{"id":77922,"date":"2026-06-17T20:21:41","date_gmt":"2026-06-17T20:21:41","guid":{"rendered":"https:\/\/exittechnologies.com\/?p=77922"},"modified":"2026-06-17T20:27:15","modified_gmt":"2026-06-17T20:27:15","slug":"softbanks-100-billion-robot-data-center-bet-has-a-disposition-problem-nobodys-talking-about","status":"publish","type":"post","link":"https:\/\/exittechnologies.com\/es\/blog\/centro-de-datos\/softbanks-100-billion-robot-data-center-bet-has-a-disposition-problem-nobodys-talking-about\/","title":{"rendered":"SoftBank\u2019s $100 Billion Robot Data Center Bet Has a Disposition Problem Nobody\u2019s Talking About"},"content":{"rendered":"<span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">Tiempo de lectura: <\/span> <span class=\"rt-time\"> 6<\/span> <span class=\"rt-label rt-postfix\">minutos<\/span><\/span>\n<p>A $6.5 billion compute company, a $5.4 billion robotics division, and a $3 billion infrastructure firm walk into a blank legal entity called Roze. SoftBank wants to take this combination public at a $100 billion valuation before it has shipped a single product, disclosed a revenue figure, or confirmed an IPO date.<\/p>\n\n\n\n<p>The pitch: deploy autonomous robots to build AI data centers faster than the US construction workforce can, closing a labor gap that the Information Technology and Innovation Foundation pegged at 439,000 workers as of November 2025. Satellite imagery analysis suggests up to 40% of AI data center construction sites face delays and Masayoshi Son is arguing that machines can solve a problem human recruiters cannot.<\/p>\n\n\n\n<p>The financial press is focused on the valuation question. That\u2019s the wrong question. The right question is what happens downstream. Specifically, what happens to the hardware inside all those data centers when the AI refresh cycle runs its course and the machines that built the facilities have to be replaced.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What You Need to Know<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>SoftBank is creating a new entity called Roze, targeting a $100 billion US IPO in the second half of 2026, with no product revenue, no confirmed IPO date, and acknowledged internal skepticism about the valuation.<\/li>\n\n\n\n<li>Roze will use autonomous robots to accelerate AI data center construction across the US, folding in SoftBank\u2019s $5.4 billion acquisition of ABB\u2019s robotics division, Ampere Computing for $6.5 billion, and DigitalBridge for $3 billion.<\/li>\n\n\n\n<li>If Roze succeeds, it accelerates the pace at which hyperscale AI infrastructure gets built and, by direct consequence, the pace at which that hardware gets decommissioned.<\/li>\n\n\n\n<li>AI hardware refresh cycles have already compressed from 5-7 years to 18\u201336 months. A faster build rate means a faster decommission rate. Your <a href=\"https:\/\/exittechnologies.com\/es\/servicios-de-centro-de-datos\/itad\/\" style=\"color: var(--e-global-color-primary) !important; font-weight: bolder !important;\">ITAD program<\/a> was not built for this cadence.<\/li>\n\n\n\n<li>The ITAD market is projected to grow from $18.6 billion in 2026 to $40.1 billion by 2035, driven largely by this compression.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">The Build That Roze Is Trying to Unlock<\/h2>\n\n\n\n<p>Roze would seek to make data center construction in the US more efficient by deploying autonomous robots to help build server farms. The construction of hyperscale data centers is one of the most labor-intensive undertakings in modern infrastructure, an activity that still depends heavily on skilled electricians, pipe fitters, and HVAC engineers.<\/p>\n\n\n\n<p>That dependency is the wall. A 60 MW data center delay can cost $14.2 million per month in lost revenue, and 41% of the construction workforce is expected to retire by 2031. Son\u2019s argument is that autonomous robots for welding, lifting, assembling, etc, can compress typical 24-to-36-month data center build timelines without waiting for a labor market that cannot meet demand.<\/p>\n\n\n\n<p>Roze is set to include SoftBank\u2019s recent $5.4 billion purchase of ABB\u2019s robotics division, Ampere Computing for $6.5 billion, and DigitalBridge for $3 billion. That\u2019s a portfolio of hard assets: robotics capability, compute architecture, and real estate infrastructure rolled into a single listed vehicle.<\/p>\n\n\n\n<p>The strategic hook is Stargate. SoftBank is leading financing for the Stargate project to develop data centers across the United States, a planned $500 billion undertaking pursued alongside OpenAI and Oracle. Roze is the construction arm that makes Stargate\u2019s timeline credible. Without it, Stargate\u2019s $500 billion target collides directly with the 439,000-worker shortage and the 40% delay risk on AI sites.<\/p>\n\n\n\n<p>What Roze does, if it works, is remove the physical bottleneck from AI infrastructure deployment. More facilities. Built faster. Coming online on compressed timelines.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The $100 Billion Question Nobody Has Answered<\/h2>\n\n\n\n<p>Arm Holdings, a far more mature business with decades of revenue history and an installed base in virtually every smartphone on the planet, currently trades at a market capitalization of roughly $145 billion. Pricing a company that does not yet exist at two-thirds of that figure is a bold pitch, even by the standards of the current AI investment cycle.<\/p>\n\n\n\n<p>SoftBank founder and Chief Executive Masayoshi Son is driving the effort, with executives reportedly targeting a valuation of about $100 billion, though the report noted that the valuation target and timeline could shift and that the plans are considered ambitious by some SoftBank executives.<\/p>\n\n\n\n<p>SoftBank has been known to back some dark horse startups. It notably sunk hundreds of millions of dollars into Zume, an AI-driven pizza delivery startup that went belly up in 2023. The difference this time is that the demand Roze is targeting is real. The labor shortage is documented. The construction delays are satellite-verified. Whether an autonomous robotics company can execute against that demand at hyperscale is a different question entirely.<\/p>\n\n\n\n<p>Roze has not yet announced a product, shared a revenue plan, or set an IPO date. For now, the company has a $100 billion target, an analyst day in July, and Masayoshi Son\u2019s backing. KPMG has been hired to prepare financials for the public offering, and Bilal Safeer, currently an executive at Arm, is expected to serve as interim CFO.<\/p>\n\n\n\n<p>The analyst day at a Texas data center site in July is the first real moment of accountability. That\u2019s when Son has to show something operational, not just a term sheet and a vision.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What Faster Construction Actually Does to Downstream Disposition<\/h2>\n\n\n\n<p>If Roze accelerates the pace of AI data center construction, even partially, the volume of hardware entering those facilities goes up. That sounds obvious. What\u2019s less obvious is what it means for your decommissioning calendar.<\/p>\n\n\n\n<p>AI hardware refresh cycles have compressed from the traditional four-to-six years down to two-to-three years. A mid-size enterprise running 500 AI GPU servers on a 2-to-3-year refresh cycle will process more decommissioned hardware volume per year than it previously handled across an entire 5-year HDD server lifecycle.<\/p>\n\n\n\n<p>Traditional enterprise racks often drew 5\u201310 kW per rack. AI compute racks now routinely demand 20\u201360 kW, and hyperscale environments can push 80\u2013100 kW per rack or more. New GPU designs based on Blackwell architecture are already driving peak rack densities in the 100+ kW range.<\/p>\n\n\n\n<p>The first major wave of GPU-dense <a href=\"https:\/\/exittechnologies.com\/es\/servicios-de-centro-de-datos\/servicios-de-desmantelamiento-de-centros-de-datos\/\">AI server decommissioning<\/a> has begun. Organizations that deployed NVIDIA GPU clusters and AI accelerator racks between 2022 and 2024 are now entering the disposal phase of their first refresh cycle. Most enterprise IT asset disposition programs were designed for five-to-seven-year horizons, not the 18\u201336 month cycles that AI infrastructure demands.<\/p>\n\n\n\n<p>The Roze buildout, if it executes, puts substantially more GPU-dense infrastructure online, faster. Every rack that gets built under Roze\u2019s accelerated timeline will need to be decommissioned on AI-era timescales. The downstream ITAD volume compounds with the build rate.<\/p>\n\n\n\n<p>Is your ITAD program structured for a recurring 18-month disposition cycle, or is it still built around a once-every-five-years event?<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What the Hardware Value Recovery Window Actually Looks Like<\/h2>\n\n\n\n<p>High-value AI assets (NVIDIA GPUs, NVMe arrays, high-speed networking gear) lose 40% of their recoverable worth within 60 days of retirement. It\u2019s the difference between recovering $800,000 from a 200-unit H100 decommission and recovering $480,000 from the same hardware 90 days later.<\/p>\n\n\n\n<p>The secondary market for AI compute hardware moves on GPU generational velocity: H100 to H200 to Blackwell. Each new architecture announcement compresses residual values on the previous generation. Timing your disposition against that curve is the variable that determines whether your decommission becomes an asset recovery event or a write-off.<\/p>\n\n\n\n<p>The market data supports this urgency. The global <a href=\"https:\/\/exittechnologies.com\/es\/servicios-de-centro-de-datos\/servicios-de-desmantelamiento-de-centros-de-datos\/\">servicios de desmantelamiento de centros de datos<\/a> market expanded from $12.12 billion in 2025 to $12.95 billion in 2026, and is projected to reach $19.94 billion by 2032 at a compound annual growth rate of 7.37%. The broader ITAD market is expected to grow from $18.6 billion in 2026 to $40.1 billion by 2035.<\/p>\n\n\n\n<p>That growth is being driven by the structural acceleration of AI infrastructure cycles: the exact dynamic that Roze is designed to amplify.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The Compliance Exposure That Scales With the Build<\/h2>\n\n\n\n<p>Faster builds mean faster retirements mean higher compliance risk per unit of time. That math doesn\u2019t change regardless of who built the data center.<br>NIST SP 800-88 governs media sanitization for federal and federally-adjacent workloads. R2v3 governs chain-of-custody documentation for certified ITAD facilities, requiring documented tracking from asset intake through final disposition. NAID-AAA certification covers secure data destruction with third-party auditing. None of these standards get easier to satisfy when your decommissioning calendar compresses from 60-month cycles to 24-month cycles.<\/p>\n\n\n\n<p>The compliance exposure compounds at scale. An enterprise running 300 AI accelerators on a three-year refresh cycle generates more chain-of-custody documentation requirements in a single disposal event than a traditional IT program generates in five years. If your ITAD vendor is issuing batch certificates instead of per-asset verification reports, that documentation gap is a liability.<\/p>\n\n\n\n<p>R2v3 specifically requires that each device be accounted for individually across the disposition chain. A certificate that says \u201c500 drives were processed\u201d doesn\u2019t satisfy R2v3\u2019s chain-of-custody requirements any more than a batch invoice satisfies an audit. The standard requires individual traceability from intake, to processing method, to final disposition status, per device.<\/p>\n\n\n\n<p>What does your current <a href=\"https:\/\/exittechnologies.com\/es\/servicios-de-centro-de-datos\/itad\/\">Proveedor de ITAD<\/a> provide? Per-asset documentation, or a batch certificate with a vendor\u2019s letterhead?<\/p>\n\n\n\n<p>Big Tech is committing more than $650 billion to AI infrastructure capex in 2026 alone, with Microsoft, Google, Amazon, and Meta accounting for the bulk of the spend. That capital is going into physical facilities. Those facilities will be filled with NVIDIA Blackwell clusters, high-bandwidth NVMe arrays, and 400GbE switching fabric. All of it will be retired on 18-to-36-month cycles. All of it will require NIST 800-88-compliant data sanitization, per-asset chain-of-custody documentation, and an ITAD partner with the certification stack to prove it.<br>Roze isn\u2019t creating the disposition wave. Roze is a bet that the wave arrives faster. The question for your data center team isn\u2019t whether to believe in the $100 billion valuation. It\u2019s whether your current ITAD program is structured for the volume and cadence of decommissioning that AI infrastructure already demands and whether you\u2019re timing your dispositions to recover value, or just to clear the cage.<\/p>\n\n\n\n<p>If you\u2019re managing GPU infrastructure on a refresh cycle and want to understand what your next decommission is actually worth, contact exIT Technologies for a hardware valuation assessment.<\/p>\n\n\n\n<p><\/p>","protected":false},"excerpt":{"rendered":"<p><span class=\"span-reading-time rt-reading-time\" style=\"display: block;\"><span class=\"rt-label rt-prefix\">Reading Time: <\/span> <span class=\"rt-time\"> 6<\/span> <span class=\"rt-label rt-postfix\">minutes<\/span><\/span>A $6.5 billion compute company, a $5.4 billion robotics division, and a $3 billion infrastructure firm walk into a blank legal entity called Roze. SoftBank wants to take this combination public at a $100 billion valuation before it has shipped a single product, disclosed a revenue figure, or confirmed an IPO date. The pitch: deploy [&hellip;]<\/p>\n","protected":false},"author":9,"featured_media":77923,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"episode_type":"","audio_file":"","podmotor_file_id":"","podmotor_episode_id":"","cover_image":"","cover_image_id":"","duration":"","filesize":"","filesize_raw":"","date_recorded":"","explicit":"","block":"","itunes_episode_number":"","itunes_title":"","itunes_season_number":"","itunes_episode_type":"","footnotes":""},"categories":[43],"tags":[],"class_list":["post-77922","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-data-center"],"acf":[],"_links":{"self":[{"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/posts\/77922","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/comments?post=77922"}],"version-history":[{"count":0,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/posts\/77922\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/media\/77923"}],"wp:attachment":[{"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/media?parent=77922"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/categories?post=77922"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/exittechnologies.com\/es\/wp-json\/wp\/v2\/tags?post=77922"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}