Tag: Supermicro

  • Supermicro’s 256-CPU Vera Rack Puts Agentic AI’s CPU Tier on the Liquid Loop

    Supermicro’s 256-CPU Vera Rack Puts Agentic AI’s CPU Tier on the Liquid Loop

    TL;DR · 30-second read

    The Short Version

    The newest AI systems don’t just answer questions. They act as “agents” that carry out tasks, such as writing and testing software. Each task needs its own walled-off workspace running on ordinary computer processors.

    Server maker Supermicro just launched a single cabinet holding 256 of NVIDIA’s new processors. It can host about 22,500 of those workspaces at once, and it is cooled by liquid piped through the cabinet instead of by fans.

    Why it matters: data centers building for AI may now need that plumbing across more of the building, not only in the rooms full of AI chips.

    Super Micro Computer (Nasdaq: SMCI) on October 23, 2026 announced the NVIDIA Vera CPU rack, a fully liquid-cooled rack holding 256 NVIDIA Vera CPUs, which adds up to 22,528 cores and 45,056 threads. Supermicro says one rack supports up to 22,500 simultaneous agent sandbox environments. The rack pairs the CPUs with NVIDIA BlueField-4 DPUs, uses NVIDIA’s third-generation MGX rack design and Supermicro’s DLC-2 direct liquid cooling stack, and is available now for quotation.

    The rack can run on its own or alongside Supermicro’s Vera Rubin NVL72 GPU racks, which the company says are now shipping. The launch comes three days after an 8-K filed October 20, in which Supermicro estimated fiscal first-quarter revenue of $18.5 billion to $19.5 billion, ahead of its $14.5 billion to $15.5 billion guidance, and attributed the beat primarily to increased demand for AI platforms.

    Executive Summary

    Supermicro has added a CPU-only rack to its Data Center Building Block Solutions (DCBBS) portfolio. The rack is built for the part of agentic AI that GPUs do not handle: running the actions an AI agent decides to take, such as executing code, calling tools, checking results and orchestrating multi-step workflows. Each of these runs in a sandbox, an isolated software environment that keeps an agent’s actions contained. Supermicro rates one rack at up to 22,500 concurrent sandboxes, roughly one per CPU core.

    The more consequential detail for facility planners is how the rack is built. It is fully liquid-cooled, it shares NVIDIA’s third-generation MGX rack architecture with Vera Rubin NVL72, and Supermicro offers a combined layout in which both rack types sit in rows served by redundant coolant distribution units and rear-door heat exchangers. In that design the CPU tier of an AI cluster draws on the same liquid-cooling and power infrastructure as the GPU tier. It does not sit in a conventional air-cooled hall next door.

    Supermicro still sells air-cooled 1U and 2U Vera servers, so liquid is an option rather than a requirement. But the rack-scale product, and the reference design that pairs it with NVL72, assumes that the CPUs serving AI agents live on the liquid loop.

    Why Agents Need a CPU Tier of Their Own

    Supermicro’s case for the product rests on a description of how agentic AI differs from chatbot-style inference. A conventional inference request is a straight line: a prompt goes in and generated text comes out. Agentic and reinforcement learning workloads branch and loop. A model proposes an action, something executes it, the result is evaluated, and the model tries again. In the company’s framing, GPUs still do the heavy lifting of generating tokens, code and queries. CPU-based sandboxes execute the actions, run the tools and orchestrate the workflow. Reinforcement learning here means training a model by letting it attempt tasks and scoring the outcomes.

    The arithmetic in the announcement is consistent with that picture. 256 CPUs delivering 22,528 cores works out to 88 cores per Vera CPU, and 45,056 threads means two threads per core. The 22,500-sandbox ceiling is therefore close to one sandbox per physical core. That is a plausible density for lightweight, isolated execution environments. It is also an “up to” figure whose real value depends on how much memory, storage and compute each agent’s sandbox consumes, and Supermicro has not defined that.

    NVIDIA’s Chris Marriott made the same argument in the announcement: supporting thousands of concurrent agent workflows requires “a new class” of high-density CPU infrastructure. The claim is directional rather than quantified. Neither company has published a recommended ratio of CPU racks to GPU racks, and that ratio is the number that would tell an operator how much extra floor space agentic workloads actually add.

    The Liquid Loop Now Extends Past the GPU Hall

    This is where the announcement matters for people who build and run data centers. Liquid cooling has so far been justified by GPU density. Racks of AI accelerators produce too much heat for air alone, so operators have plumbed coolant to those rows. Supermicro’s Vera CPU rack is described as 100% liquid-cooled through its DLC-2 direct liquid cooling stack, in which cold plates sit directly on the chips. It is built on the same third-generation MGX architecture as Vera Rubin NVL72. Supermicro presents that shared architecture as a way to standardize deployment.

    The optional combined design spells out the facility consequence. Vera Rubin NVL72 racks and Vera CPU racks are laid out together with hot-aisle containment, which walls off the hot exhaust side of the racks so heat does not mix back into the room. The rows are served by redundant in-row coolant distribution units (CDUs), the pump-and-heat-exchanger units that move heat from rack coolant into the building’s water system, and by rear-door heat exchangers, radiators mounted on the back of a rack that capture residual heat. In that layout, every CPU rack added for agent sandboxes is another load on the CDUs, the facility water loop and the containment scheme. It is not a cheap air-cooled addition at the edge of the hall.

    The affected groups are colocation providers, enterprise data center teams and the engineers sizing cooling plants for AI clusters. Their liquid-cooling capacity plans have been built around GPU rack counts. If buyers adopt the rack-scale CPU product rather than the air-cooled 1U and 2U alternatives, CDU capacity and piping need to cover the CPU rows as well. The size of that addition cannot yet be calculated, because Supermicro has not disclosed the rack’s power draw or heat load. What can be said is that the reference design treats the CPU tier as liquid-cooled floor space by default.

    800 VDC and Batteries in Place of a UPS

    The power design follows the same logic of standardizing around the GPU rack. Customers can feed the compute rack with conventional AC power, or with an 800-volt DC sidecar power rack, a dedicated rack beside the compute racks that converts and distributes power. The sidecar option allows battery backup units to replace a conventional uninterruptible power supply (UPS), the room-scale battery system that keeps equipment running through a utility outage.

    Supermicro says this approach frees data center floor space, can improve end-to-end power conversion efficiency, and lowers both capital and operating costs. These are reasonable design goals, since fewer conversion steps generally mean less energy lost as heat. The announcement does not quantify any of them, however. Operators evaluating the option will want efficiency figures, battery ride-through times and an account of how a sidecar-based design interacts with their existing electrical distribution and redundancy standards.

    Integration Is the Product, and the Margin

    Security is offered as a hardware feature. NVIDIA’s OpenShell Reference Architecture runs on the Vera CPUs and applies least-privilege, policy-based controls to agents at runtime. NVIDIA Sentry runs on the BlueField-4 DPUs, network cards with their own processors, as an out-of-band monitor. It operates separately from the host system, enforces policy, provides attested telemetry and, according to the announcement, can quarantine an agent within milliseconds if it oversteps its limits. Moving that monitoring off the host CPUs is meant to preserve compute for the sandboxes themselves. Independent validation of these claims will matter to buyers putting autonomous agents into production.

    Supermicro’s broader pitch is end-to-end delivery. One team covers site survey, design, rack- and cluster-level integration and testing, on-site deployment and ongoing support. That positioning fits the company’s financial profile. In the October 20 business update filed with the SEC, Supermicro estimated first-quarter fiscal 2027 revenue of $18.5 billion to $19.5 billion and gross margins of 12.0% to 12.5%, both above guidance. It attributed the revenue beat primarily to increased demand for AI platforms and the margin improvement primarily to customer and product mix. The figures are preliminary and unaudited. They are also thin margins on very large volumes, which helps explain why the company emphasizes integrated racks, cooling and services over standalone servers. The Vera CPU rack is quote-only today and contributed nothing to that quarter. Investors will hear more on the November 3 earnings call.

    Background

    Super Micro Computer, based in San Jose, California, designs and manufactures servers, storage and networking systems in the US, Taiwan and the Netherlands. It has become one of the largest suppliers of AI server racks built around NVIDIA platforms, and it increasingly sells complete data center packages, which it calls DCBBS, covering power, liquid cooling and deployment services. In an 8-K filed October 15, the company reported that Mark Long had been appointed chief financial officer, succeeding David Weigand.

    NVIDIA’s Vera Rubin platform pairs its Vera CPU with its Rubin GPU. Vera Rubin NVL72 is the rack-scale GPU system that Supermicro says it is now shipping. The new Vera CPU rack extends the platform into CPU-heavy workloads created by AI agents, using the same third-generation MGX modular rack design so that both rack types can be deployed and cooled as one system.

    Sources

    Source: Supermicro introduceert het NVIDIA Vera CPU-rack: een volledig vloeistofgekoelde oplossing op rackschaal met 256 CPU’s, als aanvulling op het NVIDIA Vera Rubin-platform voor agentische AI op grote schaal. Supermicro’s announcement of its liquid-cooled, 256-CPU rack for agentic AI sandboxes.

    Primary sources: Super Micro Computer Form 8-K filed October 20, 2026; Supermicro Provides First Quarter of Fiscal Year 2027 Preliminary Business Update (Exhibit 99.1); Super Micro Computer Form 8-K filed October 15, 2026 (Chief Financial Officer Transition).

  • Supermicro’s 144-GPU OCP Racks Bet AI Capacity Ships by the Rack, Not the Server

    Supermicro’s 144-GPU OCP Racks Bet AI Capacity Ships by the Rack, Not the Server

    TL;DR · 30-second read

    The Short Version

    Supermicro, a large American maker of the computers that run AI services, is now selling complete AI equipment cabinets that arrive ready to plug in, instead of individual machines that have to be assembled on site.

    One version packs 144 AI chips into a single cabinet a bit taller than a person. A matching cooling unit is built to carry away about as much heat as a hundred kitchen ovens running flat out.

    The bet is that companies racing to build AI will pay for speed and simplicity. Supermicro has not said what the cabinets cost or who is buying them.

    Super Micro Computer (Nasdaq: SMCI) on October 14, 2026 announced an expanded portfolio of Open Compute Project (OCP) hardware for AI and high-performance computing, timed to the OCP Global Summit in San Jose (October 12–15). The lineup spans OCP Open Rack v3 (ORv3) and Open Rack Wide (ORW) racks with centralized 48-volt DC power busbars, liquid-cooled GPU systems including a 2OU NVIDIA HGX B300 node that allows up to 144 GPUs per rack, a 250 kW in-rack coolant distribution unit (CDU), rear-door heat exchangers rated from 10 kW to 120 kW, and 800G and 1.6T Ethernet switches, including co-packaged optics designs.

    Supermicro packages the components under its Data Center Building Block Solutions (DCBBS) brand as pre-integrated rack- and cluster-scale deployments, managed through its SuperCloud software suite, and says the OCP solutions are available now. The company did not disclose pricing, named customers or shipment volumes.

    Executive Summary

    Supermicro’s announcement is less about any single server than about where the boundary of the product sits. By building on OCP rack designs that replace each server’s own power supplies with a shared 48V busbar, and pairing them with its own coolants, in-rack CDUs, rear-door heat exchangers, switches and management software, the company is offering the entire rack as the deliverable. CEO Charles Liang framed the strategy as integrating compute, networking, power, cooling, software and services into validated rack-scale solutions to shorten the path from order to production.

    That matters because AI hardware has outgrown the assumptions most data centers were built on. A rack holding 144 liquid-cooled GPUs cannot simply be wheeled into a room designed for air-cooled servers. Supermicro’s answer is to ship the power distribution, liquid cooling and residual-heat removal as part of the rack, including rear doors it says remove exhaust heat without changes to a data hall’s air conditioning.

    The release is strong on specifications and light on commercial evidence. It names no customers, prices, lead times or deployment-time figures for Supermicro’s own racks, so the central promise of faster time-to-online remains a claim rather than a demonstrated result.

    Why the Busbar Makes the Rack the Product

    In a conventional data center, every server is a self-contained appliance: it has its own power supplies, its own power cords and its own fans, and it can be bought, racked and replaced individually. The ORv3 and ORW racks in Supermicro’s portfolio break that model. They replace per-server power supplies and cabling with a centralized 48-volt DC busbar, a copper rail running up the back of the rack that is fed by 1OU power shelves, with 2OU backup power units alongside. Servers slide in and draw power from the rail. The same logic applies to cooling: Supermicro’s DLC-2 direct-to-chip system moves heat through cold plates into a coolant loop served by an in-rack CDU, and its 2OU co-packaged optics switch takes power from the busbar and connects to the liquid loop through blind-mate QD08 quick-disconnect fittings.

    The 144-GPU figure shows what that integration buys. Each NVIDIA HGX B300 node holds eight GPUs in 2OU (OCP’s “OpenU” unit of rack height), and Supermicro says up to 18 of them fit in a 21-inch ORv3 rack, for 144 GPUs. Those nodes occupy 36 OU of a 44OU or 48OU rack; the remaining space has to hold the power shelves, switches and, where used, a 4U CDU. None of those 144 GPUs does useful work until power, coolant and network are all present and validated together, which means the integration risk sits at the rack, not the server. Supermicro’s bet is that buyers would rather have that risk absorbed in its factory than discover problems on their own data-hall floor.

    For operators, neoclouds and enterprise buyers, the practical consequence is that capacity planning shifts to rack-level interfaces: a busbar power feed, a facility-water connection and a network uplink. Facilities teams spec around those few connection points rather than hundreds of individual server cords. The trade-off is that the unit of failure, delivery and replacement grows too. A delayed or defective rack holds back 144 GPUs, not eight.

    Cooling Aimed at Halls Built for Air

    Direct-to-chip liquid cooling captures heat from the hottest components, but some heat still leaves the rack as warm air. Supermicro’s rear-door heat exchangers (RDHx), coil-and-fan units mounted on the back of ORv3 and ORW racks, are designed to remove that exhaust heat before it reaches the room. The company lists six models rated from 10 kW to 120 kW, in AC and DC versions, with the DC models plugging directly into the rack busbar. They include N+1 redundant fans and an anti-condensation system to protect nearby equipment from moisture. Separate chiller doors are rated at 20–100 kW for ORv3 racks and 60 kW for ORW racks, controllable through Redfish, SNMP and web interfaces.

    Supermicro says the rear doors let operators raise rack density in both new AI facilities and existing data centers without modifying data-hall air conditioning. That is the commercially important claim, because much of the world’s installed colocation capacity was designed for air-cooled racks at far lower densities than AI systems need. If the residual heat can be handled at the rack, an existing hall can host denser equipment without a mechanical rebuild of its air systems.

    The retrofit does not become free, however. Rear doors and CDUs both reject heat into a facility water loop, so the question for an older building moves from air handling to water: pipe routing, flow capacity and supply temperature to each rack position. The 250 kW 4U in-rack CDU is specified for 45°C warm-water cooling. Warmer supply water can often be cooled with dry coolers rather than energy-intensive chillers, which supports Supermicro’s efficiency pitch, but it still requires plumbing to the rack that many air-era halls do not have.

    Open Standards, Bundled Delivery

    OCP’s value to buyers has traditionally been interoperability: if racks, power shelves and cooling follow published specifications, components from different suppliers should work together, giving purchasers leverage over any single vendor. Supermicro leans on that heritage. It describes the systems, racks and rear doors as OCP Accepted, contributed its MGX 1.2 rack series for NVIDIA Vera Rubin NVL72 and Vera CPU platforms to the OCP program, and designed its two coolants, DLC-1 for standard liquid-cooled racks and the ultra-low-conductivity DLC-2 for specialized cold plates, to meet the OCP PG25 baseline specification. OCP marketing vice president Dirk Van Slyke endorsed the approach in the release.

    At the same time, DCBBS bundles Supermicro coolants, CDUs, doors, switches, its Enterprise Advanced SONiC network operating system and its SuperCloud management software into one delivery. That is a reasonable product strategy: open specifications reduce lock-in at the interface level, while a single integrator reduces the coordination burden and offers one point of accountability. Buyers will weigh how much of the portability OCP promises survives once the rack arrives as a tested Supermicro system.

    The speed claim deserves the most scrutiny. Faster time-to-market and time-to-online are the core of the pitch, yet Supermicro offers no figure for how much faster its integrated racks reach production than component-level builds. The only deployment-time multiple in the release, 1.3x, belongs to NVIDIA’s photonic switches compared with pluggable-transceiver networks.

    The Network Moves Into the Rack Budget

    Networking is being pulled into the same integrated envelope. Supermicro’s 2OU co-packaged optics (CPO) switch, built on Broadcom Tomahawk 6 silicon, delivers 102.4 Tbps across 128 ports of 800G, draws 40–60 VDC from the rack busbar and is liquid-cooled. Co-packaged optics place the optical components next to the switch chip rather than in plug-in transceiver modules at the front panel, an approach aimed at cutting power and failure points.

    DCBBS will also include NVIDIA Spectrum-X Ethernet Photonics switch systems built on the 102.4 Tb/s Spectrum-6 ASIC with 200 Gb/s SerDes. NVIDIA’s claims, carried in the release, are 5x better energy efficiency, 10x longer mean time between interruptions, 5x longer uptime for AI applications and 1.3x faster deployment compared with networks built on pluggable transceivers. Those are vendor-stated comparisons without published test conditions, but NVIDIA says the product is in production. Offering both Broadcom- and NVIDIA-based switching lets Supermicro sell into either camp, and reinforces the rack-as-product thesis: once switches run on busbar power and liquid cooling, they become another resident of the integrated rack rather than a separate network purchase.

    Background

    Super Micro Computer, based in San Jose, California, designs and manufactures servers, storage, switches and related software, with production in the United States, Taiwan and the Netherlands. Its business has long rested on modular “building block” designs that let customers configure systems for specific workloads, and the AI boom has pushed that model up from individual servers to full racks and clusters under the DCBBS brand. The company’s fiscal year ends June 30. In an 8-K filed October 8, Supermicro said it had appointed Shesha Krishnapura, a 35-year Intel veteran whose experience includes high-performance computing, AI infrastructure and data center operations, as an independent director, while Judy Lin, a director since April 2022 and chair of the nominating and corporate governance committee, is retiring from the board.

    The Open Compute Project began as an effort by large web operators to standardize and simplify data center hardware. Its Open Rack designs moved power out of individual servers and into shared rack-level busbars, an idea that has gained relevance as AI accelerators push rack power and heat well beyond what traditional air-cooled enterprise rooms were built for. GPU makers have moved in the same direction, with NVIDIA’s NVL72 systems and AMD’s Helios platform designed around entire racks rather than individual servers.

    Sources

    Source: Supermicro breidt geïntegreerde open computing-projectportfolio uit om de implementatie van AI-datacenters op rackschaal te versnellen — Supermicro’s October 14, 2026 announcement of its expanded OCP rack-scale portfolio for AI data centers.

    Primary sources: Super Micro Computer, Inc. Form 8-K filed October 8, 2026 (board appointment and director retirement).

  • Supermicro’s 120kW Rear Doors Shift the AI Retrofit Limit From Airflow to Heat Rejection

    Supermicro’s 120kW Rear Doors Shift the AI Retrofit Limit From Airflow to Heat Rejection

    TL;DR · 30-second read

    The Short Version

    Computer chips that run artificial intelligence throw off enormous heat. Supermicro, a major server maker, now sells ten models of a water-cooled door that bolts onto the back of a server cabinet and soaks up that heat before it reaches the room.

    The largest door handles about as much heat as 80 household space heaters running at once. That lets older buildings, designed around fans and air conditioning, host far hotter equipment without major rebuilding.

    The catch: the heat still has to leave the building, and Supermicro has not fully explained how.

    Supermicro announced in a July 15, 2026 press release that it has expanded its rear door heat exchanger (RDHx) portfolio to ten models, with cooling capacities from 10kW to 120kW per door and up to 240kW at the rack level. The doors are part of its Data Center Building Block Solutions (DCBBS) program, which packages servers, racks, power, cooling, management software and deployment services into validated, pre-integrated systems.

    Supermicro says the doors mount to standard EIA, ORv3 and MGX racks, can serve as a primary liquid cooling method or be paired with its direct-to-chip cold plates, and can be installed in new or existing data centers without major facility modifications or dedicated facility chilled water.

    Executive Summary

    Supermicro is broadening the least invasive rung of its liquid cooling ladder. A rear door heat exchanger replaces the back door of a server cabinet with a liquid-filled coil and fans; hot exhaust passes through the coil and leaves the cabinet cooled. The new lineup spans ten models from 10kW to 120kW per door, which lets operators size the door to anything from a single dense server stack to a rack-scale AI system.

    The pitch is aimed squarely at the large installed base of air-cooled data halls that were never designed for AI hardware. If a door can capture heat at the cabinet, an operator can place a high-density GPU rack in an older room without plumbing liquid to every server or rebuilding the hall’s airflow.

    That solves one constraint and exposes others. A door that captures 120kW of heat still has to hand that heat to something that carries it outdoors, and a rack that needs 120kW of cooling needs roughly that much power delivered to it. For operators, the doors move the retrofit question from the room to the building’s heat rejection plant and electrical distribution.

    A Door That Turns an Air-Cooled Hall Into a Liquid-Assisted One

    Traditional data halls cool servers the way an office cools people: chilled air is pushed into the room, servers draw it through their chassis, and hot exhaust is returned to air handlers. That works until the heat in a single cabinet becomes so concentrated that the room cannot move enough air to carry it away. GPU servers built for AI training and inference routinely reach that point, which is why the industry has been moving toward liquid cooling.

    A rear door heat exchanger is the gentlest version of that move. Instead of routing coolant into each server, which is what direct-to-chip cold plates do, it intercepts the exhaust at the back of the cabinet. The servers stay air-cooled internally; the room simply never sees most of their heat. Supermicro says its doors fit standard EIA racks (the long-established 19-inch format), ORv3 (the Open Compute Project’s Open Rack version 3) and MGX (Nvidia’s modular server reference architecture), which covers most of what operators already have on the floor or are buying for AI.

    The ten-model range is the practical news. A 10kW door and a 120kW door solve different problems, and offering both lets an operator match cooling to the rack rather than overbuilding. Supermicro’s stated features include intelligent fan control, N+1 redundancy (one more fan or component than strictly needed, so a single failure does not stop cooling) and anti-condensation protection, which matters because a cold coil in a humid room can drip onto equipment.

    The Retrofit Limit Moves From Airflow to Heat Rejection

    A rear door changes where heat is captured, not how much heat exists. Nearly every watt a server draws ends up as heat, so a door absorbing 120kW is handing 120kW of thermal energy to its coolant, and at Supermicro’s stated rack-level maximum of 240kW, twice that. In an air-cooled hall, the traditional ceiling on rack density is airflow: how much cold air the room can deliver and how much hot exhaust it can return before it recirculates into server intakes. By neutralizing exhaust at the cabinet, a rear door largely removes that ceiling.

    What remains is the plant behind the door. Captured heat has to reach the outdoors through chillers, dry coolers or cooling towers, and a building’s heat rejection capacity was sized for the hall it was originally designed to be. Supermicro says the doors deploy without the need for dedicated facility chilled water or extra hardware, but it has not said what the coil connects to instead. If the heat is dumped back into the room by a liquid-to-air unit, the existing air conditioning absorbs it and the old limit returns in another form. If the doors tie into an existing water loop, that loop’s spare capacity becomes the constraint.

    The same logic applies to electricity. A rack that needs 120kW of cooling is drawing roughly that much power, and older halls were generally not built to deliver that at every position. Supermicro’s option of DC-powered doors fed from rack busbars simplifies powering the door itself, not the upstream feed to the rack. For colocation providers and enterprises weighing an AI retrofit, the door shifts the practical question from whether a rack can be cooled to how many such racks the building’s cooling plant and switchgear can actually support.

    Rear Doors and Cold Plates Are Complements, Not Rivals

    Supermicro positions the doors both as a standalone option and as a companion to its direct-to-chip (D2C) liquid cooling. That pairing reflects how dense AI systems are actually cooled. Cold plates mounted on processors and GPUs remove the heat from the hottest chips directly into liquid, but memory, power supplies, storage and networking gear in the same rack still shed heat into the air. A rear door can capture that residual heat so the room does not have to.

    For operators, the hybrid approach offers a staged path: start with doors on air-cooled GPU servers in an existing hall, then add cold-plate systems as liquid distribution is built out, with the doors continuing to handle what the plates miss. Supermicro has not published how it splits heat between the two methods in a combined deployment, which is the figure facility engineers would need to size the loops.

    Bundling Is the Strategy; the Economics Are Still Claims

    The broader move is DCBBS itself. Supermicro wants to sell not just servers but the rack, power, cooling, management software and installation as one validated package, and says that approach reduces integration risk, lowers total cost of ownership and shortens time-to-online. Real-time monitoring of temperature, pressure, flow rate and pump status through Redfish (an industry standard interface for managing hardware), SNMP (a long-standing network monitoring protocol) and Supermicro’s SuperCloud Composer software is meant to fold cooling into the same management plane as the servers.

    Single-vendor integration has genuine appeal when AI hardware lead times are tight and every week a facility sits unfinished is lost revenue. The trade-off buyers weigh is flexibility: a tightly validated stack is simpler to deploy but ties more of the facility’s design to one supplier’s roadmap. Supermicro’s cost and speed benefits are stated rather than quantified, so buyers will need site-specific figures before treating them as planning assumptions.

    Background

    Super Micro Computer (NASDAQ: SMCI) is a San Jose, California server and storage maker that designs and manufactures its systems in the United States, Taiwan and the Netherlands. It has become one of the principal suppliers of GPU servers for AI workloads and, through its Data Center Building Block Solutions program, increasingly sells complete rack-scale and data center-level systems rather than individual machines.

    Liquid cooling has moved from niche to mainstream as AI hardware has pushed per-rack power far beyond what traditional air-cooled halls were built to handle. Operators generally choose among rear door heat exchangers, direct-to-chip cold plates and immersion cooling, often combining methods. Rear doors are typically the least disruptive to install, which makes them a common first step for existing facilities.

    Sources

    Source: Supermicro Expands End-to-End DCBBS Liquid Cooling Portfolio with Rear Door Heat Exchangers for High-Density AI and HPC Infrastructure, Supermicro’s July 15, 2026 announcement of a ten-model rear door heat exchanger line rated 10kW to 120kW per door.