Datacenter GPU Repair -- Japan

H100 failed?
We fix what NVIDIA won't.

BGA rework, HBM reflow, VRM replacement on H100 and H200 boards. Fixed pricing, 1–2 week turnaround, 90-day warranty.

Get a Quote How It Works

What We Repair

We diagnose to the component level and fix what's actually broken. No board swaps.

NVIDIA H100 SXM5 NVIDIA H100 PCIe NVIDIA H200 SXM5

HBM Memory Defects

HBM3/HBM3e stack failures -- ECC errors, bandwidth degradation, or complete module death. We do BGA rework to replace individual HBM stacks.

VRM Failures

Blown MOSFETs, dead power stages, or voltage regulator faults. Usually shows up as intermittent crashes or the board not posting at all.

Thermal Damage

Thermal cycling cracks solder joints and delaminates substrates over time. We X-ray to find the fractures, then reflow or reball the affected area.

PCB Trace Issues

Cracked or severed traces from mechanical stress or corrosion. Micro-soldering and trace jumper repair under microscope.

Connector Damage

Bent pins, cracked housings, or lifted pads on NVLink, NVSwitch, or power connectors. Common after repeated insertion or from thermal expansion.

We work on H100 and H200 only. No A100 or older models, no B200/B100 (those are still under NVIDIA warranty), and no boards with catastrophic physical damage -- cracked substrates, severe corrosion, etc. We'll tell you upfront if a board isn't repairable.

Your GPU Died. Now What?

NVIDIA won't repair H100s or H200s. A replacement costs $25K+ and takes weeks to arrive -- if it's even in stock. You need that node back in production. Here's how repair compares.

Repair with Ensei Buy Replacement
Cost $2,000 – $8,000 $25,000 – $40,000+
Timeline 1–2 weeks 2–8 weeks (procurement)
Warranty 90-day repair warranty Standard NVIDIA warranty
Availability Immediate — ship your unit Subject to supply / allocation

Repair runs 10-25% of replacement cost. Even if you're ordering a new unit, repairing the dead one gives you a working spare or a board you can resell.

How It Works

Ship it to us or we come to you. Two paths, same result: working GPU.

Ship-to-Us

01

Tell Us What Failed

GPU model, how many, what symptoms you're seeing. We get back to you within 1 business day with next steps.

02

Ship It Over

Send the board to our Japan facility. We accept shipments from anywhere.

03

We Diagnose and Fix

Full diagnostics in 1-3 days. We send you a report with what we found and a quote. Repair takes 3-7 days after your go-ahead.

04

You Get It Back

Repaired, burn-in tested, shipped back. 90-day warranty on the repair.

On-Site Repair

01

Tell Us What Failed

GPU model, quantity, symptoms, and your location. We get back to you within 1 business day.

02

We Schedule a Visit

Pick a time that works for your ops window. We come to you -- available globally.

03

Diagnose and Fix On-Site

Our engineer works at your facility. No shipping, no waiting for return logistics. Boards stay in your datacenter.

04

Test and Hand Off

Validation testing on your hardware, on-site. 90-day warranty on every repair.

Not sure what's wrong? Send it for diagnostics only ($350). You get a full failure report with X-ray images -- no obligation to repair.

Running 100+ GPUs?

You already know the failure rate on a fleet this size. Stop scrambling after each one. Get a maintenance contract and have a repair path ready before the next board goes down.

Maintenance Contracts

Pre-negotiated rates, priority queue, and a known repair path for your fleet. When a board dies at 2 AM, you already know the process.

Volume Discounts

10–49 units: 15% off. 50–99 units: 20% off. 100+: custom pricing.

Priority Turnaround

Contract boards go to the front of the line. You get your GPU back faster because we already know your hardware.

On-Site Maintenance

Our engineer works in your datacenter. No shipping delay, no chain-of-custody headaches. Available globally.

Talk to Us About Fleet Pricing

Frequently Asked Questions

How much does repair cost?
$2,000-$8,000 depending on what failed. HBM rework is at the higher end, connector repair at the lower. You get an exact quote after diagnostics -- no surprises.
What's the turnaround time?
1-2 weeks total. Diagnostics take 1-3 business days. Once you approve the quote, repair is 3-7 business days. Priority queue available for fleet contracts.
What if you can't fix it?
You pay the $350 diagnostic fee and nothing else. We send you a report showing what failed and why it's not repairable. No repair charge.
Do you offer on-site repair?
Yes. Our engineer comes to your datacenter with tooling and parts. No shipping, no chain-of-custody concerns. Available globally -- tell us where you are and how many boards need work.
Do you ship internationally?
Yes. You cover inbound shipping to Japan. We cover return shipping. We handle customs documentation for GPU hardware.
What warranty do you offer?
90 days on every repair. If the same fault comes back within that window, we fix it again at no charge. You also get a detailed repair report with X-ray images and test results.
Which GPUs do you repair?
H100 SXM5, H100 PCIe, and H200 SXM5. That's it. No A100 or older, no B200/B100 (still under NVIDIA warranty), no consumer cards.
Who is Ensei?
Ensei Limited is a Hong Kong-registered company with repair operations in Japan. We do board-level repair on datacenter GPUs, with direct access to component supply chains in China for parts sourcing.

Tell Us What Died

Fill this out and we'll get back to you within 1 business day with next steps and a preliminary estimate.

Rather just email? repair@ensei.dev