Last April, I sat in our network operations room watching our senior engineer swap out a dead Infinera G42 transceiver. It was the third one that week from the same batch. The batch I'd sourced. He didn't say anything—the click of the module unlatching was judgment enough.
He pulled out an Infinity Pro multimeter to check the voltage at the switch port. Clean. That ruled out the power supply and the line card. The problem was the module itself. The module I'd bought to save us money.
I've been procurement manager at a 40-person regional ISP for six years now. Our network infrastructure budget runs about $800,000 a year, and I've negotiated with twenty-plus vendors over every chunk of it. I built our internal cost tracking system myself, mostly because I got burned by hidden fees twice early in the job and decided I wasn't going to let that happen again. I like to think I do the math before I spend.
But this one, I did the math on the wrong things.
The Deal That Looked Too Good to Be True
It started pretty simple. We were expanding our backbone capacity, and the engineer's plan called for 50 Infinera G42 transceivers. The line cards support both Infinera-branded optics and third-party compatible modules. Our engineer has been around long enough to know that compatible doesn't mean garbage—he just also knows it means nothing until you qualify it.
I ran the comparison. The original Infinera G42 came in at $2,200 per unit. Vendor A quoted $1,350. Vendor B came in at $980.
When I first started buying compatible optics, I assumed the lowest quote was the best choice. Actually, I didn't even think of it as an assumption—it was just how procurement works, right? Three budget overruns later, I know better.
Vendor B's quote was about 55% cheaper than the branded units. On a 50-unit order, that's around $61,000 in paper savings. For context, that's basically our entire power budget for one of our smaller sites for a year. It felt like a win before I even signed anything.
I asked their sales engineer about testing. 'Fully tested,' he said. I asked about warranty. 'Three years.' Then I read the actual warranty language: it covered dead-on-arrival units only, not units that failed while in service. That was a red flag I should have taken seriously. But the savings were so persuasive that I rationalized it. DOA coverage is standard, right? If the module survives the initial burn-in, how bad could it be?
I signed the PO.
Pilot Phase: Everything Looked Fine
Look, I'm not reckless. I convinced the engineers to run a pilot before we deployed all 50 units. Ten modules went into lower-priority links at two sites. They came up clean. Our switches—we run a mixed fleet, including Cisco—recognized them without a single protocol complaint. DOM readings were within spec. Optical power was in the right ballpark. After two weeks, the engineering team signed off and we deployed the remaining 40 units.
For the first month, I was honestly feeling good. The network was stable. Traffic was humming. I was already spending those savings in my head.
Then module #17 died during a routine firmware update. Not a graceful failure—just dead. Then modules #22 and #31 failed within the same maintenance window, and the NOC started seeing flapping links on one of the backbone paths.
The engineer started digging. He used the Infinity Pro to verify the power rail on the line card—clean. So we knew the switch side was fine. But then he spotted it: several modules' TX power had drifted. The transceivers' internal diagnostics still reported normal values, but the actual light output was far below spec. They'd learned to lie about their own health. The only reason we caught it was because link margins started dropping on monitoring.
So the Infinity Pro helped us narrow it down, but the real diagnosis came from comparing the module's reported numbers against actual optical measurements. A multimeter can't catch everything, and neither can a diagnostic register.
Within 90 days, eight of the fifty modules had either failed outright or degraded enough to pull out of production. Sixteen percent failure rate.
The TCO Math
When I got over being defensive, I pulled up our cost system and ran the full TCO on that order, the same spreadsheet I use for every major purchase.
- 8 failed modules at $980: $7,840 sunk
- 5 truck rolls at $300 each: $1,500
- Expedited replacement shipping: $480
- Engineer time, roughly 14 hours at our loaded rate: $1,190
That's $11,010 in direct failure costs. Actually, call it $11,000. In the middle of the spreadsheet, I stopped and just stared at the number for a minute. It wasn't the biggest line item we've ever had. But I knew I'd caused it by ignoring the points that should have disqualified Vendor B from the start.
Now the 'savings' start looking a lot different. $61,000 paper savings minus $11,010 in direct failure costs gets you to about $49,990. Still positive. Still a number that looks like a win.
But then you add the softer costs: two customer-facing links degraded during peak hours. We had an account manager on the phone asking why their site was crawling. That one I can't put a nice spreadsheet number on, but I can tell you it cost us more than the math suggests. Trust burns fast.
And here's the thing: that $49,990 assumes zero further failures. But the batch was statistically suspect. We had another 42 modules from the same production run. I spent six months waiting for the other shoe to drop. The gap between paper savings and actual savings was way bigger than I expected, and the gap between actual savings and 'was this worth it?' was bigger still.
The cheaper option resulted in an $11,000 lesson, plus some relationship damage, plus a procurement process that now questions every 'discount' I bring to the table. It took three years and about 150 orders for me to understand that vendor relationships matter more than vendor capabilities. Vendor B had a great price and absolutely no engineering depth behind it.
Per FTC guidelines (ftc.gov), advertising claims have to be truthful, substantiated, and not misleading. 'Fully compatible' is exactly the kind of phrase that deserves scrutiny. I learned to ask for the qualification report, not the datasheet. If a vendor can't show you test results on the actual Infinera platform—with real capture files and QA sign-off—then their compatibility claim is just marketing.
The Switch Question: Same Logic, Same Blind Spot
Around the same time, our CFO asked me the classic question: 'switches vs Cisco—why are we still paying the premium?' I actually had the same blind spot on that one.
We compared a white-box switch at roughly half the price of the Cisco equivalent. The hardware spec sheet was comparable. The features, on paper, were comparable. But our engineers knew the Cisco CLI in their sleep. They could troubleshoot an issue in 20 minutes. With the white-box alternative, they'd be digging through vendor-specific commands, forum threads, and support tickets that take two days to get an answer.
That's not a knock on white-box switching. At least, that's been our experience with our team size. If you have a big team with deep expertise, the math can tip toward white-box. For a lean shop like ours, the Cisco premium was the cheaper option. Same logic as the optics decision. It's not about which vendor is 'better.' It's about what you're capable of supporting and what every hour of your engineers is worth.
The question everyone asks is 'which is better, switches or Cisco?' The question they should ask is 'what will this cost me over three years, including engineering time and downtime risk?'
How We Buy Now
Since that project, we've changed our approach. This is the policy that's held up for the past 10 months.
- Minimum three quotes. We compare at least three vendors on a full TCO basis, including warranty terms and RMA process, not just per-unit price.
- Real test reports. The vendor has to prove compatibility with the exact platform we're using. Screenshots and capture files from real equipment, not a generic test board. If a vendor hesitates, we walk.
- Warranty language is a deal-breaker. DOA-only warranty is disqualifying. We want coverage for in-service failures and a clear RMA process.
- Batch traceability. We ask for confirmation that all modules in our order come from one qualified batch. No Frankenstein assortments.
The goal isn't to be difficult. It's to make sure we're comparing the full cost of ownership, not just the line item. I'd rather spend 10 minutes explaining this to a new vendor than deal with mismatched expectations later. An informed customer asks better questions and makes faster decisions.
So what would I tell someone in my position? Don't buy the cheapest transceiver. Buy the one with the most evidence behind it. And when you find a vendor that provides real test data and stands behind their warranty—those are the relationships worth keeping. That's the kind of supplier who actually saves you money, even if their price isn't the lowest.
Bottom line: the savings on compatible Infinera optics are real. But you have to earn them. The price quote was 40% lower. The real cost of owning that batch was a lot closer to what we were trying to avoid. That's the number I wish I'd seen before I signed the PO.
