Who else runs NVIDIA’s repair headache
A landscape for the founders. This report answers one question: which other industries and companies run a broken after-sales-repair process that looks like NVIDIA’s — so we know who to interview. It matches on the shape of the work and where it hurts, not on anyone’s financials. Written to be read start-to-finish; every technical term is explained the first time it appears. Per our research method, it ends in questions and a call list, not a verdict.
How to read this (30 seconds)
The prior report on this topic got lost in warranty accounting — whether each company’s warranty costs were rising or falling. That was a distraction. This one ignores money entirely and asks a simpler thing: does this company run the same kind of repair-and-return process NVIDIA runs, and does it break in the same place? If yes, they’re worth a conversation, because they feel a pain we understand.
The punchline up front: where this pain shows up, it’s almost always in the same spot — not inside any one company’s software, but in the handoffs between companies (the customer, the manufacturer, the dealer or repair partner, the shipping company). Most big companies have decent internal systems. Far fewer have a system that spans the gap to their partners, so that gap gets run on email and spreadsheets. To be precise about how sure we are: we found this pattern stated outright in three industries — hospital equipment, cars, and aviation, each with specific evidence below — and likely by structure in the rest, meaning the multi-company setup is clearly there but we’d need an interview to confirm it still hurts. That’s not a weakness in the plan; confirming those is literally what the call list is for. And to head off the obvious objection — “don’t all companies struggle to share data?” — the specific thing here isn’t generic data-sharing, it’s the repair workflow: reconciling which parts got used, adjudicating who owes for a warranty claim, and tracking one physical unit’s journey against a deadline, across companies that each keep their own books.
Part 1 — NVIDIA’s process, in plain English
First, the vocabulary, defined once and reused throughout:
- RMA (Return Merchandise Authorization): the permission slip a manufacturer issues when a customer wants to send a broken product back. Also the name for the whole return-and-repair process. Say it “R-M-A.”
- Field service: instead of shipping the broken thing back, the manufacturer sends a technician to fix it where it sits (think an elevator or a hospital scanner).
- Depot: a central repair shop that broken units get shipped to.
- CRM (e.g., Salesforce): the software where customer requests — “tickets” or “cases” — are logged and tracked.
- ERP (e.g., SAP): the company’s financial and inventory backbone, where parts, stock, and money are recorded.
- Contract manufacturer (or ODM): an outside factory (Foxconn, Quanta, Wistron) that builds — and often repairs — a company’s products for them.
- 3PL (third-party logistics): the outside shipping-and-warehouse company that physically moves things.
- Spares pool: a shared stock of working units kept on hand, so a customer gets a replacement immediately while their broken one goes off to be fixed.
- SLA (Service Level Agreement): the contractual promise for how fast a repair or replacement will happen (e.g., “30 days”). The clock you’re judged against.
The story of one broken chip. A customer’s AI server fails in their data center. They open a ticket in a web portal. NVIDIA checks that it’s a real, covered unit and decides whether to repair it or just send a replacement. A replacement ships out; the broken unit heads back — usually not to NVIDIA directly, but to a contract manufacturer’s repair line, moved by a shipping partner. The factory fixes it, using a mix of parts NVIDIA owns and parts the factory buys. The repaired unit goes back into the spares pool, ready for the next customer. Then everyone tries to reconcile what happened: which serial number went where, which parts got used, did we hit the SLA.
The six steps, boiled down: (1) customer opens a ticket → (2) manufacturer checks coverage and decides repair-or-replace → (3) replacement ships / broken unit is routed to a repair site or a technician is dispatched → (4) the work crosses several companies, each holding its own piece of the story → (5) parts get consumed and have to be reconciled against the ERP → (6) the fixed unit comes back and the loop closes.
Where it actually breaks. From our NVIDIA conversations, the pain lives in three clusters [Interview: Lonny Orona, 2026-05-12; Alex Zhu, 2026-05-27]:
- No shared record once the ticket is open. The moment the case leaves the CRM, coordination falls back to email threads and spreadsheets. Delivery notes, tracking numbers, serial numbers, ship-to addresses — all updated by hand. The manufacturer’s version of events and the customer’s version drift apart.
- Nobody can measure the loop. There’s no end-to-end dashboard, so nobody actually knows whether they’re hitting the 30-day promise. In NVIDIA’s own words, they “don’t know the score of the game.” Nothing automatically flags a case that’s stuck.
- The other companies are a black box. NVIDIA can’t easily see what the customer, the integrator, or the repair factory is doing. Parts the factory used get reconciled by hand, weeks later.
The unifying thread: the process works fine inside any single building. It falls apart every time it has to cross from one company to another. And it broke because volume grew faster than anyone built the tooling — a process that was fine “when we were a chip company” buckled once returns went from a trickle to a flood.
Part 2 — What we’re actually matching on
Here’s the important shift from the last report. The old report keyed on whether the broken thing was physically removable and shipped to a depot. That filter threw out anything serviced in place — elevators, turbines, hospital scanners.
That was too narrow. Think about a repair technician dispatched to fix a machine on-site. Nothing gets shipped anywhere — but there’s still a ticket opened in a CRM, still a coverage check, still parts pulled and reconciled against an ERP, still a handoff to a service partner, still an SLA clock ticking. The workflow and its pain are the same whether the unit comes back or the technician goes out. So field service is fully in scope, and that roughly triples the number of industries worth looking at.
So the real match criteria — all about process, none about money — are:
- Does the company run a ticket-driven repair/return process that has to bridge a customer system (CRM) and an inventory system (ERP)?
- Does that process cross several organizations (customer, manufacturer, dealer or integrator, repair partner, shipper)?
- Is the coordination between them manual — email, spreadsheets, phone — and unmeasured end-to-end?
And one crucial test we apply to every industry: “mature” does not mean “solved.” An old, established company can still run this on spreadsheets. The question isn’t “have they been doing repairs for 40 years” — it’s “have they actually connected the systems across their partners, or does a person still stitch it together by hand?” Almost universally, we found the answer is: the inside is modern, the seams between companies are not.
Part 3 — The industry landscape
Below, grouped from closest-to-NVIDIA outward. For each: does the process match, is the pain real, is it genuinely solved or just old-and-still-manual, and who to call. Every company named here is a cold prospect — we have no existing relationship in any of these industries, which is itself worth noting.
Group A — The near neighbors (high-value electronics and equipment)
1. Data-center and enterprise IT hardware — servers, storage, networking gear (Dell, HPE, Cisco, Lenovo, Supermicro, Pure Storage, NetApp, Arista). This is NVIDIA’s own neighborhood and the closest match by far. The process is identical: web RMA portal, coverage check, cross-ship a replacement, route the dead unit to a repair depot that’s often run by the same contract manufacturers NVIDIA uses. Is it solved? The giants (Dell, HPE, Cisco) have been service companies for decades and are likely further along than NVIDIA — which makes them a great “is maturity actually the answer?” test. The sharper pain is probably at the faster-growing, younger hardware companies (Supermicro, Pure, Arista) that, like NVIDIA, grew up as product companies and bolted on a returns operation later. Who to call: at Dell/HPE/Cisco, a Director of Reverse Logistics or Depot Repair Operations; at a fast-grower like Supermicro, the Head of Customer Service Operations. Why similar: same outsourced, multi-party repair loop. Why different: the incumbents may already be where NVIDIA wants to get.
2. Telecom / network equipment — the gear that runs phone and internet networks (Ericsson, Nokia, Ciena, plus the carriers like Verizon and AT&T). Strong match, with a twist: carriers keep spares pre-positioned near cell towers so a technician can swap a failed radio in hours. The interesting pain sits with the carrier, not the manufacturer — a big carrier runs equipment from five different makers and has to reconcile repairs across all of them and several independent repair shops. There’s even a niche tool (Flow Catalyst) built just to manage multi-vendor spares and repairs — a tell that the big systems don’t cover it. Who to call: a VP of Network Field Operations at a Tier-1 carrier (likeliest acute pain), or a Head of Spares & Repair Operations at Ericsson/Nokia. Why different from NVIDIA: the worst pain is felt by the multi-brand operator, not the single manufacturer.
3. Medical devices and imaging — MRI/CT scanners and hospital equipment (GE HealthCare, Siemens Healthineers, Philips).
This turned out to be the strongest non-NVIDIA pain match in the whole scan. A scanner fails; often the hospital’s own in-house technicians look first, then the manufacturer sends a field engineer, while individual components (coils, X-ray tubes) get pulled and swapped from a parts depot. Three or more organizations each hold a piece: the hospital’s biomedical team, the manufacturer, and often an independent multi-brand service company. And despite this being a mature, heavily-tooled industry, the record-keeping across those parties is still manual — one industry source estimates hospitals lose about $2.3M a year each to duplicate manual data entry between their maintenance system and their ERP, with an average hospital juggling 35,000+ devices [Public: Oxmaint, 2025]. That’s NVIDIA’s exact pain — a manual bridge between the ticket system and the inventory system — in a completely different industry. Who to call: a VP of Service Operations at GE HealthCare or Siemens Healthineers; and separately, a Director of Clinical Engineering at a large hospital network — the multi-brand buyer who feels the cross-company mess most acutely. Why it’s a top target: mature industry, genuinely unsolved seams.
4. Semiconductor manufacturing equipment — the machines that make chips (Applied Materials, Lam Research, KLA, ASML). Partial match — and a useful cautionary case. The machines are bolted into a customer’s chip factory and serviced in place, but their sub-modules get pulled and refurbished, so there’s a real parts loop. However, these companies treat service as a core business (it’s a huge share of their revenue), run deeply on SAP, and have even automated their parts warehouses with robots. They look similar but may be substantially solved — worth one interview mainly to confirm that and calibrate. Who to call: a Director of Service Parts Operations at Applied Materials. Why cautious: expect “we’ve mostly handled this.”
5. Industrial automation and machine tools — factory robots, controllers, CNC machines (ABB, Fanuc, Rockwell, Siemens, plus machine-tool makers like Mazak and DMG Mori). Solid match as a field-service business, with an extra wrinkle: much of this gear is sold and serviced through system integrators — middlemen who own the customer relationship, so the manufacturer often can’t see the field ticket at all. The biggest makers may be well-tooled (many use Syncron for parts). The sharper pain is likely at the machine-tool builders and across the integrator channel. Who to call: a Director of Customer Support & Field Service at Rockwell (heavy integrator channel), or Head of Service at a machine-tool builder. Why similar: the integrator is an extra organizational seam, just like NVIDIA’s system integrators.
6. Scientific and lab instruments — mass specs, analyzers, lab equipment (Thermo Fisher, Agilent, Danaher, Waters). Match, but likely the least painful in this group. These run on service contracts with guaranteed response times, so the front-end (coverage checks, dispatch) tends to be well-tooled — that’s their revenue engine. The open question is the parts-and-refurb loop and any multi-brand service they do. Who to call: a Director of Field Service Operations at Thermo Fisher’s service arm. Why cautious: front end probably solved; probe the parts loop.
Group B — The field-service world (units fixed in place or through dealers)
These are the industries the old report wrongly excluded. Nothing gets shipped back — but the same ticket-to-closeout workflow runs, and the same seams break.
7. Aviation maintenance (MRO) — keeping aircraft flying (airlines like Delta and United; engine and part makers; independent repair shops like AAR and StandardAero). MRO means “maintenance, repair and overhaul” — the aftermarket service industry for aircraft. This is the textbook multi-party version: a part is swapped at the gate, a shared pool provides the spare, the failed part goes to a shop, and warranty claims bounce between the airline, the manufacturer, the part supplier, and the repair shop. Industry marketing describes the pain almost word-for-word as NVIDIA’s — “claims tracked in spreadsheets, emails, and memory.” Is it solved? The systems inside each company are mature (aviation invented a lot of this). The between-company claim reconciliation is where it’s still manual. Who to call: a Director of Warranty & Supplier Recovery at a major airline’s maintenance division, or a VP of Aftermarket Services at an independent shop like AAR. Note: this same pattern extends to military aircraft sustainment, which is a documented bullseye on the process — but that’s a government buyer, and per our standing policy I’m flagging it for you to decide, not pursuing it.
8. Heavy equipment, farming, and construction machinery (Caterpillar, John Deere, Komatsu, CNH). Strong match with a distinctive extra player: the dealer. When a bulldozer or tractor breaks, the customer calls an independent franchised dealer — not the manufacturer. The dealer files the warranty claim, holds its own records, and the manufacturer later tries to recover costs from the part supplier. So there’s a three-way reconciliation (dealer ↔ manufacturer ↔ supplier) that’s a known friction point. The parts side is industrialized (mature core-exchange programs, often on Syncron/SAP), but the dealer-to-manufacturer claim adjudication is the likely still-manual seam. Who to call: a Director of Warranty Operations or VP of Dealer Systems at Caterpillar or John Deere. Why similar: the dealer is exactly the kind of extra organization NVIDIA struggles to see across.
9. Automotive and EV service (Ford, GM, Stellantis; and EV charging networks like ChargePoint and EVgo). Very strong match — the industry describes its own pain in NVIDIA’s terms. A warranty claim can touch five parties (customer, dealer, manufacturer, part supplier, sometimes a third-party administrator), and one industry source notes most manufacturers “only have a process for two of them,” with claims processors manually pulling data from five systems to validate one claim. That a whole crop of warranty-automation startups exists (WarrCloud, NextGen) tells you the built-in tools don’t close it. EV charging is a separate, younger pain: an operator runs thousands of scattered charging stations, and without a system tracking actual repair response against the SLA, they can’t even prove uptime — reported reliability far overstates reality. Who to call: a Director of Warranty Operations at a car maker; or a VP of Network Reliability at ChargePoint/EVgo. One difference to note: the charging operator owns its own stations, so it’s fixing its own assets rather than a customer’s — a slightly different shape.
10. Elevators, escalators, and building systems — elevators plus HVAC, fire, and security systems (Otis, KONE, Schindler, Johnson Controls, Carrier).
Pure fix-it-in-place field service, at enormous scale — Otis alone maintains over 2.4 million units. A fault is reported, a technician is dispatched, parts are ordered, a repeat visit closes it, all against an SLA. Is it solved? The leaders have invested heavily (Otis runs on SAP with dedicated field apps) and may be closer to a closed loop than NVIDIA. But building systems often run through mechanical contractors, re-adding the multi-company seam, and the mid-market contractor tail is largely still manual — only about half of service organizations use field-service software at all, dropping to under a third for small firms [Public: FSM market stats, 2025]. Who to call: a VP of Service Operations at Otis or KONE (test whether they’ve truly closed the loop), or a Head of Connected Service at Johnson Controls. Watch-out: the still-manual pain concentrates in smaller firms that aren’t NVIDIA-scale buyers.
11. Power, energy, and oil-field equipment (GE Vernova and Siemens Energy turbines; SLB, Halliburton, Baker Hughes oil-field tools). Two different shapes here. Big installed turbines are serviced in scheduled on-site “outages” under long-term contracts — a planned version of the loop. Oil-field rental tools are a closer match: tools are rented, used, returned, and “redressed” (inspected and rebuilt for the next job) across many field locations — a genuine return-and-reconcile loop. There’s even a manufacturer joint venture (Aero Alliance) where two companies literally share the record by design. Who to call: a VP of Rental Tools at SLB or Halliburton (the closest match), or a Director of Outage Services at GE Vernova. Why interesting: rental-tool reconciliation across scattered sites is a candidate for manual mess.
12. Distributed device fleets — ATMs, checkout/point-of-sale terminals, kiosks, office printers and copiers (NCR, Diebold Nixdorf, Xerox, Ricoh, Canon). One of the cleanest matches. Thousands of identical machines are scattered across retail and office sites; when one breaks, it’s triaged remotely, then either a technician is dispatched (often from an outside field-service firm, not the manufacturer) or the unit is swapped and sent to a repair depot. The manufacturer frequently never touches the unit — an outside firm does, holding its own records. The tooling to close this loop exists and is actively marketed (ServiceMax sells exactly this), which cuts both ways: some fleets are solved, but the heavy marketing of “visibility across partners” implies many aren’t. Who to call: a VP of Global Field Service at NCR or Diebold Nixdorf; a Director of Managed Print Services at Xerox/Ricoh; or the Head of Operations at a third-party field-service firm (they see the coordination gaps across all their clients). Why valuable: the third-party firm may already be the shared record the manufacturers lack — worth learning who owns the truth.
13. Rail and transit — trains and signaling (Wabtec, Siemens Mobility, Alstom). Match via “unit-exchange”: a failed component (a traction motor, a brake module) is swapped for a rebuilt one, and the failed one goes to an overhaul shop. The org chart is rich — the train operator, a leasing company that owns the trains, the manufacturer or an independent maintainer, and parts suppliers. Condition-monitoring platforms exist, but the multi-owner claim-and-parts reconciliation is where manual work likely persists. Who to call: a Director of Component Overhaul Operations at Wabtec, or a Head of Fleet Maintenance Services at Siemens Mobility or Alstom. Note: Alstom and Siemens make good money on this service, so parts of it may be “solved and monetized.”
Part 4 — Three patterns that cut across everything
1. Where the pain is real, it’s at the seam between companies, not inside them. In the industries where we saw direct evidence (hospitals, cars, aviation) — and, we suspect, in the others by structure — the system inside each company (the airline’s maintenance software, Caterpillar’s dealer system, Otis’s SAP) is far more mature than the handoff between companies (airline↔manufacturer↔supplier claims; dealer↔manufacturer↔supplier). And “the handoff” specifically means the repair-workflow handoff: reconciling parts used, settling warranty claims, and tracking a unit against its deadline. If there’s a wedge here, it’s that between-company reconciliation, which the big single-company software doesn’t span. This is the same thing that breaks at NVIDIA. The honest bound: it’s confirmed in a few industries and a hypothesis to test in the rest — which is why Part 5 pairs an expected-”still-manual” call with an expected-”already-solved” one.
2. “Who feels the pain” isn’t always the manufacturer. At NVIDIA, the manufacturer feels it. But in telecom, healthcare, and device fleets, the worst pain is felt by whoever holds many brands at once — the carrier, the hospital network, the third-party service firm. That’s a different kind of customer with a different budget. Worth deciding early: are we selling to the manufacturer, or to the multi-brand operator in the middle?
3. When a niche tool exists, the big systems haven’t closed the loop. Everywhere we found a small, purpose-built tool (multi-vendor spares in telecom, maintenance-to-ERP connectors in hospitals, warranty-claim startups in automotive), it was a signal that the giant incumbents (SAP, ServiceMax, Syncron) leave the cross-company gap open. That gap is the recurring white space.
Part 5 — Who to interview first
If the goal is conversations, here’s where the learning-per-call looks highest:
- A large hospital network’s Director of Clinical Engineering — the strongest evidence of an unsolved, multi-party, mature-industry pain. Expect “yes, it’s still manual, and it’s a nightmare.”
- Applied Materials service operations — the disconfirming interview. Expect “we’ve mostly solved this.” The gap between this answer and the hospital answer is the real signal about where the opportunity is.
- A fast-growing IT hardware company (Supermicro, Pure, Arista) service lead — the closest NVIDIA-twin: a product company that grew into a returns problem.
- A car maker’s or heavy-equipment maker’s warranty operations director — to understand the dealer-in-the-middle seam, which is the richest multi-party version.
- A Tier-1 telecom carrier’s field operations VP — to test the “multi-brand operator feels it worst” pattern.
A useful pairing: run interview #1 (expected “still manual”) against #2 (expected “solved”) back-to-back. Whichever way the surprise falls is the most valuable thing we could learn.
Part 6 — Honest caveats and open questions
- We have zero existing relationships in any of these industries. Every name here is a cold prospect. The whole point of this report is to start those conversations, not to conclude anything from a desk.
- “Mature-but-manual vs. genuinely solved” is a judgment we made from the outside. For most industries it needs one real conversation to confirm. We flagged our best guess for each; treat it as a hypothesis to test, not a fact.
- The manufacturer-vs-aggregator question is unresolved and matters a lot. It changes who we sell to and what we’d build. It’s answerable with a couple of targeted interviews (one manufacturer, one multi-brand operator per industry).
- Defense aircraft sustainment is the single closest process match we found — but it sits behind a government buyer, so per our standing policy it’s flagged here for a human decision, not pursued as a research direction.
- This is process-similarity only. We deliberately said nothing about market size, willingness to pay, or economics. A similar process is a reason to have a conversation — it is not, by itself, a reason to build.
What would change this picture: any single interview that reveals an industry we tagged “still manual” has quietly solved the cross-company loop (kills it as a target), or that one we tagged “solved” is secretly a mess (promotes it). Those surprises are the goal.
Sources: the reusable process pattern is drawn from the NVIDIA reverse-logistics interviews Lonny Orona, 2026-05-12 and Alex Zhu, 2026-05-27; the internal software landscape rma-reverse-logistics-warranty-software-landscape-2026-07-08; and public/industry sources on each industry’s service operations (labeled inline). Companion (financial-lens) scan: nvidia-like-rma-pain-target-scan-2026-07-13 — deliberately answers a different question.