Lambda

  Executive Briefing

Lambda is a San Francisco-based GPU cloud company that brands itself the "Superintelligence Cloud" — an AI infrastructure specialist focused entirely on renting NVIDIA GPU compute to researchers, AI startups, frontier labs, Fortune 500 enterprises, and even hyperscalers. Founded in 2012 by twin brothers Stephen Balaban and Michael Balaban, the company spent its first decade evolving from a facial-recognition API into a GPU hardware reseller, and ultimately into one of the most consequential neoclouds of the large-language-model era. As of mid-2026, Lambda operates tens of thousands of GPUs across liquid-cooled AI factory facilities in California, Chicago, Atlanta, and Kansas City, with a stated ambition of managing 3 GW of AI compute and roughly one million GPUs by 2030.¹

The company's trajectory accelerated dramatically between 2023 and 2026 as demand for GPU capacity overwhelmed public cloud supply. Lambda raised a reported $480 million Series D at an estimated $2.5 billion valuation in February 2025, followed by a reported $1.5 billion-plus Series E led by TWG Global in November 2025 at an estimated post-money valuation of approximately $5.9 billion — though these figures are reported estimates and not officially confirmed.²³ A $1 billion senior secured credit facility led by J.P. Morgan closed in May 2026, upsized roughly four times from an earlier $275 million facility, underscoring the company's scale of capital deployment.⁴ NVIDIA is both a strategic investor (entering in the Series D) and a major customer, having signed a reported $1.5 billion GPU lease-back agreement in September 2025 for 18,000 GPUs over four years.⁵

In May 2026, Lambda assembled a new institutional-grade executive team, signaling a deliberate shift from founder-led startup to capital-intensive infrastructure operator. Michel Combes — former CEO of Sprint, SoftBank International, Brightspeed, and Alcatel-Lucent — became CEO, while co-founder Stephen Balaban moved to CTO and Michael Balaban to CPO. John Donovan, former CEO of AT&T Communications, became Chairman of the Board. Jerry Hunter, formerly COO of Snap and an early AWS infrastructure leader, joined as Vice Chairman of Compute Delivery.⁶ The executive formation reflects a deliberate positioning of Lambda alongside the institutional capital deployment scale of major data-center operators rather than the scrappy startup model that got the company to a billion dollars in annualized revenue.

Lambda's closest direct competitor is CoreWeave (which IPO'd in March 2025 at a valuation that reached approximately $65 billion), though Lambda competes across a wider band — from individual researchers seeking affordable hourly H100 access, to multi-billion-dollar enterprise and hyperscaler agreements. Its NVIDIA Exemplar Cloud certification, zero-egress-fee policy, and ML-optimized Lambda Stack software give it a differentiated position. The company's reported gross margin was approximately 61% on cloud-only revenue in H1 2025, with reported annualized revenue of approximately $760 million — though different sources diverge on the 2025 full-year figure, and all revenue numbers should be treated as reported estimates.⁷

  At a Glance

  Origins & Founding

Lambda was founded in 2012 in San Francisco by identical twin brothers Stephen Balaban and Michael Balaban. Stephen, who studied Computer Science and Economics at the University of Michigan, had been training neural networks from that year onward — a remarkably early starting point before deep learning had achieved its ImageNet-moment cultural breakthrough. The company's initial product was the Lambda Face API, a facial-recognition service launched in direct response to Facebook's abrupt shutdown of the Face.com API in 2012, which stranded roughly 45,000 developers who had built products on top of it.⁸ Within a year, the new API had attracted more than 1,000 active developers and was handling over 5 million API calls per month.

Stephen's pedigree extended into Apple's device AI world: he was the first engineering hire at Perceptio, a mobile on-device AI startup that Apple acquired in 2015 and whose work fed into on-device neural processing on iPhone. He has also published at NeurIPS and SPIE. Michael served as CTO through early 2026, reflecting a technical co-founder split typical of deep-tech startups, before transitioning to CPO as the company brought in operational leadership.

The founding insight — that AI practitioners were being underserved by generic infrastructure — evolved through successive product forms. The facial-recognition API gave way to GPU-accelerated workstations and servers designed specifically for deep learning researchers, a market that was almost nonexistent when Lambda entered it. From there the logical extension was cloud compute, and the cloud business eventually eclipsed the hardware lines entirely. Lambda's first plug-and-play deep learning supercomputer, the Lambda Blade server, launched at under $20,000 in 2017, pricing that was genuinely disruptive for a market where comparable capability had previously cost multiples more.

  History & Timeline

    2012–2016: API and early hardware

Lambda launched as a facial-recognition API business after Facebook's shutdown of Face.com created an underserved developer community. The Lambda Face API reached 1,000+ developers and 5 million monthly API calls within a year of launch. The company received seed funding tranches totaling approximately $4 million from Gradient Ventures (Google's AI-focused fund), 1517 Fund, and Bloomberg Beta between 2015 and 2018. Lambda was formally incorporated in March 2015. As the company's founders observed deep learning researchers struggling to acquire the GPU hardware they needed, they began building workstations and servers purpose-configured for training neural networks.

    2017–2020: GPU hardware and Lambda Stack

Lambda's GPU workstation line gave the company significant reach inside research institutions — by 2022 its hardware was reportedly deployed at 97% of US universities. The Lambda Stack software, a pre-configured environment bundling NVIDIA drivers, CUDA, cuDNN, PyTorch, and TensorFlow, was adopted by more than 50,000 ML teams, and remains a key differentiator. Lambda launched its GPU Cloud in 2018, beginning the transition from hardware-first to cloud-first. A first plug-and-play deep learning server under $20,000 debuted in 2017.

    2021–2022: Cloud, capital, and the LLM inflection

A $15 million Series A in July 2021 funded the expansion of cloud infrastructure. The Lambda Tensorbook laptop — a collaboration with Razer — was released in 2022, marking an attempt to extend the brand to individual ML practitioners. Revenue reached approximately $20 million in 2022. The Series A was followed by $39.7 million in additional venture funding, also in 2022. The company's pivot toward cloud compute was now central rather than complementary.

    2023: Series B and hypergrowth

Lambda closed a $44 million Series B led by Mercato Partners in March 2023. Revenue scaled from roughly $20 million in 2022 to approximately $250 million in 2023 — a more than 12x increase driven by the sudden explosion in demand for GPU compute following ChatGPT's November 2022 launch. The company's paying customer base roughly doubled from approximately 5,000 to over 10,000 by the end of 2024.

    2024: Institutional capital and the cluster era

February 2024 brought a $320 million Series C at a reported valuation of approximately $1.5 billion, led by the US Innovative Technology Fund with participation from B Capital, SK Telecom, and T. Rowe Price.⁹ In April 2024, Macquarie Group extended a $500 million debt facility specifically for GPU purchases — an unusual arrangement illustrating how GPU procurement had become a capital-markets-scale problem. A Lambda Inference API offering serverless LLM access launched in December 2024 but was wound down by mid-2025 in favor of instance-based deployments. Revenue reached approximately $425 million for 2024, representing roughly 70% year-over-year growth.⁷

    2025: Series D, NVIDIA partnership, and the $1.5B Series E

February 2025 saw the close of a $480 million Series D at a reported valuation of approximately $2.5 billion, with NVIDIA joining as a strategic investor alongside ARK Invest, G Squared, Pegatron, Supermicro, Wistron, and Wiwynn.¹⁰ The NVIDIA investment formalized a partnership that was already operational at the infrastructure level. Lambda became the NVIDIA AI Excellence Partner of the Year in 2024.

The pace of infrastructure expansion accelerated through 2025: the first NVIDIA HGX B200-accelerated clusters were deployed in Columbus in June 2025; NVIDIA SHARP protocol integration (delivering 45–63% bandwidth improvement) was announced in July 2025; a JPMorgan $275 million credit facility was arranged in August 2025; and in September 2025, NVIDIA signed its reported $1.5 billion GPU lease-back agreement with Lambda covering 18,000 GPUs over four years.⁵

In November 2025, Lambda closed a $1.5 billion-plus Series E led by TWG Global (a $40 billion firm backed by Thomas Tull and Mark Walter), with co-investment from the US Innovative Technology Fund, NVIDIA, and ARK Invest.² The post-money valuation was estimated at approximately $5.9 billion based on secondary-market data, though Lambda did not officially disclose the figure. In the same month, Lambda announced a multibillion-dollar multi-year agreement with Microsoft for Lambda to deploy AI infrastructure powered by tens of thousands of NVIDIA GPUs, including GB300 NVL72 systems — a landmark deal that positioned Lambda as a capacity provider to hyperscalers, not merely a competitor for end-user workloads.¹¹

Also in 2025, Lambda deprecated both its Lambda Inference API and Lambda Chat (which had hosted open-source models including DeepSeek R1), and in August 2025 permanently discontinued all legacy on-premise hardware lines — the Vector workstations, Scalar/Hyperplane servers, and Tensorbook laptop — concentrating entirely on cloud infrastructure.⁸

    2026: New leadership, AI factories, and pre-IPO

At GTC 2026 in March, Lambda made several significant announcements: the launch of Bare Metal Instances providing direct hardware access without hypervisor overhead; designation as an early launch partner for NVIDIA's Vera CPU (88 cores, 1.2 TB/s memory bandwidth, targeting RL and agentic workloads); and the reveal of a 10,000-plus GPU AI factory powered by NVIDIA GB300 NVL72 systems using NVIDIA Quantum-X800 InfiniBand Photonics Co-Packaged Optics switches — one of the largest such deployments of that switching fabric globally.¹

In May 2026, Lambda assembled its new institutional C-suite: Michel Combes as CEO, John Donovan as Chairman, Jerry Hunter as Vice Chairman of Compute Delivery, Charles Fisher (ex-Turo) as CFO, and David Connolly (ex-Altice) as Chief Legal Officer.⁶ The same month, a $1 billion senior secured credit facility led by J.P. Morgan closed, oversubscribed and upsized approximately four times from the August 2025 facility.⁴

Lambda is reported to be in pre-IPO financing discussions, targeting approximately $350 million in convertible notes led by Mubadala Capital at an approximately 20% discount to anticipated IPO price, with an IPO targeted for H2 2026 — though neither the convertible round nor the IPO timeline has been formally confirmed as of June 2026.¹²

  What They Offer — Products & Platform

Lambda organizes its commercial offering into three main compute tiers, all underpinned by zero egress fees and the pre-configured Lambda Stack environment.

On-Demand Cloud is Lambda's core hourly-billed product, offering single- and multi-GPU instances across the full current NVIDIA GPU lineup — RTX 6000 Ada, V100, A10, A6000, A100 PCIe and SXM in both 40 GB and 80 GB configurations, GH200, H100 SXM, and B200 SXM. Billing is at the minute level, and instances launch pre-configured with Lambda Stack (PyTorch, TensorFlow, CUDA, cuDNN, and current NVIDIA drivers). This tier is designed for researchers, developers, and startups who need immediate GPU access without long-term commitments.

1-Click Clusters are production-ready InfiniBand-connected GPU clusters ranging from 16 to 2,000-plus GPUs in H100 or B200 configurations, provisioned within minutes. Clusters ship with a choice of orchestration framework — Kubernetes, Slurm, or dstack — and are intended for model training, large-scale fine-tuning, and distributed inference workloads. Reserved pricing for one-year commitments is available.

Private Cloud offers dedicated, air-gapped clusters for enterprises with heightened security and compliance requirements. Private Cloud deployments carry SOC 2 Type II certification and support thousands of GPUs in isolated configurations serving regulated industries including finance, healthcare, and government.

Bare Metal Instances, launched at GTC in March 2026, eliminate the hypervisor layer to provide direct hardware access at the physical server level. This offering targets frontier model training workflows where hypervisor overhead creates meaningful performance drag at the largest scales.

Lambda Stack — the pre-configured ML software environment originally built for workstations — remains freely available and is used by more than 50,000 ML teams. Adopted across 97% of US university research programs at peak hardware distribution, it has become a standard environment for AI research.

Professional services are offered to enterprise customers, covering AI roadmaps, proofs of concept, workflow design, and model optimization.

The Lambda Inference API (serverless LLM access, December 2024 – mid-2025) and Lambda Chat (open-source model hosting including DeepSeek R1) have both been deprecated as Lambda focuses on compute infrastructure rather than managed model serving.

  Technology & Infrastructure

Lambda's infrastructure strategy is built around density, networking performance, and direct NVIDIA hardware partnership rather than general-purpose cloud architecture.

    GPU Fleet

Lambda operates across the full current NVIDIA compute stack: H100 SXM and PCIe (80 GB), H200, B200 SXM, GH200 (96 GB), A100 SXM (40 GB and 80 GB), A100 PCIe (40 GB), A6000, A10, V100, and RTX 6000 Ada. Lambda is a launch partner for both the NVIDIA GB300 NVL72 and Vera Rubin NVL72 Superclusters, with H2 2026 availability, and for the NVIDIA Vera CPU (88-core, 1.2 TB/s memory bandwidth, 1.8 TB/s CPU-to-GPU connectivity) targeting reinforcement learning and agentic workloads.¹ Lambda's AI factory in Kansas City houses 10,000-plus NVIDIA Blackwell Ultra GPUs and came online in early 2026. The press figure of approximately 25,000 GPUs total (cited November 2025) is likely materially understated by mid-2026 given ongoing deployments.

Lambda holds NVIDIA Exemplar Cloud status, certifying that GPU performance is within 5% of NVIDIA's own published baselines — a designation that requires passing structured performance validation and distinguishes Lambda from commodity GPU cloud providers where hardware configurations may deviate significantly from reference.

    Networking

Cluster connectivity uses NVIDIA Quantum-2 InfiniBand at the majority of deployments and NVIDIA Quantum-X800 InfiniBand with Photonics Co-Packaged Optics at the AI factory scale. Lambda integrated NVIDIA SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) in July 2025, delivering reported bandwidth improvements of 45–63% for collective communication workloads typical in distributed training.¹³ DPU-based fabric management and out-of-band telemetry complement the networking layer.

    Data Centers

Lambda operates a colocation model across purpose-built AI factory sites:

• LAX01 — 21 MW at Prime Data Centers' campus in Vernon, California
• Chicago — 23 MW single-tenant facility via EdgeConneX
• Atlanta — EdgeConneX facility
• Columbus — first NVIDIA HGX B200 cluster deployments (June 2025)
• Kansas City — repurposed 2009 data center, 24 MW initial capacity scalable to 100-plus MW, housing 10,000-plus NVIDIA Blackwell Ultra GPUs (online early 2026)

All facilities are liquid-cooled. Lambda is an NVIDIA Platinum Sponsor partner and participates in the NVIDIA Fleet Intelligence Early Access Program for infrastructure telemetry and optimization.

    Software Stack

Lambda Stack provides a unified environment across all instance types: NVIDIA drivers, CUDA, cuDNN, PyTorch, TensorFlow, and supporting libraries, maintained and updated by Lambda. Cluster orchestration is available via Kubernetes, Slurm, and dstack. Bare Metal Instances expose the physical layer directly for workloads that cannot tolerate hypervisor overhead.

  Pricing & Performance Position

Lambda's on-demand pricing positions it well below AWS, Azure, and Google Cloud on equivalent GPU hardware, and comparably or slightly below CoreWeave at most configurations.

On-demand single-GPU pricing (as of June 2026, approximate):

Competitive comparison (H100 SXM on-demand, approximate mid-2025):

Lambda's H100 on-demand pricing sits between CoreWeave (which is lower due in part to reserved-capacity contracts) and the major hyperscalers. The more meaningful comparison for enterprise customers is total cost of ownership: Lambda charges zero egress fees where AWS, Azure, and GCP impose per-GB egress charges that can be substantial at training and inference scales. Lambda also bundles the Lambda Stack environment, eliminating the CUDA/driver configuration overhead that consumes material engineering time on generic clouds.

Reserved pricing (one-year commitment): approximately $1.85/GPU-hr for H100 clusters and approximately $3.49/GPU-hr for B200 clusters. Reported cloud gross margin was approximately 61% in H1 2025, and approximately 50% overall including the now-discontinued hardware lines.⁷

  People & Leadership

The leadership transition completed in May 2026 was deliberate in character: co-founder Stephen Balaban described it publicly as assembling a team to "power gigawatt-scale AI infrastructure" — an explicit acknowledgment that the next phase of Lambda's growth requires institutional operating experience rather than the startup DNA that got it to a billion-dollar revenue run rate.⁶

Michel Combes (CEO, effective May 2026) brings a career arc through the capital-intensive infrastructure industries that AI compute most resembles. He served as CEO of Alcatel-Lucent (restructuring the company from crisis before its Nokia acquisition), CEO of Sprint, CEO of SoftBank International, and most recently Chairman and CEO of Brightspeed, a fiber broadband buildout company. His background managing large-scale network rollouts and balance-sheet-heavy infrastructure companies maps directly to Lambda's GPU factory ambitions.

John Donovan (Chairman) spent the bulk of his career at AT&T, ultimately as CEO of AT&T Communications and as CTO and Group President of Technology and Operations. He brings both board governance experience and deep familiarity with operating infrastructure at national scale.

Jerry Hunter (Vice Chairman of Compute Delivery) was COO of Snap and one of the early infrastructure leaders at AWS — experience combining hyperscale cloud operations with delivery execution.

Stephen Balaban (Co-founder, CTO) remains central to Lambda's technical direction and hardware partnerships. His move from CEO to CTO follows the pattern of founders who remain the technical authority while operational leadership scales alongside capital deployment. His history publishing at NeurIPS and SPIE and his early career at Perceptio (acquired by Apple) grounds the company's ML-infrastructure credibility.

Michael Balaban (Co-founder, CPO) moved from CTO to CPO in the May 2026 reorganization, focusing on product strategy across the cloud platform.

Charles Fisher (CFO) joins from Turo (the peer-to-peer car-rental marketplace) and Charter Communications, bringing structured-finance and public-company-readiness experience ahead of the expected IPO.

David Connolly (Chief Legal Officer) joins from Altice, where he served as General Counsel through significant corporate transactions.

  Funding, Ownership & Business

Lambda has raised a total of approximately $2.36 billion in equity since founding, progressing through a classic seed-to-growth arc before entering the large-scale infrastructure debt markets characteristic of data-center companies.

Funding history:

Lambda's business model is GPU-hours: on-demand pricing and multi-year reserved contracts generate the bulk of revenue, with enterprise private-cloud contracts and professional services as smaller contributors. The revenue trajectory — approximately $20 million in 2022, $250 million in 2023, $425 million in 2024, and approximately $760 million annualized in 2025 (per Sacra; other sources diverge) — reflects the GPU-demand wave driven by LLM training and inference at scale.⁷ All figures are reported estimates.

A distinctive aspect of Lambda's capital structure is the use of large-scale debt facilities for GPU procurement rather than equity alone. The Macquarie $500 million GPU-purchase facility (April 2024) and the J.P. Morgan-led $1 billion senior secured facility (May 2026) reflect a data-center-operator capital logic: long-term assets (GPUs under reserved contract) can support debt financing, a model Lambda has adopted ahead of most GPU-cloud peers.

The NVIDIA $1.5 billion GPU lease-back arrangement further illustrates the unusual capital dynamics: NVIDIA leasing GPU capacity from Lambda — the company it has also invested in — reflects both Lambda's position as the most capable large-scale deployment partner in NVIDIA's ecosystem and the GPU maker's interest in demonstrating real-world deployment of its newest hardware.⁵

  Customers & Partnerships

Lambda's customer base exceeds 10,000 paying accounts as of 2024, up from roughly 5,000 in 2023, with more than 200,000 AI developer sign-ups.⁸ The customer mix spans individual researchers to the largest technology companies and government agencies.

Named customers and segments include:

• Apple, Microsoft, Tencent — large enterprise and hyperscaler customers
• Department of Defense — government/defense workloads
• MIT, Stanford, Harvard, Caltech, Kaiser Permanente — academic and healthcare research
• AI frontier labs, AI startups, and Fortune 500 enterprises across finance, healthcare, and government

The most significant partnership announced to date is the Microsoft multibillion-dollar multi-year agreement (November 2025) under which Lambda deploys AI infrastructure powered by tens of thousands of NVIDIA GPUs, including GB300 NVL72 systems, on Microsoft's behalf.¹¹ This positions Lambda not merely as a self-serve GPU marketplace but as a white-label infrastructure layer for the world's largest technology companies — a fundamentally different and higher-margin relationship than on-demand access.

NVIDIA occupies multiple roles simultaneously: hardware supplier (Lambda purchases at scale), strategic equity investor (Series D onward), capacity buyer (the $1.5 billion lease-back), and certification authority (Exemplar Cloud status, launch partner for GB300 NVL72, Vera CPU, and Vera Rubin NVL72). This depth of entanglement with NVIDIA is arguably Lambda's single most important structural advantage.

Additional partnerships include Supermicro (AI factory server clusters using Blackwell hardware, announced August 2025), and access to the NVIDIA Fleet Intelligence Early Access Program for infrastructure telemetry.

  Competitive Position

Lambda occupies a carefully defined niche between commodity GPU marketplaces and hyperscaler general-purpose clouds. Its closest direct competitor is CoreWeave, which IPO'd in March 2025 and reached a market capitalization in the tens of billions of dollars. CoreWeave has historically priced H100 on-demand slightly lower than Lambda (approximately $2.99–$3.19/hr versus Lambda's $3.99–$4.29/hr as of mid-2025), but Lambda counters with zero egress fees, the Lambda Stack pre-configuration, startup-accessible on-demand availability, and direct NVIDIA partnership depth.

Against marketplace-model competitors like RunPod (approximately $2.39/hr for H100, peer/marketplace model) and Vast.ai, Lambda differentiates with enterprise-grade SLAs, SOC 2 Type II certification, dedicated multi-thousand-GPU clusters with InfiniBand fabric, and the institutional customer relationships those features enable.

Against the hyperscalers — AWS, Azure, and Google Cloud — Lambda holds meaningful price advantages (30–62% below comparable configurations, per its own comparisons), zero egress fees, and provisioning speed (minutes versus hours). The hyperscalers' advantages are ecosystem depth, global region coverage, and the integration with broader cloud services; Lambda's counter is ML-specific optimization and the ability to provision dense GPU clusters rapidly without the enterprise sales cycles hyperscalers require.

Lambda's NVIDIA Exemplar Cloud certification is a significant moat: it requires passing NVIDIA's structured performance benchmarks and provides a verified performance guarantee that neither marketplaces nor most neoclouds can match. The combination of that certification, the NVIDIA equity investment, and the GB300/Vera/Vera Rubin launch-partner status gives Lambda priority access to the next generation of GPU supply — which has historically been the binding constraint in the GPU cloud market.

Reported GPU availability challenges in 2024 — a documented approximately 64% same-day success rate for A100 provisioning, and user complaints about H100 on-demand scarcity — represent a known weakness that the ongoing data-center buildout is designed to address.⁸

  Notable Events

November 2025 — Series E and Microsoft deal: The $1.5 billion-plus Series E led by TWG Global and the concurrent announcement of the multibillion-dollar Microsoft infrastructure agreement together established Lambda at a qualitatively new scale of capitalization and customer relationship.²¹¹

September 2025 — NVIDIA lease-back: NVIDIA's reported $1.5 billion agreement to lease GPU capacity from Lambda — a company NVIDIA has also invested in — is structurally unusual and signals the depth of the two companies' interdependence.⁵

August 2025 — Hardware exit: Lambda permanently discontinued all on-premise hardware product lines (Vector workstations, Scalar and Hyperplane servers, Tensorbook laptop), completing the pivot from hardware company to cloud-only infrastructure operator.

March 2026 — GTC 2026 announcements: Lambda unveiled Bare Metal Instances, early Vera CPU launch-partner status, the Kansas City 10,000-plus GPU Blackwell Ultra AI factory, and one of the world's largest deployments of NVIDIA Quantum-X800 InfiniBand Photonics Co-Packaged Optics switching.¹

May 2026 — C-suite formation and $1B credit facility: The simultaneous appointment of Michel Combes as CEO and the close of the J.P. Morgan-led $1 billion credit facility marked the transition to institutional-scale infrastructure operator.⁴⁶

March 25–26, 2025 — us-south-1 outage: A UPS maintenance event at the southern US data center cascaded into approximately 16 hours of networking and cooling disruption, highlighting the operational risks inherent in running dense GPU clusters at scale.

2024 — NVIDIA AI Excellence Partner of the Year: Lambda received NVIDIA's flagship partner award, formalizing its position as the preferred third-party cloud for NVIDIA GPU deployments.

  Outlook & Roadmap

Lambda's stated target is 3 GW of AI compute under management and approximately one million GPUs by 2030 — a scale that would place it among the largest dedicated AI compute operators globally.¹ The roadmap toward that target has several near-term signposts.

The Kansas City AI factory (24 MW initial, scalable to 100-plus MW, 10,000-plus NVIDIA Blackwell Ultra GPUs) is operational as of early 2026, adding meaningful capacity beyond the approximately 25,000 GPUs disclosed in late 2025 press materials. NVIDIA GB300 NVL72 and Vera Rubin NVL72 Superclusters are planned for H2 2026 availability, extending the GPU fleet to the next hardware generation. NVIDIA STX-based platform instances are also in the roadmap.

The assembly of the new executive team with telecom and large-scale infrastructure pedigree — Michel Combes, John Donovan, Jerry Hunter — is the clearest signal of Lambda's strategic direction: the company is positioning for the same capital-intensity profile as major data-center REITs and network operators, where asset buildout is financed through a combination of long-term contracted revenues and structured debt, rather than the equity-heavy startup model.

The Microsoft agreement and the NVIDIA lease-back demonstrate a capacity-provider model that is distinct from the end-user GPU marketplace Lambda began with: at sufficient scale, Lambda can be the infrastructure that hyperscalers and chip makers themselves deploy on. This dual role — serving AI researchers and startups through self-serve access while operating dedicated capacity for the largest technology companies — offers Lambda a diversified revenue base that neither pure marketplace competitors nor individual hyperscaler AI divisions can easily replicate.

A pre-IPO convertible note round of approximately $350 million led by Mubadala Capital (at a reported ~20% discount to anticipated IPO price) is reported to be in process as of mid-2026, with an IPO targeted for H2 2026. Neither the convertible round nor the IPO timeline has been formally announced; both should be treated as unconfirmed pending official disclosure.¹²

---

  References

1. Lambda at GTC 2026 — Building the Superintelligence Cloud
2. Lambda Raises $1.5B+ Series E led by TWG Global — SuperbCrew
3. Lambda IPO news and valuation — Forge Global
4. Lambda closes $1 billion senior secured credit facility — Lambda blog
5. Lambda and NVIDIA lease-back — Data Center Dynamics
6. Lambda assembles leadership team — Lambda blog
7. Lambda company research — Sacra
8. Lambda company profile — Contrary Research
9. Lambda Series C — various sources
10. Lambda Series D — NVIDIA investment
11. Lambda Microsoft multibillion-dollar agreement — Lambda blog
12. Lambda pre-IPO convertible note round — Data Center Dynamics
13. NVIDIA SHARP protocol integration — Lambda announcement

  References

1. Lambda at GTC 2026 — lambda.ai/blog. ↩ ↩² ↩³ ↩⁴ ↩⁵

2. Series E and TWG Global — SuperbCrew; post-money valuation of ~$5.9B is an estimate from secondary-market data. ↩ ↩² ↩³

3. IPO and valuation data — Forge Global; treat as reported/estimated. ↩ ↩²

4. $1B credit facility — Lambda blog. ↩ ↩² ↩³

5. NVIDIA lease-back — Data Center Dynamics; amount and GPU count reported, not officially confirmed by Lambda. ↩ ↩² ↩³ ↩⁴

6. Leadership team — Lambda blog. ↩ ↩² ↩³ ↩⁴

7. Revenue and margin — Sacra; all revenue figures are reported estimates. Sacra's ~$760M annualized 2025 figure diverges from other sources; treat accordingly. ↩ ↩² ↩³ ↩⁴

8. Company history and product details — Contrary Research. ↩ ↩² ↩³ ↩⁴

9. Series C details — reported across multiple press sources. ↩

10. Series D details — reported across multiple press sources; NVIDIA co-investment confirmed via Lambda announcements. ↩

11. Microsoft agreement — Lambda blog. ↩ ↩² ↩³

12. Pre-IPO round — Data Center Dynamics; unconfirmed as of June 2026. ↩ ↩²

13. NVIDIA SHARP integration — reported in Lambda announcements July 2025. ↩