Fluence Launches GPU Cluster Auctions to Bypass AI Hardware Brokers

Fluence GPU Cluster Auctions: Solving the Pricing Problem With Fluence — Ai4

Ai4

Auction-based marketplaces eliminate intermediary broker markups on AI compute, establishing transparent clearing prices while allowing data center operators to award capacity based on creditworthiness and SLAs rather than the highest bid.

AI-assisted summary of Fluence GPU Cluster Auctions: Solving the Pricing Problem With Fluence — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by Ai4 on August 15, 2026, this conference session features Justin Jose, Head of Business Development at Fluence, discussing the bottlenecks of reserved GPU procurement. The presentation outlines why the AI market needs a transparent pricing mechanism and demonstrates how Fluence GPU Cluster Auctions introduce competitive bidding to streamline infrastructure access for buyers and providers.

## TL;DR The presentation addresses how hyperscaler dominance and multi-layered broker markups have created a fragmented, opaque marketplace for reserved AI compute. Fluence solves this by introducing forward and reverse auction mechanisms that enable transparent price discovery across Tier 3 and Tier 4 data center inventory, including NVIDIA Hopper and Blackwell clusters. Rather than relying solely on highest monetary bids, the platform balances pricing against SLAs, latency, MTTR, and creditworthiness through masked matching and direct peer-to-peer contracts. Managing over 6,500 live GPUs globally, Fluence operates as a real-time pricing index that eliminates secondary margin stacking.

## Atmosphere & style The presentation maintains a polished, pragmatic, and authoritative tone suited for an industry conference audience of enterprise buyers and infrastructure operators. Justin Jose delivers a structured, business-oriented walkthrough that methodically diagnoses systemic supply-chain friction before presenting Fluence's auction mechanics as a practical remedy. The session feels direct, informative, and market-savvy, focused entirely on the commercial realities and operational trade-offs of securing high-performance AI compute.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Hyperscale cloud providers have spent over $500 billion securing priority GPU access, leaving other AI teams facing severe market fragmentation. - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Traditional GPU procurement is weighed down by opaque broker networks that add multi-tier markups and obscure baseline rates and operator identities. - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Fluence GPU cluster auctions aggregate verified data centers, SLAs, MTTR, and contracts under a single transparent platform. - [02:07](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=127s) In the current supply-constrained market, forward auctions let qualified buyers bid upward from a reserve price on upcoming cluster capacity. - [02:07](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=127s) Fluence supports reverse auctions for future supply-surplus cycles, allowing buyers to post requirements and hardware providers to bid downward. - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Clearing decisions factor in operational trade-offs—such as uptime SLAs, latency, and MTTR—alongside raw bid pricing. - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Available inventory spans Tier 3 and 4 data centers across NVIDIA Hopper and Blackwell architectures (H100, H200, B200, B300, and GB300). - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Suppliers retain full discretion over bidder selection based on creditworthiness and use case, meaning the highest monetary bid is not guaranteed to win. - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Bidding operates under masked identities until transactions finalize, protecting competitive market positioning and preventing speculative reselling. - [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) The platform currently hosts over 6,500 live GPUs across North America, Europe, and APAC, with upcoming expansions into LATAM and the Middle East. - [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) Direct peer-to-peer contracting between buyers and operators establishes an open, real-time GPU pricing benchmark across hardware models. - [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) Fluence has 6 to 7 new Blackwell B300 clusters scheduled to come online within 12 to 15 weeks.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) The Broken GPU Economy: Fragmented Supply, Opaque Pricing, and Manual Workflows

High-performance compute has become the vital heartbeat of artificial intelligence, yet accessing reserved GPU capacity remains deeply dysfunctional. Hyperscalers have poured over $500 billion into securing direct, priority access to hardware, leaving everyone else scrambling for limited scraps scattered unevenly across the globe. Teams attempting to lease GPU clusters face severe fragmentation—finding a fraction of a cluster in Asia, another piece in North America, and some nodes in Europe, with no unified index to determine what constitutes a fair market clearing price.

Transactions today are plagued by layers of brokers, where hardware passes through multiple intermediaries who each mark up the price before it reaches the end user. Buyers are left in the dark about who the actual data center operator is, what baseline pricing was originally offered, or whether their negotiated rate is competitive. To eliminate these manual workflows and opaque negotiations, Fluence launched its GPU cluster auctions platform. The platform brings the entire stack—verified suppliers, tier-rated data centers, transparent pricing, service level agreements (SLAs), and mean time to repair (MTTR)—under one single hood.

- **Hyperscaler Dominance:** Over $500 billion invested by hyperscale cloud providers to lock down priority GPU hardware access. - **Core Market Pain Points:** Fragmented global supply, opaque pricing with no benchmark clearing rates, and friction-filled manual broker workflows. - **Platform Scope:** Direct integration of verified data centers, transparent SLAs, MTTR metrics, and contractual master service agreements (MSAs).

---

[02:07](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=127s) Mechanics of Forward and Reverse Auctions for GPU Capacity

To accommodate shifting market dynamics, Fluence structured its marketplace around two distinct auction models: forward auctions and reverse auctions. In today’s supply-constrained environment, suppliers hold the upper hand. When upcoming capacity is scheduled to come online—such as a 128-node cluster slated for delivery 10 weeks out—Fluence vets the supplier, facility, SLAs, and technical specifications before listing the cluster with a base reserve price. Qualified, pre-vetted bidders then compete upward in real time, giving buyers complete visibility into where their bids sit relative to the baseline and whether they currently hold the winning position, all while preserving the provider's margins without intermediary markups.

The marketplace is also engineered for future cycles where GPU supply may eventually outpace demand. In a market with surplus capacity, the system switches to reverse auctions. In this scenario, a qualified enterprise buyer posts their compute requirements and parameters to the marketplace, prompting hardware providers to compete by bidding lower prices to service the demand. Whether capacity is scarce or abundant, the platform enables transparent price discovery rather than backchannel bargaining.

```mermaid flowchart TD accTitle: Fluence Dual Auction Model Flowchart

subgraph MarketDynamics["Market Conditions"] Constrained["Supply-Constrained Market<br/>(Current Phase)"] Surplus["Supply-Surplus Market<br/>(Future Phase)"] end

subgraph ForwardAuction["Forward Auction Flow"] S1["Supplier lists vetted cluster<br/>(e.g., 128-node cluster)"] --> B1["Qualified buyers bid upwards<br/>from base price"] B1 --> M1["Supplier evaluates bids based on<br/>price, use case & creditworthiness"] end

subgraph ReverseAuction["Reverse Auction Flow"] D2["Buyer posts compute requirements<br/>& SLA criteria"] --> S2["Hardware providers bid downward<br/>to win workload"] S2 --> M2["Buyer awards contract directly<br/>to optimal provider"] end

Constrained --> ForwardAuction Surplus --> ReverseAuction ```

---

[03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Beyond Price: SLAs, Cluster Lineup, and Bidder Matchmaking

Price alone does not dictate the clearing of an auction on Fluence. A realistic evaluation of high-performance compute requires balancing the bid amount against operational reliability and geographic suitability. A lower clearing price may mean sacrificing uptime—dropping from a 99.99% SLA to 99%—or accepting higher latency and longer MTTR windows. The platform lays out these exact metrics so buyers can make calculated trade-offs for their specific training or inference workloads.

Fluence features active inventory spanning NVIDIA Hopper and Blackwell architectures, including H100s, H200s, B200s, and B300/GB300 series sourced directly from Tier 3 and Tier 4 data centers. The matchmaking process operates with masked identities until a transaction is finalized to protect competitive positioning. Crucially, suppliers are not forced to accept the highest monetary bid: "The highest bid is not guaranteed to get the supply." Facility operators retain full discretion to evaluate the creditworthiness, corporate standing, and technical use case of the bidder before signing. Simultaneously, buyers undergo stringent legal and credit checks to guarantee that contracts are executed with verified end users rather than speculative resellers.

- **Available Architectures:** NVIDIA H100, H200, Blackwell B200, B300, and GB300 clusters. - **Infrastructure Standards:** Sourced directly from Tier 3 and Tier 4 data centers with enforced SLAs and MTTR parameters. - **Discretionary Clearing:** Providers select winning bids based on comprehensive criteria including financial stability, reputation, and workload compatibility, rather than purely top-line price. - **Identity Privacy:** Counterparty identities remain masked throughout the bidding process until the match is awarded and direct contracting begins.

---

[05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) Real-Time Management, Scale, and an Open Pricing Index

Both suppliers and buyers manage their transactions via a unified dashboard that tracks compute requests, incoming bids, active contracts, and matchmaking status in real time. For providers offloading expiring customer contracts or commissioning new compute racks, listing a cluster pushes immediate notifications to 10 to 20 qualified institutional buyers at once. Today, Fluence hosts more than 6,500 live GPUs spanning North America, Europe, and the Asia-Pacific region, with plans underway to expand into emerging data center hubs across the Middle East and Latin America (LATAM).

By facilitating direct peer-to-peer contracting between buyers and infrastructure operators—bypassing brokers and without Fluence acting as an intermediary reseller—the platform functions as a real-time GPU pricing index for the broader AI sector. This index establishes transparent valuation across hardware models from NVIDIA to AMD, granting enterprises visibility into true uninflated rates. Looking ahead, Fluence is bringing six to seven new Blackwell B300 clusters online within the next 12 to 15 weeks, inviting organizations at the conference to initiate cluster requests online or connect in person at booth 1251.

- **Current Footprint:** 6,500+ live GPUs distributed across North America, Europe, and APAC, with planned expansion into Latin America and the Middle East. - **Direct Contracting:** Buyers execute master service agreements directly with data center operators, eliminating secondary margin stacking. - **Pipeline Deployment:** 6 to 7 new B300 clusters scheduled to go live over the next 12 to 15 weeks. - **On-Site Presence:** Accessible via the platform QR code and booth 1251 at the event.

## Notable quotes > Reserved GPU capacity has become one of the biggest bottlenecks for AI teams, yet accessing it still relies on slow, opaque negotiations. — Justin Jose

> [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) High-performance compute has become the vital heartbeat of artificial intelligence, yet accessing reserved GPU capacity remains deeply dysfunctional. — Justin Jose

> [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Hyperscalers have poured over $500 billion into securing direct, priority access to hardware, leaving everyone else scrambling for limited scraps scattered unevenly across the globe. — Justin Jose

> [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Transactions today are plagued by layers of brokers, where hardware passes through multiple intermediaries who each mark up the price before it reaches the end user. — Justin Jose

> [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) The platform brings the entire stack—verified suppliers, tier-rated data centers, transparent pricing, service level agreements (SLAs), and mean time to repair (MTTR)—under one single hood. — Justin Jose

> [02:07](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=127s) Whether capacity is scarce or abundant, the platform enables transparent price discovery rather than backchannel bargaining. — Justin Jose

> [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) A realistic evaluation of high-performance compute requires balancing the bid amount against operational reliability and geographic suitability. — Justin Jose

> [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) The highest bid is not guaranteed to get the supply. — Justin Jose

> [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) Facility operators retain full discretion to evaluate the creditworthiness, corporate standing, and technical use case of the bidder before signing. — Justin Jose

> [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) By facilitating direct peer-to-peer contracting between buyers and infrastructure operators—bypassing brokers and without Fluence acting as an intermediary reseller—the platform functions as a real-time GPU pricing index for the broader AI sector. — Justin Jose

## Mentioned Justin Jose, Fluence, Ai4, NVIDIA, NVIDIA Hopper, NVIDIA Blackwell, H100, H200, B200, B300, GB300, AMD

## Video description Justin Jose - Head of Business Development, Fluence

Reserved GPU capacity has become one of the biggest bottlenecks for AI teams, yet accessing it still relies on slow, opaque negotiations. This session examines why the market needs a better pricing mechanism and how Fluence GPU Cluster Auctions introduce transparent price discovery for reserved AI infrastructure. We’ll explore how competitive bidding benefits both buyers and providers, simplifies access to reserved GPU clusters, and creates a more efficient market for AI compute. Join this session to learn how auction-based infrastructure sourcing can help solve one of AI’s most pressing challenges.

[Source](https://www.youtube.com/watch?v=TnQ2J6dgdsI)

中文简报

Fluence推出GPU集群拍卖以绕过AI硬件中间商 基于拍卖的市场消除了AI算力领域的中间商层层加价，建立起透明的清算价格，同时允许数据中心运营商依据信用度与SLA而非单纯依据最高出价来分配算力。 本内容探讨了超大规模云厂商的主导地位与多层经纪商加价如何导致预留AI算力市场高度碎片化且缺乏透明度。Fluence通过引入正向与反向拍卖机制解决了这一痛点，在涵盖NVIDIA Hopper和Blackwell集群等Tier 3与Tier 4数据中心库存中实现透明的价格发现。该平台并非单纯依赖最高出价，而是通过匿名撮合与点对点直接签约，在价格与SLA、延迟、MTTR及信用度之间取得平衡。Fluence在全球管理着超过6,500块在线运行的GPU，作为一个实时定价指数运作，有效消除了二级市场的多层利润叠加。 超大规模云厂商已投入超过 5000 亿美元锁定优先 GPU 算力，导致其他 AI 团队面临严重的资源碎片化与多层中介加价困境；Fluence 为此推出 GPU 集群拍卖平台，将认证数据中心、SLA、MTTR 及标准化合同整合于统一平台。 (0:04) 平台构建了灵活的双向拍卖机制：在当前供应受限周期中，合格买家围绕基准底价开展顺向竞价；而在未来的供求过剩周期中，则支持买方发布算力参数、供应商向下压价竞标的逆向拍卖。 (2:07) 最终出清决策综合权衡价格、在线时长 SLA、延迟和修复时间（MTTR），目前上线了来自 Tier 3 与 Tier 4 数据中心的 NVIDIA H100、H200 及 Blackwell B200、B300、GB300 集群；供应商享有根据买家信用及技术用例挑选客户的完全裁量权，最高出价并非绝对胜出。 (3:34) Fluence 现已在北美、欧洲及亚太地区聚合了超过 6,500 块活跃 GPU，并计划在未来 12 至 15 周内交付 6 至 7 个全新 Blackwell B300 集群，促成买家与运营商直接签署 MSA 协议以剥离中间商溢价。 (5:33)

Reading summary

Fluence has launched a GPU cluster auction platform designed to bypass opaque broker networks and establish open market pricing in an artificial intelligence hardware landscape heavily dominated by hyperscalers. Structural Fragmentation and Broker Intermediary Markups (0:04) According to Fluence, hyperscale cloud providers have invested over $500 billion to lock down priority access to high-performance GPUs, forcing other organizations to assemble compute from fragmented clusters dispersed across Asia, North America, and Europe. This scarcity is intensified by traditional procurement models reliant on layers of brokers who mark up hardware without revealing baseline costs, operational data center identities, or market benchmark rates. To counter these opaque manual negotiations, Fluence centralizes verified Tier 3 and Tier 4 data center inventory under direct contractual master service agreements. By standardizing technical disclosures, mean time to repair (MTTR) metrics, and service level agreements (SLAs), the platform aims to eliminate intermediary margins and restore visibility to foundational leasing costs. Dual Auction Mechanics for Variable Compute Cycles (2:07) The platform utilizes two distinct auction structures to adapt to shifts in hardware availability. In the current supply-constrained market, Fluence runs forward auctions: after vetting an operator's facility and specifications for upcoming capacity—such as a 128-node cluster available 10 weeks out—the system sets a reserve price. Pre-vetted institutional buyers then submit real-time upward bids against full visibility of the baseline and competing positions. Fluence also engineered reverse auctions for anticipated future cycles where GPU supply may outpace enterprise demand. Under surplus conditions, qualified buyers post specific workload and SLA parameters to the exchange, prompting infrastructure operators to bid prices downward to secure contracts. Both models are intended to replace backchannel bargaining with transparent, real-time price discovery. Operational Reliability and Discretionary Bid Matchmaking (3:34) Clearing an auction on Fluence does not depend solely on the monetary value of a bid. The platform requires buyers to weigh pricing against operational reliability, as lower clearing rates can entail accepting an uptime drop from 99.99% to 99%, longer MTTR windows, or elevated latency across workloads running on NVIDIA Hopper and Blackwell systems (including H100, H200, B200, and B300/GB300 architectures). Crucially, data center operators retain full discretion over awards, with Fluence noting that the highest monetary bid is not guaranteed to secure capacity. Suppliers evaluate bidders based on corporate standing, creditworthiness, and technical use cases. To prevent speculative reselling and safeguard market strategy, counterparty identities remain masked during bidding until contracts are executed directly with verified end users. Current Infrastructure Scale and Benchmark Indexing (5:33) Fluence reports hosting more than 6,500 live GPUs across North America, Europe, and the Asia-Pacific region, alongside expansion plans targeting emerging data center hubs in the Middle East and Latin America. Listing unassigned compute notifies 10 to 20 pre-qualified buyers simultaneously, accelerating placement for facilities opening new racks or re-leasing expiring customer contracts. By facilitating direct peer-to-peer contracting without acting as a reseller, Fluence positions its platform as an independent valuation index for NVIDIA and AMD hardware across the AI sector. The company anticipates deploying six to seven new Blackwell B300 clusters onto the platform within the next 12 to 15 weeks. While Fluence introduces a structured mechanism to counter broker markups and establish clearing benchmarks, the marketplace's viability as an industry-wide pricing index will depend on sustaining liquidity across both sides of the exchange amid intense hyperscaler competition for leading-edge silicon.

精炼阅读摘要

针对超大规模云服务商垄断资源以及中介转售推高成本的现状，Fluence推出了GPU集群拍卖平台，试图通过前向与反向竞拍机制及点对点直签模式，解决算力市场的供给碎片化与定价不透明问题。 算力供给碎片化与传统中介弊端 (0:04) 在人工智能算力需求激增的背景下，超大规模云服务商已投入超过5000亿美元锁定优先硬件访问权，导致其余机构只能争夺零散分布于北美、欧洲和亚洲的配额。由于缺乏统一的市场出清基准，现有算力交易高度依赖多层中间经纪商，硬件在多次转手中被层层加价，买方既无法得知数据中心实际运营方的身份，也难以判断最终协议价格是否公允。为消除人工低效协商与定价黑箱，Fluence构建了GPU集群拍卖平台，将经过验证的供应商、分级数据中心、透明定价指标以及服务等级协议（SLA）与平均修复时间（MTTR）统一整合进标准化体系。 兼顾供需周期的双向竞拍模式 (2:07) 为适应动态变化的算力市场供求关系，Fluence设计了前向拍卖与反向拍卖两种互补机制。在当下算力供给受限的环境中，平台采用前向拍卖：供应商在通过对其机房设施、SLA和技术规范的严格审核后，为即将交付的集群（例如10周后上线的128节点集群）设立保留底价，预审合格的买方实时向上竞价，买方可直观掌握竞价排名，供应商也能在不被中介截留利润的前提下保障自身收益。在未来供给可能超过需求周期的假设下，平台支持切换至反向拍卖，由企业买方发布具体的算力参数与SLA指标，由各硬件提供商向下竞价竞争订单，从而依靠供需机制而非暗箱协商实现透明的价格发现。 综合服务指标与非单纯价高者得 (3:34) 在Fluence的匹配机制中，报价并非促成交易的唯一依据，算力的实际匹配需权衡资金出价、运行稳定性与地理网络延迟。较低的出清价格可能伴随着SLA从99.99%降至99%或MTTR增加的可靠性妥协，平台公开这些指标以供买方根据模型训练或推理需求做出取舍。目前平台库存涵盖来自Tier 3和Tier 4数据中心的NVIDIA Hopper与Blackwell架构，包括H100、H200、B200、B300及GB300等硬件。竞标过程中双方身份保持匿名保护；根据平台规则，最高出价并不保证能够获得算力配额，数据中心运营商保留最终裁量权，可在签约前综合评估买方的信用背景、公司资质及技术使用场景，以防止投机性转售商扰乱真实终端用户的供给。 规模扩张与点对点定价指数功能 (5:33) 目前Fluence平台已连接北美、欧洲和亚太地区超过6500张处于运行状态的GPU，并计划向中东及拉丁美洲等新兴算力枢纽延伸。供应商发布新集群或转让即将到期的客户合同时，系统可实时向10至20家合格机构买方推送通知，促成买卖双方直接签署主服务协议（MSA），Fluence自身不作为加价转售的中介。通过促成基础设施运营商与终端买方的点对点交易，该平台力图建立覆盖NVIDIA与AMD等硬件的实时行业定价指数，反映真实未加价的市场水平；此外，Fluence预计将在接下来的12至15周内陆续上线6至7个新的Blackwell B300集群。 Fluence通过引入结构化竞拍和直接签约机制，为缓解算力市场的层层加价与信息不对称提供了可操作方案；但其市场透明度能否长期确立，仍取决于未来硬件供给充裕阶段反向竞拍的参与深度，以及对多样化异构硬件建立权威定价基准的实际能力。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2026 年 8 月 15 日在 YouTube 发布，主讲人为 Fluence 业务拓展负责人 Justin Jose。视频主要围绕当前预留 GPU 算力市场存在的定价不透明、流程缓慢等瓶颈展开，介绍了 Fluence GPU 集群拍卖平台如何通过透明的竞价机制与基础设施采购模式解决 AI 行业的算力痛点。 ## TL;DR Fluence 针对当前超大规模云厂商垄断与算力中介层层加价导致的供给碎片化和定价黑盒问题，推出了专业的 GPU 集群拍卖平台。平台在当前供不应求的环境下采用正向拍卖机制，并在未来可能出现的供过于求阶段支持反向拍卖，同时覆盖 H100、B300 等高端算力。通过对 Tier 3/4 数据中心与最终买家实施严格的双向资质审核及匿名竞价，Fluence 摒弃了单纯‘价高者得’的模式，促成买卖双方直接签署协议，致力于打造全球化、透明的实时 GPU 算力定价指数。 ## Atmosphere & style 演讲整体呈现出专业、务实且直击行业痛点的商业路演风格。主讲人以清晰利落的逻辑拆解复杂的算力供需困局，语气坚定且充满说服力。观看体验如同置身一场高效的 B2B 基础设施发布会，既有对宏观算力资本博弈的冷静洞察，又展示了产品落地机制的严谨与规范。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) 超大规模云服务商投入超 5000 亿美元锁定算力，导致散布在全球的剩余 GPU 资源极度匮乏且定价不公允。 - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) 当前 GPU 算力市场存在供给碎片化、定价不透明以及人工谈判流程繁琐三大核心痛点。 - [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Fluence 推出 GPU 集群拍卖平台，通过整合真实数据中心、实时定价、SLA 与 MTTR 指标提供一站式交易机制。 - [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 平台设计了正向拍卖与反向拍卖两种模式，以动态适应算力短缺或未来可能出现的供应过剩环境。 - [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 在当前供不应求的正向拍卖中，买方可在基准底价上公开竞价，全程实时透明展示出价与领先状态。 - [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 算力定价不仅取决于单价，还综合考量 SLA 可用性等级（如 99.99% 与 99%）、地理区域及 MTTR 修复时效。 - [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 平台覆盖 NVIDIA H100、H200、Blackwell B200、B300 以及 GB300 等主流高端算力集群。 - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 实行严格双向准入背调，仅对接 Tier 3 和 Tier 4 数据中心，并审核买方信用与法人主体以杜绝黄牛倒卖。 - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 竞价过程中买卖双方身份匿名脱敏，且最终匹配并非单纯价高者得，数据中心可根据信用与业务契合度自主决定。 - [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 交易达成后买卖双方直接签署合同，Fluence 仅作为撮合与定价平台，彻底去除多层中介转手加价。 - [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) 平台目前已汇聚超 6,500 张活跃 GPU，覆盖北美、欧洲、亚太等多个国家，并上线了多个预计 12 至 15 周内交付的 B300 集群。 - [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) Fluence 的长远愿景是构建覆盖 NVIDIA、AMD 等全品类的实时 GPU 定价指数，并将供给网络扩展至中东与拉美地区。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) GPU 经济的市场困局与 Fluence 拍卖平台的诞生 在人工智能行业，高性能算力已成为绝对的核心命脉，但当下的 GPU 算力经济却处于一种极度别扭且低效的状态。全球各地的 GPU 定价极度不一致，算力资源的获取渠道也被严密把守。各大超大规模云服务商（Hyperscalers）豪掷超过 5000 亿美元以锁定算力供应，这导致散布在全球的剩余硬件资源极度匮乏，绝大多数 AI 团队根本无从得知公允的市场价格究竟是多少。 ```mermaid flowchart TD accTitle: Fluence GPU 解决的市场痛点与平台机制 subgraph PainPoints ["当前市场三大痛点"] P1["供给碎片化 (Fragmented Supply)"] P2["定价不透明 (Opaque Pricing)"] P3["人工流程繁琐 (Manual Workflows)"] end subgraph Platform ["Fluence 拍卖解决方案"] A1["正向拍卖 (Forward Auction)<br/>供不应求：买方向上竞价"] A2["反向拍卖 (Reverse Auction)<br/>供过于求：供应商向下竞价"] M1["双向严格资质背调与匿名匹配"] D1["直接与 Tier 3 / Tier 4 数据中心直签"] end PainPoints --> Platform ``` 针对这一行业痼疾，Fluence 正式推出了 GPU 集群拍卖平台，旨在彻底解决困扰行业的三大核心痛点：供给碎片化（Fragmented Supply）、定价不透明（Opaque Pricing）以及工作流高度依赖人工（Manual Workflows）。过去，买家若想租赁 GPU 集群，往往只能在亚洲、北美或欧洲拼凑碎片化资源，不仅无法获得完整的本地化集群，更没有任何基准指数来判断自己拿到的报价是否合理，不知道中间转手了多少道中介。Fluence 将真实服务商、数据中心、实时定价、服务等级协议（SLA）以及平均修复时间（MTTR）全盘整合透明化，提供了一站式的交易机制。 - **超大规模云厂商垄断**：头部巨头投入超 5000 亿美元抢夺即时算力访问权，挤压了中小团队的硬件获取空间。 - **中介转手加价**：传统模式下 GPU 往往经过多层代理商倒手，买方难以获知数据中心的原始出厂底价。 --- ### [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 机制设计：正向拍卖与反向拍卖的动态调节 在算力极其紧缺的年代，市场上充斥着各种掮客与中介，进一步加剧了资源错配。如今，不仅获取硬件本身困难重重，获取“合理价格”更是难上加难。为此，Fluence 在平台上引入了两种不同的拍卖机制——正向拍卖（Forward Auction）与反向拍卖（Reverse Auction），以便在不同的宏观供需环境下动态运行。 ```mermaid flowchart LR accTitle: 正向拍卖与反向拍卖机制对比 subgraph Forward ["正向拍卖（当前供不应求）"] S1["供应商挂出集群<br/>（如 128 节点 B300）"] --> B1["买方在底价基础上<br/>向上竞价 (Bid Upwards)"] B1 --> R1["供应商保障利润<br/>直面数据中心无中介"] end subgraph Reverse ["反向拍卖（未来供大于求）"] D2["买方发布算力需求"] --> S2["多个供应商<br/>向下竞价 (Bid Lower)"] S2 --> R2["以最优成本<br/>满足买方算力缺口"] end ``` 在当前算力短缺的市场中，平台主要运行正向拍卖。假设某数据中心在 10 周后将交付一个包含 128 个节点的 B300 集群，Fluence 会对供应商、数据中心物理设施、SLA、MTTR 及各项服务进行严密审核并上线。通过资质认证的买家可以在设定的底价基础之上公开向上竞价。整个竞价过程买家端完全实时可见，清楚了解基准底价、最高出价以及自己是否处于领先位置。反之，如果在未来几年内算力市场出现供应过剩，平台将切换为反向拍卖模式：由买方提出明确需求，持有空闲算力的各家供应商向下竞价，以极具竞争力的价格满足采购需求。 - **定价维度的综合考量**：成交价格并非唯一决定因素，结算时必须综合权衡 SLA 指标（如 99.99% 与 99% 的可用性差异）、地理区域以及 MTTR 响应时限。 - **算力型号覆盖**：平台实时上线了 NVIDIA H100、H200、Blackwell B200、B300 以及 GB300 等主流高端算力集群。 --- ### [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 严格双向准入背调与非单纯“价高者得”的匹配逻辑 为了确保算力履约的严肃性，Fluence 建立了一套严苛的双向资质验证流程。在供应端，平台仅对接 Tier 3 和 Tier 4 级别的高等级数据中心，提前审核并固化企业法律主体、主服务协议（MSA）及 SLA 指标；在需求端，平台同样会对所有买家进行严谨的信用调查与法人主体审核，查明其是否为真实的最终用户（End User），杜绝算力黄牛倒买倒卖。 在拍卖流程中，买卖双方的真实身份均处于匿名脱敏状态（Masked），直到竞价匹配完成。值得注意的是，平台并非简单的“最高价中标”模式。当正向竞价结束时，数据中心拥有最终选择权，能够综合考量出价方的信用评级状况与具体业务用例的契合度，从而自主挑选最心仪的长期合作伙伴。 ```mermaid flowchart TD accTitle: 交易审核与匿名匹配流程 subgraph Vetting ["严格双向资质背调"] V_Sup["供应商审核：<br/>Tier 3/4 DC、SLA、MTTR、MSA"] V_Buy["买方审核：<br/>信用调查、法律主体、最终用户认证"] end subgraph AuctionProcess ["匿名竞价与决策"] Mask["身份脱敏 (Masked Identities)<br/>实时透明竞拍"] Match["撮合完成，展示出价池"] Decision{"数据中心自主决策<br/>(非单纯价高者得)"} end Vetting --> Mask Mask --> Match Match --> Decision Decision --> Deal["买卖双方直接签署合同<br/>（Fluence 不介入中介抽成）"] ``` --- ### [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) 实时算力网络看板、定价指数与全球扩展蓝图 Fluence 为每位用户提供了实时动态看板，支持实时监控所有提交的算力请求与接收到的竞价状态。当供应商拥有刚走出合同期的闲置集群或新购入的硬件时，既可自主上架，也可委托平台批量录入。集群一旦发布，系统会单次直接精准推送给 10 到 20 家合格买家，帮助供应商快速锁定利润并拓展客源，同时买家也能彻底绕开“倒手三遍、层层加价”的 Hopper 算力中间商，直接与数据中心原厂签署协议。 目前，Fluence 平台上已汇聚了超过 6,500 张实时在线的 GPU，广泛分布在北美、欧洲及亚太等地区的 4 至 5 个国家。Fluence 的长远目标是将整个竞价撮合生态打造为一个实时的全品类 GPU 定价指数（Live Pricing Index），覆盖 NVIDIA、AMD 等跨品牌全线算力产品，以此作为行业价格锚点。随着全球算力基建布局的推进，Fluence 正加速将供给网络拓展至中东以及拉美（LATAM）地区。 - **实时规模**：平台现有超 6,500 块 GPU 处于活跃可竞拍状态。 - **新品集群上线**：平台已上架 6 至 7 个 B300 集群，预计在接下来的 12 至 15 周内正式上线交付。 - **无中介直签**：合同由供应商与需求方直接缔结，Fluence 仅作为撮合与定价平台，不充当中介环节。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) 在人工智能行业，高性能算力已成为绝对的核心命脉，但当下的 GPU 算力经济却处于一种极度别扭且低效的状态。 — Justin Jose > [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) 各大超大规模云服务商（Hyperscalers）豪掷超过 5000 亿美元以锁定算力供应，这导致散布在全球的剩余硬件资源极度匮乏，绝大多数 AI 团队根本无从得知公允的市场价格究竟是多少。 — Justin Jose > [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Fluence 正式推出了 GPU 集群拍卖平台，旨在彻底解决困扰行业的三大核心痛点：供给碎片化（Fragmented Supply）、定价不透明（Opaque Pricing）以及工作流高度依赖人工（Manual Workflows）。 — Justin Jose > [00:04](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=4s) Fluence 将真实服务商、数据中心、实时定价、服务等级协议（SLA）以及平均修复时间（MTTR）全盘整合透明化，提供了一站式的交易机制。 — Justin Jose > [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) Fluence 在平台上引入了两种不同的拍卖机制——正向拍卖（Forward Auction）与反向拍卖（Reverse Auction），以便在不同的宏观供需环境下动态运行。 — Justin Jose > [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 如果某数据中心在 10 周后将交付一个包含 128 个节点的 B300 集群，Fluence 会对供应商、数据中心物理设施、SLA、MTTR 及各项服务进行严密审核并上线。 — Justin Jose > [01:42](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=102s) 反之，如果在未来几年内算力市场出现供应过剩，平台将切换为反向拍卖模式：由买方提出明确需求，持有空闲算力的各家供应商向下竞价。 — Justin Jose > [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 在供应端，平台仅对接 Tier 3 和 Tier 4 级别的高等级数据中心，提前审核并固化企业法律主体、主服务协议（MSA）及 SLA 指标。 — Justin Jose > [03:34](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=214s) 平台并非简单的“最高价中标”模式。当正向竞价结束时，数据中心拥有最终选择权，能够综合考量出价方的信用评级状况与具体业务用例的契合度。 — Justin Jose > [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) 买方也能彻底绕开“倒手三遍、层层加价”的 Hopper 算力中间商，直接与数据中心原厂签署协议。 — Justin Jose > [05:33](https://www.youtube.com/watch?v=TnQ2J6dgdsI&t=333s) Fluence 的长远目标是将整个竞价撮合生态打造为一个实时的全品类 GPU 定价指数（Live Pricing Index），覆盖 NVIDIA、AMD 等跨品牌全线算力产品，以此作为行业价格锚点。 — Justin Jose ## Mentioned Justin Jose, Fluence, Ai4, Hyperscalers, NVIDIA, AMD, NVIDIA H100, NVIDIA H200, Blackwell B200, Blackwell B300, GB300, Hopper ## Video description Justin Jose - Head of Business Development, Fluence Reserved GPU capacity has become one of the biggest bottlenecks for AI teams, yet accessing it still relies on slow, opaque negotiations. This session examines why the market needs a better pricing mechanism and how Fluence GPU Cluster Auctions introduce transparent price discovery for reserved AI infrastructure. We’ll explore how competitive bidding benefits both buyers and providers, simplifies access to reserved GPU clusters, and creates a more efficient market for AI compute. Join this session to learn how auction-based infrastructure sourcing can help solve one of AI’s most pressing challenges. [Source](https://www.youtube.com/watch?v=TnQ2J6dgdsI)

Source quotes

[0:04] Justin Jose: 在人工智能行业，高性能算力已成为绝对的核心命脉，但当下的 GPU 算力经济却处于一种极度别扭且低效的状态。 [0:04] Justin Jose: 各大超大规模云服务商（Hyperscalers）豪掷超过 5000 亿美元以锁定算力供应，这导致散布在全球的剩余硬件资源极度匮乏，绝大多数 AI 团队根本无从得知公允的市场价格究竟是多少。 [0:04] Justin Jose: Fluence 正式推出了 GPU 集群拍卖平台，旨在彻底解决困扰行业的三大核心痛点：供给碎片化（Fragmented Supply）、定价不透明（Opaque Pricing）以及工作流高度依赖人工（Manual Workflows）。 [0:04] Justin Jose: Fluence 将真实服务商、数据中心、实时定价、服务等级协议（SLA）以及平均修复时间（MTTR）全盘整合透明化，提供了一站式的交易机制。 [1:42] Justin Jose: Fluence 在平台上引入了两种不同的拍卖机制——正向拍卖（Forward Auction）与反向拍卖（Reverse Auction），以便在不同的宏观供需环境下动态运行。 [1:42] Justin Jose: 如果某数据中心在 10 周后将交付一个包含 128 个节点的 B300 集群，Fluence 会对供应商、数据中心物理设施、SLA、MTTR 及各项服务进行严密审核并上线。 [1:42] Justin Jose: 反之，如果在未来几年内算力市场出现供应过剩，平台将切换为反向拍卖模式：由买方提出明确需求，持有空闲算力的各家供应商向下竞价。 [3:34] Justin Jose: 在供应端，平台仅对接 Tier 3 和 Tier 4 级别的高等级数据中心，提前审核并固化企业法律主体、主服务协议（MSA）及 SLA 指标。 [3:34] Justin Jose: 平台并非简单的“最高价中标”模式。当正向竞价结束时，数据中心拥有最终选择权，能够综合考量出价方的信用评级状况与具体业务用例的契合度。 [5:33] Justin Jose: 买方也能彻底绕开“倒手三遍、层层加价”的 Hopper 算力中间商，直接与数据中心原厂签署协议。 [5:33] Justin Jose: Fluence 的长远目标是将整个竞价撮合生态打造为一个实时的全品类 GPU 定价指数（Live Pricing Index），覆盖 NVIDIA、AMD 等跨品牌全线算力产品，以此作为行业价格锚点。

Source quotes

Justin Jose: Reserved GPU capacity has become one of the biggest bottlenecks for AI teams, yet accessing it still relies on slow, opaque negotiations. [0:04] Justin Jose: High-performance compute has become the vital heartbeat of artificial intelligence, yet accessing reserved GPU capacity remains deeply dysfunctional. [0:04] Justin Jose: Hyperscalers have poured over $500 billion into securing direct, priority access to hardware, leaving everyone else scrambling for limited scraps scattered unevenly across the globe. [0:04] Justin Jose: Transactions today are plagued by layers of brokers, where hardware passes through multiple intermediaries who each mark up the price before it reaches the end user. [0:04] Justin Jose: The platform brings the entire stack—verified suppliers, tier-rated data centers, transparent pricing, service level agreements (SLAs), and mean time to repair (MTTR)—under one single hood. [2:07] Justin Jose: Whether capacity is scarce or abundant, the platform enables transparent price discovery rather than backchannel bargaining. [3:34] Justin Jose: A realistic evaluation of high-performance compute requires balancing the bid amount against operational reliability and geographic suitability. [3:34] Justin Jose: The highest bid is not guaranteed to get the supply. [3:34] Justin Jose: Facility operators retain full discretion to evaluate the creditworthiness, corporate standing, and technical use case of the bidder before signing. [5:33] Justin Jose: By facilitating direct peer-to-peer contracting between buyers and infrastructure operators—bypassing brokers and without Fluence acting as an intermediary reseller—the platform functions as a real-time GPU pricing index for the broader AI sector.

Original transcript

[0:04] Graph all uh in the next 10 minutes  we are going to try and dissect the [0:10] awkward GPU economy out there. So the GPU market  is what we're all here for. All of his work in [0:16] the AI space. We need a lot of high performance  computing power and GPU is the heartbeat of it. [0:22] In this current day market uh GPUs are priced  differently across spread across different regions [0:29] in the world and access to it is heavily guarded.  Hyperscalers have poured over $500 billion to get [0:35] easy instant access to these GPUs and uh what does  that leave us with? We have very little hardware [0:42] spread across the world and most of us do not have  what a fair price is for any of these GPUs. And [0:49] with that intention, we've launched GPU auctions  platform on Fluence and told me that platform [0:54] is intended to address three different issues  that PC in the market. Yeah. Fragmented supply, [1:01] opaque pricing and manual workflows. So if you  are in the market trying to lease out GPUs in the [1:07] space, you would see that you wouldn't get a full  cluster across the regions that you want them. [1:12] You probably get them some in Asia, some in North  America, some in Europe. And you really don't have [1:17] any index to see if the price that you get is  actually a fair price or not. What's the clearing [1:22] price? How many hands have these GPUs flown in  through? Who is the provider at the end of it all? [1:28] So the auctions platform is clearly defining  all of this for you wherein you get the right [1:34] provider, the right data center, the right  pricing, the SLAs, the MTR all under one hood. [1:42] uh a couple of years back uh when it was difficult  to get the GPUs you had the brokers in the space [1:49] what they added into the market is your lack of  access to the right GPUs the right hardware in the [1:54] right places now it's not easy to get GPUs anymore  but getting the right price for GPUs is becoming [2:01] even more tougher so what we do at Fluence  with the auctions platform is we've created [2:07] two different auctions the forward auctions and  the reverse auctions in the current day market [2:12] supply is limited. So suppliers have the higher  hand uh to go in the market and pick the right [2:19] demand for it. So if there's a good supply  coming in the market, let's assume there's a [2:24] 128 node B300 cluster coming into the market  say 10 weeks from now. We vet the supplier, [2:30] we wet the DC, we vet the SLAs, the MTR,  the services and list them on our platform. [2:36] We also wet the qualified biders for it and the  biders are free to choose and bid upwards from the [2:41] base price for all clusters. Thus, what happens  is for each supplier, your margins are safe. [2:47] You directly deal with the data center and there's  no middleman involved. And for all the biders, you [2:53] can see at real time what your bidding prices are.  So, if you're a client, you can see exactly where [2:59] your baseline is, what the maximum is, and if  you're winning the bid or not. On the other hand, [3:04] if there comes a day in a couple of years from  now where you have a lot of supply on the market, [3:09] it would reverse the whole process. It would be a  reverse auction wherein the demand side of things. [3:14] A qualified buyer less demand on the marketplace  and each supplier who has the supply for it [3:21] can bid a lower price expect ensuring that  the demand is fairly serviced. So the market [3:26] can go both ways on our platform depending on  the supply demand scarcity in the current day. [3:34] So what we do is we list out all supplies that we  get all verified buyers and list them in metrics [3:40] that you can see and define it. So the price  is not always the clearing factor. You should [3:47] factor in what the SLAs are. Do you get a 99.99  SLA or do you get a 99 SLA? Is it in the region [3:53] that you wanted? Is it is the MTR within uh  the range that you want for your use case? So, [4:00] not always the case that the  clearing price is the lowest [4:04] and that's something that all of us have to factor  in when we start bidding for GPUs in the market. [4:11] Here's a list of GPUs that we have on our  platform. They're all live. They're all for you to [4:16] go and bid on. Uh we have H100s, H200s, Blackwell  B200s, 300s, and GB300s coming live directly from [4:25] the data centers directly to your use cases.  And uh all of these have been vetted before. So [4:32] what the process for us looks like is every  supplier that creates a profile gets vetted by [4:36] us. They're qualified. Their entities, their  SLAs, their MSAs are all on our platform. [4:42] All the identities are masked until the bid is  serviced. So when we have a supplier giving you [4:48] say a black will be B300 and we have qualified  biders who have bid beyond the base price [4:54] and when the matchmaking is done the supplier  then gets to pick from the best bids out there. [5:00] The highest bid is not  guaranteed to get the supply. [5:04] The data center at that point has the higher hand  to choose the creditworthiness of the client, [5:08] the right use case for the client and then pick  who they want to deal with. On the other hand, [5:14] the qualified buyers, we vet all the buyers on the  platform for their credit checks, for their legal [5:19] entities, where are they placed, are they actually  going to be the end user or would there be another [5:23] end user down the line from them and get them on  the platform. So, rest assured on the demand side [5:29] everything is checked. On the supply side, you  have tier three, tier four DCs working with us, [5:33] great GPU providers out there. There's a dashboard  which every profile has. Inside the dashboard you [5:40] could see all the compute requests that you've  placed, all the bid requests that you've received. [5:44] You can do all the matchmaking and pick your  demand or supply from these dashboards. And this [5:49] are updated in real time. So everything that you  see on the platform is available as of now today. [5:56] If you have fresh clusters coming out of contracts  or if you're buying new clusters, you could always [6:01] list them, add them manually or have it sent  over to us. So we would list them for you and [6:06] each cluster listing gets pushed off to all  qualified buyers in one shot. So if you have [6:11] a new cluster coming in, rest assured every  listing goes to 10 to 20 odd different buyers [6:17] getting you the right price at the right margins  safeguarding your future as well. At this point, [6:24] we have in excess of 6,500 GPUs live across four  or five different countries, different continents. [6:32] Uh, North America, Europe, and Asia Pacific is  where we have the largest suppliers coming in. [6:39] Why is it that we need this platform at this  point? For suppliers, it's critical that you [6:44] save your margins. Your single listing reaches  everyone that needs to know about those listings. [6:50] You get the right kind of use cases attracted  to yourself. You have the right companies with [6:54] the right credit profiles coming to you. For  buyers, you see the real market pricing. As in [6:59] at this point, it's safe to say that when you buy  a Hopper series model, it may have gone through [7:05] three different hands adding their own margins to  it. And you absolutely have no clue what the data [7:10] center pricing was. So you get the real deal at  the real pricing and you contract directly with [7:15] the supplier and the demand, not through any  middleman involved, not with fluence either. [7:22] What we could do with this is turn the whole  GPU bidding process into a live pricing index. [7:28] What this pricing index gives you is the right  to know exactly what the real-time pricing are [7:33] for all models whether it be Nvidia's AMDs and  across the board. Now this can be turned into a [7:40] fullborn outbound series. you can buy, attract  more clients, attract more suppliers to you [7:46] and the next step for us is to ensure that we have  supply across all continents. So we do not have [7:51] much supply in in the Middle East in LATAM but  sure that we are heading towards that bit where [7:56] we see a lot of data setters coming in over there.  Here is a link to our QR which is to the auctions [8:02] platform. Feel free to hit us up. We are there in  booth number 1251. We have launched at least six [8:10] to seven clusters of B300 in on our platform which  goes live in the next 12 to 15 weeks. So either [8:17] you can post your request on the platform or hit  us up at route number 1251. Right. Thank you.