AI rollouts fail for boring reasons.
Too many vendors. Too many hand-offs. No RACI.
Racks arrive before pipework. Servers boot before cooling is signed off.
And benchmarks?
Running on clusters never laid out for the airflow they need.
This company formed to kill that chaos. From data-centre design to cluster layout to DLC pipework. From power metering to performance benchmarking. A single screen that shows whether the thing you paid for is actually delivering.
One partner.
Every dependency.
Who are they?
The GPU Audio Companion Issue #
Want the GPU breakdown without the reading? The Audio Companion does it for you, but only if you’re subscribed. If you can’t see it below, click here to fix that.
Company Background
Gatti started at the sharp end of deployments.
Founder Christof Stührmann led technical rollouts at scale and watched good parts fail as systems. Everyone delivered their piece. No one stitched it together.
The answer wasn’t more paperwork, but a different operating model.
So they built it.
Own the line end-to-end. Design the room. Fit the room. Rack the gear. Cable it. Cool it. Meter it. Burn it in. Benchmark it.
The company began with cluster-level performance work.
That work pulled them down the stack, into placement, racking, airflow, and liquid.
Throughput is won or lost in the depths, because it is here that rack density becomes either a steady state or a ticket queue. In turn, such granular focus closed the gap between colocation vendors and AI operators. With Gatti serving as both the deployment owner and planner for both stakeholders, points of friction where projects typically wobble are mitigated.
And with late drawings, surprise loads, misplaced trays, blind spots in airflow, and acceptance criteria tied to spec sheets rather than live jobs no longer an issue, all stakeholders can focus on the only thing that matters:
Customer outcomes.
How They Work
Design - Start with the workload. Map thermal and power envelopes, not just nameplate values. Choose layouts that avoid dead ends. Leave space for maintenance and growth, not artistic cable trays that look good on day one and trap you on day ninety.
Planning - Publish a schedule that joins civil, electrical, mechanical, and IT tasks into one critical path. Flag where change breaks acceptance. Lock inventory dates and spares. Tie each dependency to a person, not a department.
Physical deployment - Prepare the site. Set up power, cooling, containment, and safety. Install racks, cable trays, and fibres properly labelled for humans, not heroes. Bring in servers, storage, networking, and optics with quality checks that prevent silent faults.
Technical deployment - Integrate systems, configure networks, validate liquid flow and pressures, test failover paths, and instrument power and thermals. Prove performance under load. Fix what drifts. Only then invite stakeholders to acceptance.
Post-deployment support - 24/7 smart hands, routine maintenance, and upgrades without disrupting live jobs. Observability that ties alarms to actions rather than dumping noise into your NOC. When growth comes, repeat the pattern, not the mistakes.
Executive Team
Gatti is UK-registered with EU operating experience, giving customers a single counterpart that can work across jurisdictions and facility standards. The team blends physical infrastructure engineers, programme managers, and production-grade software and systems talent.
Christof Stührmann (Founder & CEO) - An industry innovator, formerly of Northern Data, who treats AI infra as a systems integration problem. Sets the “one owner, no excuses” model and stays close to delivery.
Michael Krolikowski (CTO) - 15+ years building resilient distributed systems. Turns requirements into architectures that survive production. Owns the interface between facility signals and cluster behaviour.
Marion McCallum (Director of Operations) - Runs governance, logistics, and programme execution. Keeps schedules real, inventories synced, and dependencies closed.
Fabian Meeßen (Director, Programme Management) - Engineer turned manager. Designs and optimises data-centre infrastructure, mentors the next layer of operators, and pushes for clean acceptance criteria.
Carmen Robbins (Operations Manager) - Customer alignment, internal coordination, and delivery control. The person who makes sure promises match the work on the floor.
Falk Stern (Tech Lead) - Networking wizard of industry renown. Deep hands-on experience with distributed systems and data-centre realities; the last call you make if everything else fails.
The Edge
Own the Integration - Most “full service” offers are a relay race. Different vendors, different incentives, and a baton that drops somewhere around cable management and liquid routing. Gatti runs one team, one plan, one schedule, one set of acceptance tests. If a task slips, there is one number to call.
Liquid, Done Properly - At today’s densities, Direct Liquid Cooling isn’t a pilot. It’s baseline. Gatti integrates DLC design and install with the rest of the build. That means pipework and manifolds laid out for maintenance and growth, proper separation between facility and IT liquid, and sensible choices around rear-door heat exchange and CDU placement. No afterthought loops that cap performance or lock you into fragile layouts.
Performance is the Contract - Delivery doesn’t stop at power-on. Gatti burns in, tunes, and benchmarks with workloads that resemble what the cluster will actually run. Acceptance is based on throughput, stability, and thermal behaviour under sustained load.
Facility-to-Node Observability - A single pane of truth with power usage, thermal headroom, and network choke points tied to node-level performance. The point is not a pretty dashboard. The point is catching three real failure modes early: thermal throttling, uneven power draw that trips protection, and optics or cabling choices that flatten throughput just when you scale a job. Operators see what matters in time to fix it.
Standards without bureaucracy - Repeatable checklists, documented install patterns, sensible labelling, and acceptance criteria written in plain English. Enough process to prevent drift. Not so much that the site freezes while a template is updated.
Recent Moves
Density, formalised - Gatti has codified liquid-ready builds with clear patterns for rear-door cooling, CDUs, and smart PDUs, and treats these as first-class components in the plan, not optional extras. That speeds design approvals and reduces rework, narrowing the failure window around airflow surprises and rogue heat islands.
Service breadth, public - What began as bespoke deployments is now a published chain: design, planning, physical install, technical integration, post-deployment support, and 24/7 smart hands, plus monitored smart PDUs. Customers get one contract that spans the rollout rather than stacking SOWs that break at hand-off.
Tooling from fieldwork - Internal software used to track acceptance, power, and thermal behaviour is being shaped into commercial products. The plan is pragmatic: keep the core tooling close to delivery at first, then expose the right pieces so customers can run their own checks without pulling a technician for every anomaly.
Scale claims and ecosystem ties - Gatti reports deployments across tens of thousands of GPUs, extensive cabling and network infrastructure, and multiple cooling installations across data centre estates. All while maintaining close alignment with major OEMs and storage partners, with installs kept inside vendor tolerances and warranty bounds.
What’s Next
The next 12 months will be a marquee year for Gatti Services.
The groundwork is in place. Customers keep validating the service. And now it’s time to scale.
That means marketing and brand development, and a broad focus on name recognition.
Gatti are exhibiting at booth 1845. The first release of Metal-as-a-Service tooling and an early Datacentre-as-a-Service platform will be on show. These are field-built tools, lifted from delivery playbooks, not greenfield experiments. Expect standardised acceptance checks, workload-aware burn-in, and a live link from facility telemetry to runtime behaviour.
All with the goal of cutting time-to-use, reducing variance between sites, and making “tokens out” the default acceptance metric.
Next up is software capability expansion.
Today’s single-pane view will gain scoped control where customers want it. Roadmap items include role-based actions, audit trails, API hooks, and policy guards that prevent unsafe changes. Integrations will extend across smart PDUs/CDUs, rear-door heat exchange, BMS/DCIM, and cluster schedulers, so facility signals and job behaviour stay aligned.
The result?
A shift from simple observability to controlled execution, feeding into a broader US build-out throughout 2026.
The focus here is threefold: stateside delivery capacity, on-shore logistics, and partner coverage for high-density fit-outs. Gatti are targeting >100,000 GPUs installed and >50 MW of completed data-centre fit-outs. That means leveraging certified DLC installers, spare hubs near major metropolitan areas, pre-validated topologies, and acceptance-as-code to maintain installations within vendor tolerances and warranty bounds.
Where customers allow, Gatti will start releasing delivery metrics that matter:
Time-to-first-job, variance between racks under sustained load, thermal headroom vs density, rework avoided, and mean time to recover from common faults.
The intent here is to incorporate program management, liquid systems, and low-level networking. Or, to put it another way, to simplify, make site quality legible, and set the standard for operators who can own outcomes across the seam between facilities and SRE.
2026 will see Gatti’s customer engagement scope widen, but their operating model will stay the same:
One owner, one schedule, one set of acceptance tests.
The tools formalise it. The US expansion scales it. But the deliverable?
That remains unchanged:
Capacity that hits performance numbers on real jobs and keeps doing so when the room is hot and busy.
The only question that remains is whether or not Gatti can execute on this vision, or if they’ll validate why the market keeps these dependencies separate.