Cirkadis Underwrites GPU Residuals to Unlock Bank Leases for Neoclouds

Cradle to Capital: Sustainable GPU Financing Through Lifecycle Management With Cirkadis — Ai4

Ai4

Lifecycle partners absorb residual hardware risk off lender balance sheets, allowing asset-light Neoclouds to finance multi-million-dollar GPU clusters through fair market value leases.

AI-assisted summary of Cradle to Capital: Sustainable GPU Financing Through Lifecycle Management With Cirkadis — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published by Ai4 on YouTube on August 18, 2026, this presentation features Ben Cartwright, CEO of Cirkadis. It explores how escalating AI and GPU hardware costs challenge traditional financing for Neoclouds, detailing how Cirkadis's lifecycle management solutions mitigate lender risk and free up capital for infrastructure build-outs.

## TL;DR The presentation examines the operational and financial hurdles of procuring massive AI GPU clusters, contrasting legacy IT Asset Disposition (ITAD) with proactive hardware lifecycle management. While standard ITAD focuses on late-stage data destruction and environmental liabilities, high-value AI servers require multi-year planning and residual value underwriting. Cartwright explains how Neoclouds leverage Fair Market Value (FMV) leases backed by customer offtake contracts, using lifecycle partners to absorb secondary market depreciation risk off commercial lenders' balance sheets. This integrated strategy enables continuous capital recycling, protects against mid-lease disruptions during industry consolidation, and promotes hardware circularity across global supply chains.

## Atmosphere & style The presentation carries a pragmatic, industry-savvy tone that unpacks high-stakes infrastructure finance with clarity and operational realism. Cartwright speaks with the candid authority of an insider, demystifying nine-figure GPU deals and balance sheet bottlenecks through clear analogies and commercial logic. The session feels focused, grounded, and intimate, punctuated by an easygoing demeanor that culminates in a lighthearted, self-deprecating joke about the small audience size in the room.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) Decommissioning AI hardware presents dual data and environmental liabilities, with AI servers carrying unique risks like proprietary cluster topologies embedded directly across hardware components. - [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) Specialized ITAD services offer secure, certified data sanitization alternatives to prevent operators from wastefully shredding multi-hundred-thousand-dollar AI hardware assets. - [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) Lifecycle management works upstream from traditional end-of-life ITAD by using secondary market pricing intelligence to optimize refresh timing and maximize value recovery. - [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) Compressed AI procurement timelines and long hardware lead times require infrastructure operators to plan decommissioning and replacement cycles one to two years in advance. - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) Emerging Neocloud startups face financing bottlenecks because traditional commercial lenders hesitate to underwrite massive GPU equipment loans for companies lacking extensive credit history. - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) Fair Market Value (FMV) leases and Special Purpose Vehicles (SPVs) allow Neoclouds to finance multi-million-dollar clusters by servicing lease payments directly through customer offtake revenue. - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) Hardware lifecycle partners unlock debt financing by underwriting the future residual value of AI clusters, giving banks the certainty needed to price and structure leases. - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) Secondary GPU market valuations are subject to shifting variables including macroeconomic conditions, sudden OEM price cuts, workload evolutions, and international export regulations. - [11:05](https://www.youtube.com/watch?v=wOQMxS-VymU&t=665s) Commercial lenders reach internal asset concentration limits on massive AI deals, requiring lifecycle partners to purchase residual value risk off balance sheets to free up lending capacity. - [11:05](https://www.youtube.com/watch?v=wOQMxS-VymU&t=665s) Data center consolidation and M&A require turnkey lifecycle support to physically extract, sanitize, recertify, and remarket displaced AI clusters mid-lease. - [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) Hardware circularity extends beyond high-end GPUs to finding secondary market utility for legacy components across global supply chains in sync with technology refresh cadences.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) The Dual Liabilities of IT Hardware Decommissioning

When an organization decommissions, retires, or reaches end-of-life on IT hardware, it confronts two primary liabilities: data liability and environmental liability. Data liability centers on securely destroying confidential information so that it never falls into unauthorized hands. Traditionally, this is addressed by the IT Asset Disposition (ITAD) industry through physical destruction (such as industrial shredders) or serialized, certified software wiping. While personal identifiable information (PII)—like patient data in hospitals—is a standard concern, AI GPU servers introduce a distinct risk profile. These machines frequently store proprietary network topologies and complex internal cluster configurations embedded across hardware components, not merely on central hard drives.

```mermaid flowchart TD accTitle: IT Asset Disposition (ITAD) Core Liabilities subgraph Decom["Decommissioning & End-of-Life"] A["Retired Hardware"] end

subgraph Liabilities["Core Liabilities"] B["Data Liability"] C["Environmental Liability"] end

subgraph Solutions["ITAD Resolutions"] D["Software Wipes & Certified Shredding"] E["Refurbishment, Demanufacturing & Reselling"] end

A --> B A --> C B --> D C --> E ```

Environmental liability dictates that equipment cannot simply be dumped into landfills; it must be sustainably repurposed, disassembled to component levels, or sent to certified smelters. However, some large enterprise operators currently elect to throw $500,000 AI hardware units directly into shredders solely to guarantee complete data destruction. While an organization has the prerogative to destroy assets it owns outright, a specialized ITAD provider offers far more efficient, value-retaining, and sustainable methods to sanitize data while preserving reusable equipment.

- Hardware retirement triggers include standard refresh cycles, corporate bankruptcies, and mergers and acquisitions. - At present, high secondary market demand ensures that the vast majority of decommissioned AI hardware is refurbished, remarketed, and redeployed rather than recycled for scrap.

---

[03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) Evolving from End-of-Life ITAD to Lifecycle Strategy

A standard ITAD provider only touches hardware at the very end of its operational life. In contrast, an IT hardware lifecycle partner works backward up the timeline into the middle of the asset's lifespan. By monitoring market intelligence, a lifecycle partner helps operators strategize optimal refresh cycles—weighing space constraints inside data centers, secondary market pricing swings, and hardware recertification options. Secondary market valuations for identical equipment can fluctuate wildly depending on channel access, with quotes varying from $100,000 to over $300,000 for the same asset.

AI infrastructure operates under compressed procurement and upgrade timelines compared to legacy enterprise IT. Traditional servers could handle routine compute loads for extended multi-year periods, but AI operators frequently sell compute capacity through offtake contracts tied directly to specific chip architectures (such as NVIDIA H200s or GB300s). Because lead times for new AI hardware often range from six months to a full year, organizations must actively plan their decommissioning and replacement cycles one to two years ahead of time.

- Lifecycle management aligns sales and business development strategies directly with hardware depreciation and procurement timelines. - Determining when to repurpose servers internally versus when to sell them onto the open market maximizes asset value recovery.

---

[06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) Capital Demands, Neocloud Financing, and Fair Market Value Leases

Procuring AI hardware requires unprecedented capital expenditures, routinely scaling across $50 million, $100 million, $500 million, or even $1 billion tranches. This has introduced a structural challenge for emerging "Neocloud" providers. Unlike traditional colocation clients that spent decades building up balance sheets to fund infrastructure, a Neocloud startup might secure a high-value customer offtake contract—for instance, generating $5 million a month over three years—yet have virtually zero credit history when approaching commercial banks for a $500 million equipment loan.

```mermaid flowchart LR accTitle: Neocloud Hardware Lease Structure subgraph Contract["Revenue"] A["Offtake Customer"] -->|"Pays $5M/month"| B["Neocloud"] end

subgraph Lease["Financing Structure"] C["Lender / Leasing Co."] -->|"Buys $500M GPUs"| D["Hardware Vendor"] C -->|"Leases Hardware ($2M-$3M/mo)"| B end

subgraph Partner["Risk Transfer"] E["Lifecycle Partner (Cirkadis)"] -->|"Underwrites Residual Value"| C end ```

To bridge this gap without forcing startups to write massive upfront checks to hardware manufacturers, financing frequently relies on Fair Market Value (FMV) leases or Special Purpose Vehicles (SPVs). Under an FMV lease, a leasing company purchases the equipment and leases it to the Neocloud for $2 million to $3 million per month, allowing the Neocloud to comfortably service the lease using its $5 million monthly contract revenue. However, traditional banks do not understand the depreciation curves of cutting-edge GPUs and fear ending up "upside down" on the lease. A lifecycle partner fills this void by underwriting the future residual value of the used AI clusters (such as Blackwell GPUs) three years out, giving lenders the underwriting confidence required to structure the transaction.

- Secondary market risks include macroeconomic bubbles, sudden OEM price cuts on new hardware, workload shifts, and shifting export control regulations (e.g., changes in trade restrictions surrounding shipments to China). - Distributing market risk across the bank, the Neocloud, and the hardware lifecycle partner makes large-scale GPU financing viable.

---

[11:05](https://www.youtube.com/watch?v=wOQMxS-VymU&t=665s) Relieving Balance Sheet Constraints and Managing Mid-Lease Disruption

Commercial banks, lenders, and private equity firms face strict internal concentration limits and cannot allocate all their capital to a single asset class. Even the largest global financial institutions become cautious when multiple AI hardware deals run into hundreds of millions or billions of dollars. Once lenders reach their internal risk thresholds, they must turn away viable Neoclouds with lucrative offtake contracts. A lifecycle partner resolves this bottleneck by purchasing the residual value risk off the lender’s balance sheet—either at origination or mid-lease—thereby freeing up lender capacity to finance additional AI build-outs.

```mermaid flowchart TD accTitle: Lifecycle Balance Sheet Risk Transfer subgraph Stage1["1. Origination & Deployment"] A["Neocloud with Offtake Contract"] --> B["Lender Capital Deployed"] end

subgraph Stage2["2. Capacity Saturation"] B --> C["Lender Hits Asset Class Limit"] end

subgraph Stage3["3. De-Risking & Capital Recycling"] C --> D["Cirkadis Buys Residual Value"] D --> E["Balance Sheet Cleared for New Leases"] end ```

Lifecycle partners also provide essential operational coverage during corporate disruptions. With industry consolidation inevitable—whether through Neocloud mergers or acquisitions by hyperscalers like Google or Amazon—integrated companies frequently inherit redundant data center footprints mid-lease. A lifecycle partner manages the physical extraction, data sanitization, recertification, and global remarketing of those displaced, high-value clusters, ensuring alignment among hardware vendors (like Supermicro, Dell, and HPE), leasing agents, and infrastructure operators.

- Unlocking residual value off lender balance sheets modernizes legacy equipment financing structures for modern AI demand. - Turnkey lifecycle support spans origination strategy, mid-lease maintenance, physical redeployment, and end-of-life disposition.

---

[15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) The Philosophy Behind Cirkadis

The presentation concludes with the origin and ethos of Cirkadis. Beyond high-end GPUs that get refurbished and remarketed for secondary AI workloads, broader IT circularity requires maximizing the lifespan of every constituent component. Even legacy components such as older CPUs and DDR3 RAM continue to find secondary demand across global electronics supply chains.

The name **Cirkadis** blends three core concepts: 1. **Circularity**: Maximizing hardware reuse and minimizing electronic waste. 2. **Cadence**: Synchronizing with the rhythmic upgrade and refresh cycles of data center technology. 3. **Cicadas**: Reflecting the multi-year developmental life cycle of the cicada insect.

The speaker closes on a lighthearted note, thanking the room and joking that the audience had successfully "doubled in size to 10 people."

## Notable quotes > [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) When an organization decommissions, retires, or reaches end-of-life on IT hardware, it confronts two primary liabilities: data liability and environmental liability. — Ben Cartwright

> [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) These machines frequently store proprietary network topologies and complex internal cluster configurations embedded across hardware components, not merely on central hard drives. — Ben Cartwright

> [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) Some large enterprise operators currently elect to throw $500,000 AI hardware units directly into shredders solely to guarantee complete data destruction. — Ben Cartwright

> [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) A standard ITAD provider only touches hardware at the very end of its operational life. In contrast, an IT hardware lifecycle partner works backward up the timeline into the middle of the asset's lifespan. — Ben Cartwright

> [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) Secondary market valuations for identical equipment can fluctuate wildly depending on channel access, with quotes varying from $100,000 to over $300,000 for the same asset. — Ben Cartwright

> [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) Because lead times for new AI hardware often range from six months to a full year, organizations must actively plan their decommissioning and replacement cycles one to two years ahead of time. — Ben Cartwright

> [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) A Neocloud startup might secure a high-value customer offtake contract—for instance, generating $5 million a month over three years—yet have virtually zero credit history when approaching commercial banks for a $500 million equipment loan. — Ben Cartwright

> [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) A lifecycle partner fills this void by underwriting the future residual value of the used AI clusters (such as Blackwell GPUs) three years out, giving lenders the underwriting confidence required to structure the transaction. — Ben Cartwright

> [11:05](https://www.youtube.com/watch?v=wOQMxS-VymU&t=665s) A lifecycle partner resolves this bottleneck by purchasing the residual value risk off the lender’s balance sheet—either at origination or mid-lease—thereby freeing up lender capacity to finance additional AI build-outs. — Ben Cartwright

> [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) Beyond high-end GPUs that get refurbished and remarketed for secondary AI workloads, broader IT circularity requires maximizing the lifespan of every constituent component. — Ben Cartwright

## Mentioned Ben Cartwright, Cirkadis, NVIDIA H200, NVIDIA GB300, NVIDIA Blackwell, Supermicro, Dell, HPE, Google, Amazon, Ai4

## Video description Ben Cartwright - CEO, Cirkadis

AI hardware and GPUs are driving up spending and changing the way the IT hardware market works. Neoclouds and others who need large build-outs face some difficult realities when looking for financing. This presentation will introduce the latest solutions that Cirkadis offers and how they can facilitate transactions, mitigate risk for lenders, and free up more capital for future investments. By the end of the presentation, you'll know why Cirkadis might be the most important company at this show.

[Source](https://www.youtube.com/watch?v=wOQMxS-VymU)

中文简报

Cirkadis承保GPU残值，助力Neocloud解锁银行租赁融资 全生命周期合作伙伴从出借方资产负债表中剥离并承担硬件残值风险，使轻资产型Neocloud能够通过公允市场价值租赁为数百万美元的GPU集群融资。 本次演讲探讨了采购大规模AI GPU集群所面临的运营与财务障碍，并将传统IT资产处置（ITAD）与前瞻性硬件全生命周期管理进行了对比。虽然标准ITAD侧重于后期的数销毁与环境合规责任，但高价值AI服务器更需要多年规划和残值承保。Cartwright阐述了Neocloud如何借助客户承购合同支持的公允市场价值（FMV）租赁，并利用全生命周期合作伙伴将二级市场折旧风险从商业贷款机构的资产负债表中剥离。这一综合策略不仅实现了资本的持续循环利用，还能防范行业整合期间租期中途发生中断的风险，并促进了全球供应链中的硬件循环利用。 AI 硬件退役面临数据安全与环保合规的双重责任；由于 AI 服务器硬件组件直接嵌入了专有的集群网络拓扑，专业 ITAD 服务可提供认证数据清除，避免企业将价值数十万美元的硬件直接粉碎。 (0:04) 硬件生命周期管理比传统 ITAD 更早介入资产周期，利用二级市场定价情报优化设备轮换；鉴于 AI 硬件交付周期长达半年至一年，运营商需提前一至两年统筹规划退役与升级。 (3:12) 新兴 Neocloud 初创企业缺乏长期信贷记录，难以获得数亿美元的传统硬件贷款，正通过公允市场价值（FMV）租赁与 SPV 架构，以客户承购协议的月收入直接覆盖设备租金。 (6:06) 生命周期服务商通过承销二手 AI 集群的未来残值，为放贷机构提供定价确定性，并在租约中期分担残值风险以释放银行的资产负债表额度，从而打通大规模 GPU 债务融资通道。 (11:05)

Reading summary

Scaling capital-intensive AI infrastructure requires shifting from reactive end-of-life IT asset disposition to proactive lifecycle management. By underwriting the future residual value of GPU clusters, lifecycle partners enable commercial lenders to finance emerging Neocloud providers while mitigating balance sheet and asset-devaluation risks. Decommissioning Liabilities and Asset Preservation (0:04) Decommissioning enterprise IT hardware introduces dual liabilities: data security and environmental compliance. While standard IT Asset Disposition (ITAD) handles personal data on hard drives, AI GPU servers introduce heightened data liability because proprietary network topologies and cluster configurations are embedded directly across hardware components. To eliminate breach risks, some operators resort to physically shredding hardware units valued at $500,000. At the same time, environmental regulations prohibit landfill disposal, mandating component disassembly, reuse, or certified smelting. Specialized lifecycle providers offer serialized software sanitization as a value-retaining alternative to physical destruction. Because current secondary market demand remains elevated, the vast majority of retired AI systems can be refurbished, recertified, and redeployed rather than scrapped. Upstream Lifecycle Strategy Amid Rapid Refresh Cycles (3:12) Traditional ITAD engages hardware only at the end of its operating life, but AI infrastructure demands lifecycle planning one to two years in advance. Procurement cycles operate under compressed timeframes, driven by customer offtake contracts tied to specific chip generations, such as NVIDIA H200s or GB300s, alongside hardware manufacturing lead times of six to twelve months. A mid-life hardware partner monitors secondary market dynamics to optimize decommissioning and replacement timing. Secondary market pricing for identical assets can fluctuate significantly based on remarketing channels, with quotes ranging from $100,000 to over $300,000. Managing these assets mid-cycle enables data center operators to balance physical floor-space constraints against hardware depreciation and secondary market recovery values. Underwriting Residual Values for Neocloud Financing (6:06) Procuring modern AI hardware demands capital tranches scaling from $50 million to over $1 billion. Emerging Neocloud startups frequently hold lucrative customer contracts—such as $5 million monthly over three years—yet lack the multi-decade credit history required by commercial banks for nine-figure loans. Financing structures consequently rely on Fair Market Value (FMV) leases or Special Purpose Vehicles, where a lessor purchases the hardware and leases it back for $2 million to $3 million per month. Because traditional commercial banks lack models to evaluate rapid GPU depreciation curves, they risk falling upside down on leases. Lifecycle partners bridge this divide by underwriting the residual value of used clusters three years forward. This shared-risk model enables lease execution while factoring in secondary market hazards, such as original equipment manufacturer price cuts, macroeconomic volatility, and regulatory shifts like export controls on shipments to China. Mitigating Balance Sheet Concentration and Lease Disruptions (11:05) Commercial lenders and private equity firms face internal concentration caps that restrict the total capital they can deploy into a single asset class. Once institutions hit these limits on multi-hundred-million-dollar AI hardware portfolios, they must halt lending to creditworthy operators. Lifecycle partners alleviate this bottleneck by acquiring residual value exposure off the lenders' balance sheets, either at origination or mid-lease, thereby restoring lending capacity for new deployments. Additionally, lifecycle management provides operational contingency planning during market consolidation. When mergers occur or hyperscalers acquire Neoclouds, operators often inherit redundant data center footprints mid-lease. Lifecycle partners manage physical de-installation, cluster recertification, and global remarketing, maintaining operational continuity across hardware manufacturers, leasing agents, and data center operators. The viability of high-capital AI financing depends on multi-year hardware circularity, extending from entire GPU clusters down to legacy components like CPUs and DDR3 memory. However, the long-term effectiveness of residual value underwriting remains contingent on secondary market absorption capacity, geopolitical export constraints, and the pace of industry consolidation.

精炼阅读摘要

Ai4举办的本次演讲指出，AI基础设施的巨额资本开支和快速迭代正在颠覆传统IT资产处置模式；通过引入涵盖残值承保的硬件全生命周期管理，不仅能解决新型云服务商缺乏信用记录导致的融资瓶颈并释放银行的资产负债表额度，还能在平衡数据安全与环保责任的同时实现硬件资产价值最大化。 AI硬件退役的特殊责任与处置转型 (0:04) 演讲指出，IT硬件在退役或达到生命周期终点时，主要面临数据安全和环境合规双重责任。传统IT资产处置（ITAD）通过工业粉碎机进行物理销毁或通过认证软件擦除来防止个人身份信息泄露。然而，AI GPU服务器具有截然不同的风险特征：它们通常将专有网络拓扑结构和复杂的内部集群配置嵌入在硬件组件本身，而非仅存储于中央硬盘上。 面对这种风险，部分大型企业由于过度担忧数据泄露，直接选择将价值50万美元的AI硬件设备送入粉碎机。演讲者强调，虽然企业拥有处置自有资产的权利，但专业的ITAD服务能够提供更高效、更具可持续性且能保留价值的数据清除方式。目前由于二级市场的强劲需求，绝大多数退役的AI硬件都会经过翻新、再营销并重新部署，而非直接报废拆解。硬件退役的诱发因素主要包括常规更新周期、企业破产以及并购活动。 生命周期策略前移与提前规划周期 (3:12) 传统的ITAD提供商通常仅在硬件运营周期的最末端介入，而全生命周期合作伙伴则将服务前移至资产寿命的中期阶段。生命周期合作伙伴通过监测市场行情，帮助运营商综合评估数据中心空间限制、二级市场价格波动以及硬件重新认证方案，从而制定最优的设备更新周期。根据演讲中的数据，相同设备在不同渠道中的二级市场估值差异显著，报价可能在10万美元到30万美元以上之间波动。 与传统企业IT相比，AI基础设施的采购和升级周期大幅压缩。传统服务器可长期承担常规计算负荷，而AI运营商的算力销售通常通过与特定芯片架构（如NVIDIA H200或GB300）直接绑定的消纳合同来进行。鉴于新型AI硬件的交付周期通常长达6个月至整整1年，企业必须提前1至2年主动规划退役与替换周期。演讲指出，这种管理模式将销售和业务拓展策略与硬件折旧及采购时间表紧密结合，并明确了服务器内部复用与公开市场出售的时机，以最大化回收资产价值。 Neocloud融资困境与FMV租赁机制 (6:06) 采购AI硬件需要规模空前的资本开支，单笔交易规模动辄达到5000万美元、1亿美元、5亿美元甚至10亿美元。这给新兴的Neocloud服务商带来了结构性融资难题。与历经数十年积累资产负债表的传统主机托管客户不同，Neocloud初创公司虽然可能已锁定优质客户合同（例如为期3年、每月产生500万美元收入的消纳协议），但在向商业银行申请5亿美元设备贷款时，往往几乎没有任何信用历史。 为解决巨额前期现金支出难题，行业常借助公允市场价值（FMV）租赁或特殊目的载体（SPV）进行融资。在FMV租赁模式下，租赁公司出资采购设备并以每月200万至300万美元的价格出租给Neocloud，使后者能通过其每月500万美元的合同收入顺利履约。然而，传统商业银行并不理解尖端GPU的折旧曲线，担心在租赁期内陷入资产贬值风险。生命周期合作伙伴（如Cirkadis）通过对使用3年后的AI集群（如Blackwell GPU）的未来残值进行承保，为贷款机构提供了结构化交易所需的承保信心。同时，演讲列举了二级市场的多项潜在风险，包括宏观经济泡沫、原始设备制造商（OEM）对新硬件的突发降价、计算工作负载转移，以及针对出口管制（如对华发货限制）的监管政策变动。 释放资方集中度限额与中途资产重组 (11:05) 商业银行、贷款机构和私募股权基金面临严格的内部行业集中度限制，无法将所有资本倾斜至单一资产类别。即便对于大型跨国金融机构，当多笔AI硬件交易规模累计达数亿或数十亿美元时，也会触发内部风险阈值，从而不得不拒绝拥有高额消纳合同的合规Neocloud项目。全生命周期合作伙伴通过在业务发起阶段或租赁中途直接从贷款方的资产负债表上购买残值风险，有效释放了后者的资本额度，使其能够继续为新的AI建设项目提供资金支持。 此外，随着行业整合的不可避免——无论是Neocloud之间的横向并购，还是被Google、Amazon等超大规模云服务商收购——合并后的企业常常在租赁期中途接盘多余的数据中心空间。全生命周期合作伙伴能够统筹管理这些高价值闲置集群的物理拆卸、数据清除、重新认证及全球再营销，并在硬件供应商（如Supermicro、Dell、HPE）、租赁代理商与基础设施运营商之间建立协同。这种一站式生命周期服务覆盖了项目发起、租期内维护、物理再部署及最终报废处置的全部流程。 Cirkadis循环理念与元件再利用 (15:37) 演讲最后阐述了Cirkadis品牌的起源与发展理念。除高端GPU被翻新并重新投放至二级AI工作负载之外，更广泛的IT循环经济还要求最大化利用每个组成部件的寿命。演讲指出，即使是旧款CPU和DDR3内存等老旧零部件，在全球电子产品供应链中依然保持着持续的次级市场需求。 “Cirkadis”这一名称融合了三大核心概念：一是循环性（Circularity），即最大化硬件再利用并减少电子垃圾；二是节奏（Cadence），即与数据中心技术的周期性升级和更新节拍保持同步；三是蝉（Cicadas），寓意效仿蝉类昆虫经历多年的发育生命周期。 AI基础设施的极速迭代对传统硬件报废管理和金融信贷模型构成了双重挑战。尽管全生命周期管理模式为解决Neocloud融资和硬件残值承保提供了操作路径，但未来二级市场的流动性与残值表现仍高度依赖于全球出口管制政策的演变、算力工作负载的转移以及OEM厂商的价格调整。

Detailed notes — zh

## Background 本内容由频道 Ai4 于 2026 年 8 月 18 日在 YouTube 平台发布。视频记录了 Cirkadis 首席执行官 Ben Cartwright 的主题演讲，主要探讨在 AI 硬件与 GPU 资本开支激增的背景下，如何通过全生命周期管理与残值风控方案，帮助新兴算力云厂商（Neocloud）化解融资瓶颈并释放投资资本。 ## TL;DR 演讲深入剖析了 AI 时代算力硬件从传统 ITAD 终端处置向“全生命周期管理”转型的必然趋势。面对 Neocloud 动辄数亿美元的 GPU 采购需求与传统银行信贷门槛之间的矛盾，公允市场价值租赁（FMV Lease）成为核心路径，而精准预测资产残值则是交易成败的关键。Cirkadis 通过在租赁初期或中期主动买断设备残值、剥离金融机构的资产负债表风险，打通了硬件翻新再营销与循环经济链条，从而为算力运营商、资金方与硬件厂商创造了多方共赢的融资与处置生态。 ## Atmosphere & style 整场演讲呈现出务实、精炼且极具行业洞察力的商业对谈风格。演讲者逻辑清晰地将底层硬件处置的实操细节与宏观金融租赁架构串联起来，语言生动直白且不乏幽默感。尽管身处专业会议现场且听众规模不大，演讲者依然保持着从容自信的节奏，在严谨剖析算力折旧与资本痛点的同时，以轻松自嘲的语调营造出亲切而富有说服力的交流氛围。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) IT 硬件退役面临数据责任与环境责任两大核心痛点，推动传统 IT 资产处置（ITAD）服务向规范化与多路径利用演进。 - [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) AI 硬件除涉及个人身份信息外，还包含内部网络拓扑与架构等敏感数据，部分巨头甚至选择将价值数十万美元的设备直接粉碎以确保合规。 - [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) 行业正从末端 ITAD 转向全生命周期合作伙伴（Lifecycle Partner），将服务重心前移至硬件服役中期，精细规划升级时机与残值最大化路线。 - [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) AI 算力采购高度绑定 1 至 3 年期的算力承购合同（Offtake contracts），加上漫长的硬件交付周期，迫使企业须提前 1 至 2 年制定退役与换代策略。 - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) 新兴 Neocloud 虽拥有充足的月度算力承购现金流，但因成立时间短、缺乏资产负债表背书，难以直接向传统银行借贷数亿美元采购顶级 GPU。 - [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) 公允市场价值租赁（FMV Lease）成为解决 Neocloud 资本痛点的核心模式，使运营商能以每月租金形式部署数亿美元算力集群并实现盈利。 - [08:48](https://www.youtube.com/watch?v=wOQMxS-VymU&t=528s) FMV 租赁模式成立的关键在于精准预测硬件的未来残值，生命周期合作伙伴凭借二级市场数据模型为金融机构提供定价与风控锚点。 - [08:48](https://www.youtube.com/watch?v=wOQMxS-VymU&t=528s) 二手 AI 硬件残值面临算力行业需求周期波动、NVIDIA 原厂定价策略调整以及地缘出口管制政策变化等多重市场风险。 - [11:52](https://www.youtube.com/watch?v=wOQMxS-VymU&t=712s) 出资机构受限于资产类别集中度与投资组合风险敞口上限，在完成数笔大规模硬件融资后往往被迫暂停承接新的 AI 租赁业务。 - [11:52](https://www.youtube.com/watch?v=wOQMxS-VymU&t=712s) Cirkadis 等生命周期服务商通过买断设备残值将贬值风险从出资方资产负债表剥离，有效释放金融资本额度以支持更多算力交易落地。 - [13:42](https://www.youtube.com/watch?v=wOQMxS-VymU&t=822s) 行业并购整合将引发大规模数据中心资产重构，对涉及硬件现场搬迁、数据擦除合规及再营销分销的端到端最终处置能力提出更高要求。 - [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) 循环经济实践不仅覆盖退役 GPU 的翻新转售，还延伸至 CPU 及 DDR3 等老一代零部件在全球新兴设备制造中的梯次重用。 - [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) Cirkadis 品牌融合了循环经济（Circularity）、硬件升级节奏（Cadence）及具备多年周期特性的蝉（Cicadas）三重核心理念。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) IT 资产退役的两大责任与 ITAD 的演变 在探讨数据中心与硬件生态时，一切通常始于 IT 硬件的退役（Decommissioning）阶段。当任何负责管理 IT 硬件的机构面临设备报废、达到使用寿命（End of life）或淘汰时，资产所有者必须处理两大核心责任：第一是**数据责任（Data liability）**，即如何确保设备上的数据彻底销毁而不落入不法之徒手中；第二是**环境责任（Environmental liability）**，废旧设备绝不能简单一扔了之，必须通过某种方式进行回收或再利用。为了解决这两大痛点，市场上诞生了专门提供此类服务的行业——IT 资产处置（ITAD, IT Asset Disposition）。 在数据责任方面，当设备（无论是笔记本电脑、通用服务器还是 GPU 服务器）退役时，通常会通过物理粉碎（Put in a shredder）或经序列化认证的专业软件擦除来销毁数据。这类退役需求不仅出现在常规的硬件升级中，也频繁发生在企业破产清算、并购（M&A）等场景下。除了医院等机构涉及的大量个人身份信息（PII）需要销毁外，在 AI 时代，GPU 服务器内部还可能包含关于内部网络架构或系统配置的敏感信息，这些信息甚至不一定仅存储在硬盘上。一个优秀的 ITAD 服务商必须清晰了解设备各处存储的数据形态并确保其彻底消除。演讲者特别提到一个极端的现实案例：有些运营昂贵 AI 硬件的大型巨头，在硬件服役短短几年后选择将整台设备直接送进粉碎机——"那是一大笔钱，单台价值可能高达 50 万美元的设备直接被扔进粉碎机，仅仅是为了彻底确认每一比特的数据都被销毁了。" 在环境责任方面，化解合规风险的手段包括对设备进行翻新再售、拆解至零部件级别重新利用，或送往冶炼厂进行原料回收。但在当前的 AI 行业阶段，由于市场算力需求极度旺盛，绝大多数 AI 硬件退役后并不会被拆解或熔炼，而是通过翻新（Refurbish）、再营销（Remarket）和重新部署（Redeploy）继续发挥价值。 - **数据责任销毁方式**：物理粉碎、经认证并记录序列号的软件擦除。 - **敏感数据范畴**：个人身份信息（PII）、AI 网络拓扑与专属配置参数。 - **环境合规路径**：设备翻新转售、零部件级拆解再利用、冶炼回收。 --- ### [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) 从终端处置到全生命周期管理：AI 硬件的刷新周期与价值波动 行业正在发生关键转型——从单纯在生命周期终点提供服务的 ITAD 公司，进化为覆盖全流程的**生命周期合作伙伴（Lifecycle Partner）**。生命周期合作伙伴的核心价值在于为客户提供深度的市场情报与战略支持。当企业管理着海量高价值 IT 资产时，核心痛点是何时将旧设备退役并引入新设备（即刷新/升级周期）。这涉及精细的策略考量：数据中心机架空间的物理限制、新旧硬件交付与腾退的时间窗口，以及二手硬件市场价格的剧烈波动。 在二手硬件的再认证与再营销环节，市场估值往往存在巨大分歧。同一台旧设备，某家估价机构可能给出 10 万美元，而另一家可能会给出 20 万甚至 30 万美元。生命周期合作伙伴的工作重心因此从时间线的"最末端"逐步前移：不再仅仅是在设备报废时处理残值，而是在硬件仍处于正常服役的中期阶段，就帮助企业规划最佳退役时机、评估内部复用可行性，并制定最大化资产残值回收的再营销路线。 AI 时代的硬件采购与传统数据中心有着根本不同。过去，传统服务器的使用寿命相对更长，只要能承载业务负载就可以继续运行；但在 AI 领域，企业往往是围绕具体的计算合同（Offtake contracts）来开展业务的。客户的算力合同通常会明确指定硬件架构（例如 NVIDIA H200 或 GB300/Blackwell），合同期限一般为 1 到 3 年。当这期合同结束、进入下一轮谈判时，客户往往要求使用最新一代的硬件。加之 AI 硬件交付周期动辄长达 6 个月到 1 年，企业必须提前 1 到 2 年在硬件退役前就制定下一代采购与处置战略。 ```mermaid flowchart LR accTitle: 从传统ITAD到全生命周期合作伙伴的转变 subgraph S1["传统 ITAD 模式 (被动终端处置)"] A["硬件报废 / 退役"] --> B["数据物理粉碎 / 软件擦除"] B --> C["基础拆解 / 冶炼回收"] end subgraph S2["全生命周期合作伙伴模式 (主动全周期管理)"] D["采购与融资支持 (残值预测)"] --> E["服役中期资产监控与空间优化"] E --> F["合同到期策略规划 (提前1-2年)"] F --> G["高溢价翻新 / 再营销 / 二手部署"] end ``` --- ### [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) 巨额采购背后的折旧难题与 Neocloud 的融资困局 随着 AI 算力集群规模的急剧扩大，硬件采购金额迅速从 5000 万、1 亿美元飙升至 5 亿甚至 10 亿美元级别。面对如此庞大的资本开支，折旧曲线（Depreciation curve）的把控变得生死攸关：硬件今天的采购成本是多少？在服役期内能产生多少营收？何时退役？退役时在二级市场上还能值多少钱？这种跨越硬件技术与金融残值的战略对话，传统银行根本无力评估，新兴的 Neocloud（新型云算力厂商）自身也缺乏专业经验，这正是硬件生命周期合作伙伴发挥不可替代作用的切入点。 现代算力市场的商业形态催生了严峻的现金流与信用矛盾。过去，能在数据中心机房（Colo）部署数千万、数亿美元硬件的通常是经营多年的老牌大型企业；如今，市场上涌现出大量刚成立不久的 Neocloud 初创公司。这些公司往往已经拿到了极具吸引力的算力承购合同——例如客户签约承诺在未来 3 年内每月支付 500 万美元来购买算力。然而，为了履行这份合同，Neocloud 需要立即采购价值 5 亿美元的顶级 GPU 硬件。当他们拿着合同找到银行申请 5 亿美元贷款时，银行的反应往往是："你是谁？你的公司网站好像昨天才刚刚上线。" 由于初创 Neocloud 缺乏信用背书和足够的资产负债表来直接借款 5 亿美元，即使通过设立特殊目的实体（SPV）或抵押贷款也困难重重。在此背景下，**公允市场价值租赁（FMV Lease, Fair Market Value Lease）** 成为最可行且最具吸引力的路径：由租赁公司出资全额买下硬件设备，再租赁给 Neocloud 使用。 - **Neocloud 商业模式测算**：每月从算力承购客户处获得 500 万美元营收，每月向租赁公司支付 200 万至 300 万美元的硬件租金，从而在无需一次性支付 5 亿美元现金的情况下实现盈利。 - **核心资本痛点**：具备每月支付租金的充沛现金流能力，但无力直接开具巨额支票向硬件供应商全款提货。 --- ### [08:48](https://www.youtube.com/watch?v=wOQMxS-VymU&t=528s) 公允市场价值租赁、残值预测与多方风险分担 在 FMV 租赁模式下，整个交易能否跑通完全取决于对**资产残值（Residual value）**的精准预测。提供资金的银行和租赁机构最担忧的风险是租赁期内发生"资产倒挂"（Upside down，即设备实际贬值速度快于租金回收与贷款本金递减速度）。银行必须明确知道：一套今天全新价值 5 亿美元的设备，在未来 3 年的生命周期中每年的折旧曲线是怎样的？租期结束时的残值究竟是多少？ 传统银行擅长评估金融信用，但对 AI 硬件一无所知；Neocloud 的核心能力在于商务拓展和算力销售。五年前，生命周期公司处理的还是退役的 Dell 和 HP 服务器，而如今顶级的生命周期合作伙伴已转型为 AI 硬件残值预测专家。他们能够基于二级市场供求模型明确给出判断："3 年之后，这批价值 5 亿美元的 Blackwell GPU 预计在二级市场仍值多少钱。"有了这一残值锚点，金融机构才能精确计算出每月租金并完成租赁交易结构的搭建。 ```mermaid flowchart TD accTitle: Neocloud FMV租赁与生命周期合作伙伴协同模式 subgraph Funding["资金与资产端"] Bank["金融机构 / 租赁公司 (提供5亿美元资本)"] LCP["生命周期合作伙伴 (精算3年Blackwell残值)"] end subgraph Operation["运营与履约端"] NC["Neocloud 算力运营商 (每月支付200-300万租金)"] Client["算力最终用户 (签订3年合同/每月支付500万)"] end LCP -.->|"提供残值评估与风险兜底"| Bank Bank -->|"采购硬件并提供FMV租赁"| NC Client -->|"支付算力承购费用"| NC NC -->|"交付算力集群服务"| Client ``` 当然，二手 AI 硬件残值预测也面临着多重市场风险： 1. **行业泡沫与需求周期**：AI 算力需求是否会发生阶段性回落。 2. **原厂定价策略冲击**：若 NVIDIA 突然对新一代或现有硬件进行官方降价，二手设备市场价格将遭遇剧烈冲击。 3. **工作负载演进与地缘合规约束**：目前出口管制限制了部分高端芯片流向特定海外市场（如中国市场）；但若未来监管政策发生变动，解禁可能会释放庞大的二手需求并推高残值，反向政策收紧则可能压低流通性。 在成熟的生态体系中，银行承担一部分信贷风险，Neocloud 承担运营与销售风险，生命周期合作伙伴承担资产处置与残值管理风险，从而使单笔数亿美元的超大规模交易不再由单一实体独吞所有风险。此外，面对 AI 领域动辄数以亿计的单笔硬件融资，即使是全球最大的银行也受限于投资组合多元化（Diversification）与单一资产类别集中度的资本约束，必须依靠这种精细化的残值风控机制来推动融资落地。 ### [11:52](https://www.youtube.com/watch?v=wOQMxS-VymU&t=712s) 资产敞口限制与残值分担机制 当融资交易的规模飙升至 1 亿、5 亿甚至 10 亿美元时，出资机构会变得极其谨慎。无论是传统租赁公司、银行还是私募股权投资机构，都不愿意在单一资产类别中承担过高的风险敞口。在实际操作中，很多出资方在完成几笔 AI 硬件的贷款或租赁交易后，由于内部风险控制的约束，决策层便会踩下刹车，明确表示不愿再增加 AI 资产类别的敞口。然而，市场需求依然非常火爆，每天都有带着优质采购合同（off-take contracts）和成熟商业模式的 Neocloud（新型云厂商）找上门寻求融资，而出资机构却受限于资产集中度上限，只能对这些优质业务望洋兴叹。 面对这种行业痛点，像 Cirkadis 这样专注于实物资产与全生命周期管理的合作伙伴应运而生。Cirkadis 愿意主动承担这些资产风险，不仅可以在租赁协议签署之初介入，也能在租赁周期中途加入。Cirkadis 对租赁机构提出解决方案：“让我们把这项风险从你们的资产负债表上剥离出去”（"Look, let's take that risk off your balance sheet."）。双方可以重构租赁合同结构，由 Cirkadis 直接买断设备的残值（residual value）。由于 Cirkadis 具备深厚的二手硬件再利用与残值变现能力，出资机构无需再在表内承担二手设备的贬值风险，从而释放出大量资本额度，能够重新在该资产类别中承接更多新业务。 ```mermaid flowchart LR accTitle: GPU 租赁残值风险剥离流程 subgraph Capital ["出资机构与客户"] Lender["租赁公司 / 私募基金"] Neocloud["Neocloud 客户"] end subgraph Lifecycle ["Cirkadis 赋能"] RiskRelief["买断残值 / 剥离负债表风险"] HardwareReuse["二手翻新 / 硬件再销售"] end Neocloud -->|"优质 AI 算力合同"| Lender Lender -->|"释放资本额度开展新业务"| Neocloud Lender -->|"转移设备残值风险"| RiskRelief RiskRelief -->|"资产处置与流通"| HardwareReuse ``` - **资本约束瓶颈**：单笔交易动辄达到数亿美元规模，迫使机构为控制集中度风险而暂停 AI 硬件租赁。 - **模式创新**：过去几年 AI 爆发式增长，暴露出传统借贷与租赁模式的不足，推动了残值剥离与重组模式的诞生。 --- ### [13:42](https://www.youtube.com/watch?v=wOQMxS-VymU&t=822s) 行业整合与全生命周期 ITAD 处置 AI 算力硬件的管理是一项贯穿始终的工程。从最初采购阶段制定生命周期规划与经济模型，到租期中段的运维、迁移与二次部署，再到期末的物理下架与处置，每一个环节都极其繁琐。随着行业发展，Neocloud 领域极有可能迎来并购整合浪潮。例如两家 Neocloud 合并，或者被 Google、亚马逊等巨头收购，整合双方 IT 基础设施的诉求便会浮出水面。此时，机房中可能存放着价值数千万甚至数亿美元的高性能硬件，但在合并后的技术架构中已不再需要，企业会要求“把这些用不上的硬件清理掉”。 硬件淘汰带来了一系列严峻的实操问题：谁负责去数据中心现场搬迁硬件？谁来承担数据擦除与安全合规的法律责任？谁来负责设备的再营销（remarketing）与分销变现？在这一连串复杂的最终处置（final disposition）链条中，市场对专业全生命周期伙伴的需求愈发迫切。一个具备端到端专业能力的合作伙伴，不仅帮助运营算力的 Neocloud 跑通商业模型，也为来自 Supermicro、Dell、HP 等硬件厂商的销售人员提供了促成交易的融资工具，让产业链上的每一方都能从中受益。 - **行业并购挑战**：企业合并导致数据中心硬件资产重构，中途退役的高价值硬件亟需合规清退与变现渠道。 - **全流程价值链**：覆盖前期架构设计、中期维护再部署、后期物理回收、数据安全责任承担与二次销售。 - **生态共赢**：为硬件制造商销售团队、算力运营商及出资方同时解决资本流动性与资产处置难题。 --- ### [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) 循环经济与 Cirkadis 的命名寓意 在硬件生命周期结束时，优秀的商业实践在于最大限度地实现循环利用，避免资源浪费。除了 GPU 本身在退役后可以被翻新、转售并在二手市场继续服役数年之外，传统 IT 资产处置（ITAD）业务中的旧服务器、旧内存（RAM）和旧 CPU 依然存在广阔的市场需求。例如，即使是较老的 DDR3 内存，在世界上很多地区的全新电子设备制造中仍被广泛采购和重用。 演讲最后，Ben Cartwright 揭示了公司名称“Cirkadis”的灵感来源：它融合了循环经济目标（circularity）、算力升级的节奏周期（cadence），以及具有多年周期性生命特征的昆虫“蝉”（cicadas）。公司将这些概念融为一体，代表了对硬件全生命周期管理的系统性思考。演讲在轻松幽默的气氛中收尾，演讲者打趣道现场听众已经“翻倍增长到了 10 个人”，圆满结束了整场分享。 - **深层循环利用**：不仅限于主流 GPU 的翻新再售，还涵盖 CPU 以及 DDR3 等老一代零部件的全球梯次利用。 - **品牌寓意拆解**： - **Circularity（循环性）**：最大化硬件资产残值与环保循环。 - **Cadence（节奏/周期）**：契合数据中心硬件升级迭代的固定节奏。 - **Cicadas（蝉）**：象征跨越数年的多阶段生命周期模型。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=wOQMxS-VymU&t=4s) 那是一大笔钱，单台价值可能高达 50 万美元的设备直接被扔进粉碎机，仅仅是为了彻底确认每一比特的数据都被销毁了。 — Ben Cartwright > [03:12](https://www.youtube.com/watch?v=wOQMxS-VymU&t=192s) 生命周期合作伙伴的工作重心因此从时间线的“最末端”逐步前移：不再仅仅是在设备报废时处理残值，而是在硬件仍处于正常服役的中期阶段，就帮助企业规划最佳退役时机。 — Ben Cartwright > [06:06](https://www.youtube.com/watch?v=wOQMxS-VymU&t=366s) 你是谁？你的公司网站好像昨天才刚刚上线。 — Ben Cartwright > [08:48](https://www.youtube.com/watch?v=wOQMxS-VymU&t=528s) 3 年之后，这批价值 5 亿美元的 Blackwell GPU 预计在二级市场仍值多少钱。 — Ben Cartwright > [11:52](https://www.youtube.com/watch?v=wOQMxS-VymU&t=712s) Look, let's take that risk off your balance sheet. — Ben Cartwright > [11:52](https://www.youtube.com/watch?v=wOQMxS-VymU&t=712s) 让我们把这项风险从你们的资产负债表上剥离出去 — Ben Cartwright > [13:42](https://www.youtube.com/watch?v=wOQMxS-VymU&t=822s) 把这些用不上的硬件清理掉 — Ben Cartwright > [15:37](https://www.youtube.com/watch?v=wOQMxS-VymU&t=937s) 现场听众已经“翻倍增长到了 10 个人” — Ben Cartwright ## Mentioned Ben Cartwright, Cirkadis, Ai4, NVIDIA, NVIDIA H200, Blackwell (GB300), Dell, HP, Supermicro, Google, 亚马逊 (Amazon) ## Video description Ben Cartwright - CEO, Cirkadis AI hardware and GPUs are driving up spending and changing the way the IT hardware market works. Neoclouds and others who need large build-outs face some difficult realities when looking for financing. This presentation will introduce the latest solutions that Cirkadis offers and how they can facilitate transactions, mitigate risk for lenders, and free up more capital for future investments. By the end of the presentation, you'll know why Cirkadis might be the most important company at this show. [Source](https://www.youtube.com/watch?v=wOQMxS-VymU)

Source quotes

[0:04] Ben Cartwright: 那是一大笔钱，单台价值可能高达 50 万美元的设备直接被扔进粉碎机，仅仅是为了彻底确认每一比特的数据都被销毁了。 [3:12] Ben Cartwright: 生命周期合作伙伴的工作重心因此从时间线的“最末端”逐步前移：不再仅仅是在设备报废时处理残值，而是在硬件仍处于正常服役的中期阶段，就帮助企业规划最佳退役时机。 [6:06] Ben Cartwright: 你是谁？你的公司网站好像昨天才刚刚上线。 [8:48] Ben Cartwright: 3 年之后，这批价值 5 亿美元的 Blackwell GPU 预计在二级市场仍值多少钱。 [11:52] Ben Cartwright: Look, let's take that risk off your balance sheet. [11:52] Ben Cartwright: 让我们把这项风险从你们的资产负债表上剥离出去 [13:42] Ben Cartwright: 把这些用不上的硬件清理掉 [15:37] Ben Cartwright: 现场听众已经“翻倍增长到了 10 个人”

Source quotes

[0:04] Ben Cartwright: When an organization decommissions, retires, or reaches end-of-life on IT hardware, it confronts two primary liabilities: data liability and environmental liability. [0:04] Ben Cartwright: These machines frequently store proprietary network topologies and complex internal cluster configurations embedded across hardware components, not merely on central hard drives. [0:04] Ben Cartwright: Some large enterprise operators currently elect to throw $500,000 AI hardware units directly into shredders solely to guarantee complete data destruction. [3:12] Ben Cartwright: A standard ITAD provider only touches hardware at the very end of its operational life. In contrast, an IT hardware lifecycle partner works backward up the timeline into the middle of the asset's lifespan. [3:12] Ben Cartwright: Secondary market valuations for identical equipment can fluctuate wildly depending on channel access, with quotes varying from $100,000 to over $300,000 for the same asset. [3:12] Ben Cartwright: Because lead times for new AI hardware often range from six months to a full year, organizations must actively plan their decommissioning and replacement cycles one to two years ahead of time. [6:06] Ben Cartwright: A Neocloud startup might secure a high-value customer offtake contract—for instance, generating $5 million a month over three years—yet have virtually zero credit history when approaching commercial banks for a $500 million equipment loan. [6:06] Ben Cartwright: A lifecycle partner fills this void by underwriting the future residual value of the used AI clusters (such as Blackwell GPUs) three years out, giving lenders the underwriting confidence required to structure the transaction. [11:05] Ben Cartwright: A lifecycle partner resolves this bottleneck by purchasing the residual value risk off the lender’s balance sheet—either at origination or mid-lease—thereby freeing up lender capacity to finance additional AI build-outs. [15:37] Ben Cartwright: Beyond high-end GPUs that get refurbished and remarketed for secondary AI workloads, broader IT circularity requires maximizing the lifespan of every constituent component.

Original transcript

[0:04] What I'm going to talk about is I'm going  to start with IT hardware decommissioning. [0:10] So when any body who manages IT hardware when  that gets decommissioned, end of life, retired, [0:15] whatever, the owner of the hardware has to think  about two liabilities. One, data liability. [0:22] Uh what do we do with the data that's on there?  and how how can it be destroyed so that uh [0:30] doesn't fall into the wrong hands. Second is an  environmental liability. You can't just throw [0:34] it away, right? It's got to be recycled  or reused somehow. So these are the two [0:39] liabilities that anybody who owns or manages IT  hardware has to think about at the end of life. [0:44] There's an industry that provides these two  services called ITAD, IT asset disposition. And [0:53] we're going to talk about the data liability  first. So when the device is decommissioned or [0:59] retired, we have to destroy the data. That can be  physical destruction by putting it in a shredder. [1:05] It can be software wiping and those are certified.  Those will be certified serialized. All of that [1:10] the data gets destroyed. And these events come  up obviously when somebody upgrades it or retires [1:16] a device whether it's a laptop or a server or  whatever. But it can also come up in the case [1:22] of a company going bankrupt. It comes up in cases  of mergers or acquisitions. And it's not just the [1:28] personally identifiable information. So hospital  may have a lot of personal information about us on [1:33] their devices that needs to be destroyed.  But also since we're talking about AI, [1:39] the GPU servers may have information about the  internal networks or configurations that also [1:46] may not be necessarily stored on hard drives.  So a good ITAD service provider knows all the [1:53] information that's contained on the device, where  it's contained on the device, and how to make sure [1:59] it's destroyed. Interestingly, there are some big  companies who operate very expensive AI hardware [2:06] and when they retire it after just a few years,  they throw the whole thing in the shredder. That's [2:12] a lot of money. That's $500,000 maybe per device  going into the shredder just to make sure they can [2:19] confirm that every last bit of data is destroyed.  uh so it's your prerogative if you've paid for [2:26] that hardware but uh there's more efficient and  maybe more sustainable ways to destroy the data. [2:32] So a good provider will be able to provide those  services. Second is the environmental liability. [2:39] So mitigating that means maybe refurbishing  and reselling the devices. Maybe it means uh [2:46] demanufacturing them down to a component level  and then reusing some of those components [2:52] or sending it for um recycling like smelter or  something like that. But for the purposes of AI [2:59] hardware at this point in the industry, most of it  is getting reused. There's still demand for it. So [3:04] refurbish, remarket, redeploy uh the devices.  That's what an ITAD provider will will do. Now [3:12] this specifically is talking about the transition  from an ITAD company to a life cycle partner. [3:19] So what do we mean by a life cycle partner? A  life cycle partner is going to be able to provide [3:24] to their client certain things about certain  information intelligence about the market. [3:29] For example, when you're managing a large number  of valuable IT assets, you're thinking about [3:35] when to decommission them and get replace them  with new devices, a refresh or an upgrade cycle. [3:41] What's the optimal time that there's can be some  strategy there that can help organizations operate [3:47] more efficiently? Maybe when it comes to financing  or timing of when they can get these devices [3:54] and get rid of the old devices out of a data  center. Sometimes space is a constraint and of [3:59] course price. Prices fluctuate. So when is the  best time to recover as much value as possible. [4:05] Um reertifying the assets, remarketing  the assets. Where can they move? [4:11] There's a big swing. Sometimes people get an old  asset, somebody will tell you it's worth $100,000. [4:16] Somebody else will tell you it's worth 200,000,  300,000. So finding a good life cycle partner. [4:22] We've gone from retiring assets. We talked about  decommissioning and retiring a couple of slides [4:26] back. Now, we're starting to work back up the  timeline to when those assets are still being used [4:32] and strategizing about when to retire them, how  to retire them, can they be reused internally, [4:38] are they going to get remarketed at the end,  and how to get the most value out of that. So, [4:43] we're starting to go from ITAD at the end of  a devices's life to life cycle where now we're [4:48] thinking about it in the middle of its life cycle.  Now talking about procurement is an interesting uh [4:55] topic because it's getting harder and harder.  It it traditionally data centers for example [5:02] the hardware upgrade cycles were different. A  device may have value for a little longer because [5:06] it was still able to process whatever load was  running on it. But when we talk about AI it's a [5:12] little different. People generally may be selling  compute contracts. They have offtake contracts [5:19] based on exactly that generation of hardware.  Maybe the contract specifies GB300's or H200s. So [5:27] at the end of that contract, which may  have a a 1, two, or threeear lifespan, [5:32] then the next contract negotiation is going to  be dependent on the next generation of hardware. [5:37] So these life cycles are starting to get  compressed where if there's a six-month [5:42] or a one-year lead time on some of the hardware  to get it in the data center now we really got [5:47] to start strategizing one or two years in  advance before we're retiring hardware. So [5:54] these things start to get into where a good life  cycle partner can help an organization strategize [6:00] their salespeople and business  development people strategize. So [6:06] then you start thinking about  the depreciation curve, right? So [6:12] when you get into needing to procure hardware  in large amounts, $50 million, $100 million, [6:18] $500 million, a billion dollars, when  we start talking about that much money, [6:25] then people really get interested in the strategy.  How much does it cost today? How much is it going [6:30] to make? When is it going to get retired? and how  much is it going to be worth then? And that's a [6:36] that's a conversation that banks don't know how to  have. The Neil cloud, they're not experts in that, [6:41] but a hardware life cycle partner, this is  their expertise. So that's why a good life [6:46] cycle partner is helpful. Then you start talking  about cash flow. And this is this is not like [6:54] definite facts or how things it's just an  example of how financing hardware may work. [7:01] um you have a NeoCloud company who's new. So in  the past, anybody who needed who had a bunch of [7:07] servers in a colo facility is spending tens  of millions, hundreds of millions of dollars [7:13] on data center hardware or had their own hardware,  that's not a company that just showed up yesterday [7:17] usually, right? Those are big companies that  have been around for a long time, grown into [7:21] that IT infrastructure. But nowadays, you've got  Neoclouds popping up. Neocloud company goes out [7:28] and gets a contract where I have a client, this  offtake contract where this guy is going to pay me [7:34] $5 million a month for this amount of compute for  three years. And I've got the contract signed, [7:39] ready to go. Now, I need to go buy this much  hardware to fulfill this contract. So, they go [7:45] to the bank, hey, I need to buy $500 million worth  of hardware. And the bank says, who are you? You [7:49] know, like your website just showed up yesterday.  So, they're going to give them $500 million right [7:54] now. They know that for the next three years  they're going to get $5 million, $10 million, [7:58] whatever per month in payments. So, they don't  have the credit to go borrow $500 million. [8:06] You could collateralize a loan or do SPV, whatever  they call them. You guys read it in the news, [8:12] but the easiest way to do it may be an FMV  lease where the leasing company buys the stuff [8:18] and they lease it to them. Now, their their  contract is they're going to get $5 million [8:23] a month in payments. Their lease payment on the  hardware, maybe it's 2 million, three million. [8:29] So, they're they can afford  the payments every month, [8:33] but they can't afford to go buy all that  equipment at once and write a check to DA. So [8:40] because that's the model that's attractive  and realistic for a lot of these companies, [8:48] then you have to know the residual value  because the bank doesn't want to lend money, [8:52] especially that much money where they're worried  about getting upside down in that lease, right? [8:59] They want to know what's the depreciation of this  that cost $500 million new. What's it going to be [9:04] worth over the next three years? and at the end  of the lease, how much is it going to be worth? [9:09] But the bank's not an expert in that. They don't  know. The Neocloud's out here selling compute [9:14] contracts. So, a life cycle company can come in  and say, "Hey, this is what we do. We deal in used [9:20] equipment." 5 years ago, it was old Dell and HP  servers, but nowadays a good life cycle partner [9:27] will be an expert in the residual value of  AI hardware. So they can say, "Hey, I think [9:32] in three years these $500 million of Blackwell  GPUs is going to be worth this much money." Now [9:40] you know how to structure the lease. You know  what the payments are every month and you can [9:44] project the residual value. So uh we've gone now  for a ITAG company who deals with recycling back [9:53] up to one that helps with strategy and now back  up to one that can even help you with purchasing. [9:59] So there there is risk, right? There's risk  in what could happen to the market. Is there [10:06] a bubble? Uh will Nvidia slash the price of their  hardware, which is definitely going to affect the [10:13] price of the used hardware. Um workload shifts,  regulations about where stuff can go. Right now, [10:20] you can't just sell it to China. So, but what  if that changes? Now there's a huge market with [10:26] a huge demand that's going to up the value  of all the used stuff. So there's a lot of [10:30] I mean that's not necessarily a risk, right?  But there is imagine in reverse. There's other [10:34] regulations or things that could impact the  residual value. But like with any business, [10:40] the bank's taking some risk, the Neocloud's taking  some risk, the life cycle partner takes some risk [10:45] altogether. It's not too much risk  for any one entity that way. Now, um, [10:55] that helps with the financing. It  helps make the financing possible [10:59] because the bank's not going to  want to take too much risk. Um, [11:05] yeah, this is the right slide for this. So, let's  go back to the scenario where a Neocloud wants to [11:11] buy. They have the offtake contracts. Now, they  need to get the hardware. So, they go to a a [11:16] lender. Somebody's going to finance it  or lease it to them. And that company [11:21] has their books. They have wherever they're  getting their capital, their money from, [11:25] they will have certain constraints  to that capital. You know, no bank [11:31] is going to exclusively just take all their  capital and lease it to one asset class. [11:37] That's risky for them. They like to diversify.  But when you talk about the dollars for this [11:46] this much money, that's a lot for any bank.  Even the bi biggest banks in the world [11:52] will pay attention when you're talking about 100  million, 500 million or a billion dollars. So [11:56] they're not going to want to take that much risk  in one single asset class. So what's happening is [12:06] it'll start to constrain them. They do a couple  of loans or leases like that. Even the private [12:11] equity and other people, they don't want to be  all in on one asset class. So their decision [12:16] makers will see, hey, we did a couple of leases  and AI stuff we don't want to do anymore. But [12:22] people are knocking on their door every day  with great contracts, great off take contracts, [12:28] new neoclouds with wonderful business models,  but they're bumping up against their constraints [12:33] that they can't go into that asset class  anymore. So what's happening now is um [12:41] a company like us, a real asset partner, a life  cycle partner who's looking at these business [12:47] models, we're willing to take that risk. So we  can even go to the leasing companies either at [12:53] the beginning of the lease or in the middle of  the lease and say, "Look, let's take that risk off [12:58] your balance sheet." And they can even restructure  those leases where they're selling the residual [13:02] value. Take it off their balance sheet. We'll  buy it because we're comfortable with that risk. [13:08] Residual value used equipment. We'll buy it. So  they're not carrying that risk on their internal [13:13] balance sheets anymore. So that frees them up to  do more business in that asset class again. So [13:21] some of the exciting things happening and  this is all new, right? this didn't really [13:26] this kind of business model  this kind of leasing structure [13:29] this is all new just in the last few years as  AI's taken off and people are are realizing [13:34] that the old models for leasing and lending didn't  really work for the needs of the market anymore. [13:42] So whether it's at origination when you're talking  about how to get the material strategize about its [13:51] life cycle and the economic model through the  middle of the lease when maybe you're even just [13:56] talking about maintenancing it or redeploying  it moving it whatever you need to do the end [14:02] of the lease when you need to go physically get  the hardware out. So imagine if a Neocloud gets [14:08] I mean it's I don't think anybody would be  surprised if if I said there's probably going to [14:12] people are expecting there to be consolidation  in this industry right so let's say two Neoclouds [14:17] merge or Google or Amazon buys one now they've  got to merge their IT infrastructure so now in [14:24] the middle of a lease you've got super high value  of hardware at a data center somewhere that maybe [14:33] they don't want that anymore they say get rid of  we've merged our operations. This we don't need. [14:38] Now you've got to even who's going to go get it?  Who's going to take responsibility for the data? [14:43] Who's going to remarket it? How are you going to  do all of those things? So all the way through [14:48] the final disposition, uh there's really an  emerging market, a a bigger need. Everyone from [14:55] the the the Neoclouds or the people who are  operating the hardware, they need the financing to [15:02] make their business model successful. the people  selling the hardware for Super Micro, Dell, HP, [15:09] the salesman wants the deal to happen. So, people  that facilitate the financing are important to [15:13] him. Um, so all the way through from the from the  start to the end of this devices life cycle, uh, [15:21] a partner who can bring this expertise to the deal  is something that makes everybody's life easier. [15:27] Uh, something that'll that is kind of needed  all the way through the life cycle. Now, [15:31] when you're talking about this kind  of specific AI GPU hardware, so [15:37] that's it. And just just quickly, the  name of the company, uh, Circadus, [15:43] because what you do with the end of  something at the end of its life cycle, [15:49] good business practice is that you want to you  want to reuse as much as possible. You don't want [15:54] to waste resources. So reusing all the components  inside that even when it's not I mean right now [16:00] GPUs will get resold and refurbished and reused  for a few more years but even old servers [16:06] the old RAM the old CPUs those things  all are in demand still even DDR3 RAM [16:13] still gets reused in new builds of new  electronics in certain parts of the world [16:18] so we provide a lot of those services too that's  from the old just traditional ITA business [16:25] So, a circularity goals there, a cadence when it  comes to upgrade cycles, and even the bug cicas, [16:32] they have a multi-year life cycle. So, we put it  all together. We came up with the name Circadus [16:38] to kind of encompass all of those concepts  for what we do with hardware life cycles. So, [16:44] that's it. Thank you. I think the  audience doubled in size to 10 people.