01 / Operators
Build infrastructure that performs reliably at scale.
Make the architecture, platform, vendor, and go-to-market decisions that build the foundation for success.
AI infrastructure advisory
We help investors, operators, and enterprises make defensible infrastructure decisions—across compute, datacenters, networking, storage, and platforms.
Start an engagement
01 / Operators
Make the architecture, platform, vendor, and go-to-market decisions that build the foundation for success.
02 / Enterprises
Assess partners, build-vs-buy choices, and sales motions into AI infrastructure companies with the realities of production in view.
03 / Investors
Stress-test investment theses, debunk the latest hype AI meme, and separate durable moats from operational constraints.
Coverage
Infrastructure decisions are connected. We bring the operator’s view across the whole system, so a good answer in one layer does not create a worse problem in another.
01
Cluster strategy, capacity, utilization, reliability, workload fit, and unit economics.
02
Power, cooling, site readiness, supply chain, rollout execution, and operational risk.
03
Fabric design, private links, multicloud architecture, latency, and cluster networking.
04
Storage patterns, data movement, file systems, enterprise requirements, and resilience.
05
SLURM on Kubernetes, virtualization, observability, monitoring, failure detection, and developer platforms.
06
Segmentation, packaging, pricing, enterprise objections, and competitive positioning.
Selected work
A sample of the work behind the people at Realtime Labs, anonymized to respect client confidentiality. We are a group of seasoned operators with hyperscale, edge, CDN, and neocloud experience, passionate about shaping the AI infrastructure landscape.
01 / Cluster operations
Prior neocloudSituation
A stealth neocloud needed to bring its first H100 cluster online to hit a contractual revenue milestone — 50 staff across 10 teams, no shared deployment playbook, no production track record.
Approach
Built the first baremetal deployment workflow end-to-end. Stood up daily node-tracking, defined exit criteria for inter-team handoffs, and routed every gate through a single Jira system of record.
Outcome
First revenue milestone delivered on schedule. Over 2,000 production changes tracked through the system. The deployment workflow became the operating standard across subsequent cluster classes.
02 / Storage strategy
Multi-PB dealSituation
An AI infrastructure provider needed to build a competitive multi-tenant storage offering against FsX Lustre and S3, with a real customer waiting on capacity decisions.
Approach
Ran vendor diligence across Weka, Vast, Alluxio, Quobyte, Tigris, Pure, Cloudflare R2, Solidigm, WD. Designed a two-tier architecture — high-performance parallel filesystem plus on-prem object — and negotiated preferential pricing.
Outcome
Closed the deal in weeks at healthy margin. First service-availability metrics shipped for S3, POSIX, and control-plane APIs. MLPerf Storage benchmarks established the public baseline.
03 / Observability
GreenfieldSituation
Engineering teams were flying blind. Node failures surfaced through customer escalations, not telemetry. Inventory reconciliation was manual across three sources of truth.
Approach
Stood up a Grafana-fronted telemetry stack. Unified inventory through a ServiceNow + Jira stack, Netbox, and a Snowflake source of record. Wired automated health checks into a hot-spare and remediation loop.
Outcome
Mean-time-to-detect dropped from hours to minutes. Automated failover became routine. Inventory reconciliation moved from a weekly meeting to a passive dashboard.
04 / Sovereign compute
Hyperscaler partnerSituation
A hyperscaler partner was scoping a sovereign cloud buildout in the 1 GW range, with 800V DC architecture and a non-trivial transition from baremetal to virtualized GPU offerings.
Approach
Reviewed the 800V DC spec, baremetal automation, and firmware-detection layer. Built a technical due-diligence template applied across site evaluations from XX MW to X GW. Scoped backbone connectivity at multi-Tbps PNI.
Outcome
Validated the BMaaS-to-VMaaS migration path for the chip OEM. Defined networking WAN scope across multiple stakeholders. Template now reusable across subsequent sovereign engagements.
05 / AI cloud networking
Stealth AI cloudSituation
A stealth AI cloud needed its first customer-facing network: a multi-tenant overlay, load balancers, and management plane purpose-built for AI training and inference — not a re-skin of enterprise cloud networking.
Approach
Architected the customer overlay end-to-end and directly contributed control-plane and datapath code. Partnered with AI server and Kubernetes engineers on on-host networking and NCCL, and led cross-layer triage when issues spanned GPU hosts and the underlying fabric.
Outcome
Overlay control plane and datapath running in production in the first AI datacenter. Design decisions across the network software stack realigned to AI workload demand.
Every engagement was load-bearing for revenue, a partnership commitment, or a board milestone. The work had to ship.
The work cut across product, engineering, ops, and commercial. Single-layer recommendations do not survive contact with production.
Each delivered a system, a template, or a workflow the team kept using — not a slide deck filed in a drive.
How we engage
Every engagement is scoped to the decision at hand and ends in a written, decision-ready output.
Discuss an engagement01
Clear, defensible assessments of a technology, vendor, or infrastructure thesis.
02
A strategic view of the landscape, inflection points, and credible alternatives.
03
Decision frameworks built around the criteria that actually matter for your context.
04
Ongoing access to operator judgment for teams making continuous high-stakes calls.
Let’s get started
Share a few details. We’ll route the conversation to the right expertise and follow up with next steps.