploy AI infrastructure powered by tens of thousands of NVIDIA GPUs, including GB300 NVL72 systems. The relationship dates back to 2018. It is simultaneously the strongest validation in the story and its largest unquantified risk, because no party has disclosed the contract's value, duration, or minimum-commitment structure.

6 · Competitive Position

The neocloud cohort has already been priced — by everyone except Lambda

Lambda's peers are no longer private comparables; most of them are tickers with quarterly disclosure. That is unusually helpful, because it means the market has already answered the question Lambda's bankers must ask: what is a gigawatt of contracted AI compute worth?

Market caps and prices from QuantLogix's own equity universe, refreshed August 15, 2026. CoreWeave and Nebius quarterly figures are Q2 2026 as reported.

CoreWeave is the yardstick, and it is a demanding one. Its Q2 2026: revenue of $2.575B, up 112% year over year; adjusted EBITDA of $1.5B; a $104B revenue backlog, up 246%, excluding more than $25B of new third-quarter commitments; 1.5GW of active power heading past 1.85GW by year-end with 4.2GW contracted; full-year guidance raised to $12.4–13.2B. It is also carrying $35B of debt and posted a $626M quarterly net loss, with 2026 capital expenditure guided to $35–39B.

Read those two paragraphs together and Lambda's strategic position is legible: it is roughly one-tenth of CoreWeave's revenue scale and is targeting by 2030 a capacity number CoreWeave has already contracted today. Lambda is not competing to be the largest neocloud. It is competing to be a differentiated one — and the differentiation it can actually claim is the developer funnel, the NVIDIA relationship, and a customer base that starts with individual researchers rather than with one hyperscaler contract.

7 · Valuation Math

11.8× trailing, or 6× forward — the entire debate in one ratio

At the $9.0B June 2026 secondary mark against ~$760M of annualized revenue, Lambda is carried at roughly 11.8× trailing annualized revenue. Against CoreWeave at ~4.6× its own 2026 guidance midpoint, that looks expensive by a factor of two and a half. But the denominators are not the same kind of number, and pretending they are is the most common error in neocloud analysis.

The fourth bar is the deal. If Lambda's revenue roughly doubles off the Q4 2025 exit rate — slower than the ~80% it just printed, and far slower than CoreWeave's 112% — the $9.0B mark is about 6× forward revenue, which sits between CoreWeave and Nebius rather than above both. On that arithmetic the secondary market is not paying a silly price; it is paying a growth-contingent one.

Which means the bear case is not "the multiple is too high." It is that the growth is unverified: no audited statements, no disclosed backlog, no customer concentration figure, no GPU depreciation schedule. CoreWeave's $104B backlog is a contractual fact a reader can underwrite. Lambda's forward revenue is currently an inference from a private mark. The IPO does not primarily change Lambda's value — it changes how much of that value a public investor is allowed to verify.

8 · Key Risks

What public-market diligence will price in

Note the shape of this matrix against Stripe's: Stripe's top risk was that it might never list, because tender offers made staying private comfortable. Lambda's top risks are all disclosure risks — things that are unknown rather than things that are bad. That is a categorically different risk profile, and a more resolvable one.

9 · Pre-IPO Exposure Routes Today

Indirect vectors before the listing