InfiniBand / High-Performance Fabric Engineer recruiters who bring you a focused shortlist
Data Center Headhunters are InfiniBand / High-Performance Fabric Engineer recruiters for data center owners, operators and builders. Send us the open role and we find working InfiniBand / High-Performance Fabric Engineer candidates, qualify each one live and give you a short list to interview. No upfront fee. You pay only if you hire.
AI clusters live or die on the fabric, so you need someone who has tuned one under load instead of reading about it.
This is what we do all day: recruit data center professionals.
No postings. No stacks of resumes. No chasing. You interview a short list.
No upfront fee. You pay only if you hire.
No upfront fee. You pay only if you hire.
We offer a candidate replacement guarantee on standard search agreements. The applicable period and exact terms are set out in the signed client recruiting agreement.
- No upfront fee
- Pay only if you hire
- Data centers only, construction to operations
- Local and 100% travel candidates
- Replacement guarantee on standard searches
- InfiniBand / High-Performance Fabric Engineer
What does this hire really come down to?
An InfiniBand or high-performance fabric engineer builds and runs the low-latency network that ties GPUs and storage together. The work covers switch and adapter configuration, subnet management, topology validation, cabling checks, congestion behaviour and firmware alignment. Ethernet specialists can be excellent and still have never watched a job stall because of one bad link.
Distinguish this seat from a data center network engineer. The general engineer keeps traffic flowing across a production network. The fabric engineer proves that a cluster reaches its expected performance, finds the single optic or cable that drags a collective operation, and works with the scheduler team and the vendor.
RoCE over Ethernet versus native InfiniBand is a real divide. Some clusters use both. Name the one you run, because skills transfer only partly and candidates will tell you honestly if asked.
What do we check before you see anyone?
Ask how they validated a fabric before handing it to users. Strong answers include link error counters, bandwidth tests, topology comparison and the order they fixed problems. Ask about the worst intermittent fault they traced. You want the symptom users reported, the evidence they gathered and how they confirmed the fix.
- Ask how they validated a fabric before handing it to users. Strong answers include link error counters, bandwidth tests, topology comparison and the order they fixed problems.
- Ask about the worst intermittent fault they traced. You want the symptom users reported, the evidence they gathered and how they confirmed the fix.
- Ask who they depended on: the cabling crew, the vendor, the job scheduling team. Fabric work is shared, and loners stall.
Where do we find these candidates?
Most fabric engineers come from research computing and national laboratory style environments, from large cloud operators that run training clusters, and from storage or HPC integrators. A few arrive from financial trading networks where latency discipline is similar. People from research computing often know the software side well and have less experience with hall-scale cable plant.
- Most fabric engineers come from research computing and national laboratory style environments, from large cloud operators that run training clusters, and from storage or HPC integrators. A few arrive from financial trading networks where latency discipline is similar.
- People from research computing often know the software side well and have less experience with hall-scale cable plant. Integrators know racks and acceptance testing but may lack long-term operating history.
Which hiring mistakes should you avoid?
Do not treat any Ethernet engineer as a substitute without asking about collective performance tests. Do not ignore cabling and optics, where many cluster faults actually sit. Do not post a role that expects design, hands-on validation and vendor management from one person without saying which matters most.
- Treating any Ethernet engineer as a substitute without asking about collective performance tests.
- Ignoring cabling and optics, where many cluster faults actually sit.
- Posting a role that expects design, hands-on validation and vendor management from one person without saying which matters most.
Your search in 3 steps
- Pin the fabric
We confirm the technology, scale and who owns the scheduler and the cable plant.
- Find validation experience
We screen for people who proved performance, not only those who configured switches.
- Run a fault walkthrough and align on support
Finalists describe one hard fault in detail so you can judge their method. We check on-call, vendor access and site presence before offers go out.
Questions that separate builders from readers
Ask what a healthy fabric looks like on the day of handover and which numbers they would refuse to sign off on. Ask how they would find one failing optic among many identical links. Candidates who have lived this answer with a sequence and a story, while others recite feature names.
Also ask what they do when the vendor and the user disagree about whose fault a slow job is. The answer shows how they carry evidence between teams.
Why the pool stays small
Few organisations run large GPU or research fabrics, so few engineers have operated one at scale. Candidates may be ready to move only when a cluster ends or a lab loses funding. Accept that timing is set by their projects.
Where do we fill this role?
What does BLS publish for this role?
The figure below comes from the national OEWS row for Computer Network Architects. Read the mapping badge before using it in a budget. A broader occupation can include jobs outside data centers. Compare your seat’s responsibilities and pay basis with the published occupation rather than treating the median as an offer quote.
Broader occupation. Computer Network Architects is a broader occupation based on the work described, not a separate data center wage series. BLS does not distinguish this title, experience level, shift, credentials or data center employers. Confirm the actual duties before using this comparison.
| Area | Median annual wage | BLS occupation |
|---|---|---|
| United States | $134,050 | Computer Network Architects (SOC 15-1241) |
Source: U.S. Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025. SOC 15-1241 Computer Network Architects, United States. View this release at BLS.
Not ready to start a search?
We tell you where qualified InfiniBand / High-Performance Fabric Engineer are working in your area and what it takes to move them. No fee, no obligation.
Paste or upload your JD. We flag what will put strong candidates off. No fee, no obligation.
Open roles
Questions
How do I make a fabric engineer brief precise enough to start?
Name the fabric type, the number of GPU nodes or racks, and who owns cabling and the job scheduler. Without them we cannot tell a native InfiniBand operator from an Ethernet specialist. Outreach starts once those facts are written down.
Would a fabric engineer leave in the middle of a cluster deployment?
Rarely, because most want to see their cluster accepted. We discover where they stand in the project and whether a later start date works for you. Accepting a delayed start often produces a better hire.
Will candidates move for a cluster build?
Some will for a large named program with a long run, especially if housing help is on offer. Others want remote support with trips to the hall for acceptance work. Tell us which one the budget supports.
What site time is typical during cluster installation?
Installation and acceptance weeks are heavy, then it settles into occasional upgrades. Give the likely number of stays and say who covers lodging and daily costs. Candidates weigh that against on-call duties.
How can I tell whether someone's fabric experience matches my cluster?
Ask for scale, topology and the workload the fabric carried. Hand them a sample of link error counters and ask what they would check next. A practised engineer reads the output quickly and in a sensible order.
Does pay for fabric engineers distort the search?
Scarce skills draw offers from cloud operators and research sites, so a low range stalls things. Share your range alongside the scope, and we tell you whether they line up. That conversation happens before anyone is approached.
Start your search
Which related roles might you also need?
Related hiring situations
How a search runs
How does The Six-Step Search work?
Every search runs the same six steps. We agree the brief with you, map where the people work, find direct contact routes, reach out personally and repeatedly, then qualify each candidate live. You only see a short list of people who fit, with notes on each, and we stay with you through offer and start.
- 01
Search Calibration
You get: A written brief you sign off on: scope, must-haves, pay band and who not to approach.
- 02
Market Mapping
You get: A map of the companies and projects where your future hire is working today.
- 03
Professional Contact Intelligence
You get: Verified direct contact routes to the people on that map, not job board replies.
- 04
Persistent Personalized Outreach
You get: Repeated, role-specific outreach to people who are not looking, in your words.
- 05
Live Candidate Qualification
You get: Each candidate is screened live against your brief before you hear their name.
- 06
Candidate Intelligence and Process Support
You get: A short list with candidate briefs, then support through interviews, offer and start.
What should you know before you send a role?
What does it cost to hire through Data Center Headhunters?
No upfront fee. You pay only if you hire. Fee terms are set out in the signed recruiting agreement.
How long until I see a short list?
It depends on the role, the location and the market. We agree a realistic timeline with you during Search Calibration and keep you updated as the search runs.
How do you handle counteroffers?
We talk through motivation, pay and counteroffer risk with each candidate during live qualification and again before an offer, so you hear about the risk before it surprises you.
Is there a guarantee?
We offer a candidate replacement guarantee on standard search agreements. The applicable period and exact terms are set out in the signed client recruiting agreement.
What don't you do?
We do not post your job on boards, send stacks of unscreened resumes or charge an upfront fee. We do not approach candidates for your search before the brief is agreed with you.
Need this person?
No upfront fee. You pay only if you hire.
We offer a candidate replacement guarantee on standard search agreements. The applicable period and exact terms are set out in the signed client recruiting agreement.