The IPO-bound GPU-cloud start-up adds one of the largest US quantitative trading firms to a customer list that already includes Microsoft and NVIDIA itself.
Lambda has signed a cloud-infrastructure deal with Hudson River Trading, one of the largest US quantitative trading firms, to supply HRT with access to NVIDIA chips, Reuters reported on Wednesday.
The agreement extends Lambda’s customer list across the high-frequency-trading category at a moment the GPU-cloud start-up is preparing for an IPO in the first half of 2026.
The deal puts HRT on a Lambda customer roster that already includes Microsoft and NVIDIA itself. Lambda announced its multibillion-dollar Microsoft agreement in November 2025, covering tens of thousands of NVIDIA GPUs, including the GB300 NVL72 systems; NVIDIA separately signed a $1.5bn four-year deal to lease back roughly 18,000 of its own GPUs from Lambda, an arrangement that made NVIDIA the start-up’s largest single customer.
Layering an HRT contract on top of that base extends Lambda’s exposure into the latency-sensitive segment of the GPU market, where customer pricing typically clears at a premium to model-training workloads.
HRT’s appetite for high-end NVIDIA capacity is not new. The firm has been running an NVIDIA-powered AI factory for its algorithmic-trading research, built on the Blackwell architecture and Spectrum-X networking, and has separately had a Google Cloud arrangement covering trading-simulation workloads since 2024.
Adding Lambda as a third compute provider points to a multi-vendor procurement posture, calibrated to keep HRT’s research workloads from being capacity-bottlenecked at any single hyperscaler during the current GPU-allocation crunch.
The commercial logic on HRT’s side fits its scale. The firm posted a record $6.4bn of quarterly trading revenue in the first quarter of 2026, according to Bloomberg, on a full-year 2025 trading-revenue figure of $12.3bn. A multi-million-dollar GPU-procurement commitment is a structurally small line item against that revenue base; what matters strategically is whether the capacity is available when the research team needs it.
Lambda has been positioning explicitly against AWS, Azure, and Google Cloud on time-to-capacity rather than price, with shorter procurement cycles and dedicated allocation as the differentiators.
For Lambda, the win lands at the cleanest possible moment in its IPO narrative. The company has hired Morgan Stanley, JP Morgan, and Citi for an offering planned for the first half of 2026, according to multiple secondary-market trackers.
An HRT logo on the customer slide, alongside Microsoft and NVIDIA, gives Lambda a verticals story that extends beyond the model-lab and hyperscaler-extension categories where it has historically concentrated.
Whether the deal materially moves Lambda’s bookings line ahead of the IPO will depend on the contract structure, which the Reuters wire describes as a multi-year supply arrangement without disclosing the headline value.
The wider GPU-allocation context is the part that the deal sits inside. Google’s $25bn TPU-cloud joint venture with Blackstone, announced last week, and the Tenstorrent takeover conversations with Intel and Qualcomm are both readings of the same structural shortage.
The current capacity-allocation problem has produced a procurement environment in which sophisticated commercial buyers, including HRT, are diversifying across three or more compute providers rather than concentrating on one.
The export-control overlay on the same trade, including the Trump-Xi licensing track on H200 sales, is keeping the supply-side picture volatile through the second half of the year.
Lambda did not disclose the contract value, the specific NVIDIA chip generations covered (Blackwell, GB300 or earlier H-series), the geographic split of the supplied capacity, or the contract duration beyond the multi-year framing. HRT has not publicly commented on the announcement.
The next visible proof point will be Lambda’s S-1 filing, expected ahead of the planned first-half IPO, where the customer-concentration disclosure will make it possible to read how material the HRT line is against the Microsoft, NVIDIA, and earlier model-lab books.
