I have unboxed a lot of miners. When you power up a fresh pallet of ASICs, you learn fast that they are not all equal. Two machines with the same model number, made in the same factory, can act like two different animals. One runs cool and steady for years. The other throws hashboard errors in the first week. That gap is not bad luck. It is the price of making chips by the millions, and it is the reason bad batches are part of this business.
People throw the phrase around like a curse word, but it has a real meaning. A bad batch is a group of miners, usually from the same production run, that fails at a much higher rate than normal. Sometimes it is one weak part shared across the whole run. Sometimes it is a design change that looked fine on paper and cooked itself in the field. And sometimes it is the silicon itself, born flawed on the wafer before anyone ever screwed it into a hashboard.
To understand why this keeps happening, you have to follow the chip from sand to steel container. Every step adds a chance for something to go wrong, and mass production multiplies that chance by the thousands.
An ASIC begins as a silicon wafer. The silicon has to be almost perfectly pure, on the order of nine nines, which is 99.9999999 percent. Then it gets patterned with light in machines that align features down to a few nanometers. A nanometer is smaller than most viruses. At that scale, a single speck of dust, a tiny misalignment, or a bad spot in the silicon can kill a chip. Worse, one defect can kill several chips packed next to each other on the same wafer.
This is why yield matters so much. Yield is the share of good chips you get off a wafer. The smaller and more advanced the chip, the harder it is to keep yield high, because there is simply more that can go wrong in a tighter space.
Now picture a full wafer. The good dies get used. The dead ones get scrapped. The map below is what that really looks like, and notice where the losses cluster. The edges of a wafer are the hardest place to get right, so failures pile up there first.
Even among the chips that pass, no two are truly identical. Small differences in doping and patterning mean some chips run fast at low voltage while others need more power to hit the same speed. Manufacturers test each chip and sort it into a bin. The best chips go in the top bin. The weaker ones go lower, and the failures get tossed.
Miners call this the silicon lottery. It is the reason two machines with the same sticker can post different hashrate and different power draw right out of the box. A batch loaded with low-bin chips will run hotter, pull more watts, and wear out faster, even if every unit technically passed the test.
Operator note. A cheap batch is not always a deal. If it is packed with low-bin silicon, you pay the difference back in power, heat, and early failures over the life of the machine.
Electronics fail in a pattern that engineers call the bathtub curve. Failures are high at the very start, drop to a low and steady rate for the long middle of life, then climb again as parts wear out at the end. The early spike has a name, infant mortality. These are the weak units that were going to fail no matter what, and they fail fast.
Good factories try to catch these before shipping. They run burn-in testing, where chips are pushed hard at high heat for hours or days to force the weak ones to break early. But burn-in costs time and money, and when demand is red hot and everyone wants machines yesterday, testing is the first corner that gets cut. Skip enough of it and the infant mortality that should have died in the factory shows up on your floor instead.
Not every bad batch starts on the wafer. Some start in the design room. A well known example came when miners flagged problems with a run of popular Antminer S19 units. According to reporting at the time, the maker had made a few changes to cut cost and speed up production. Components were bunched onto one side of the board, traditional circuit board material was swapped for aluminum plating, and a small controller that let the system manage each hashboard on its own was removed.
Each change sounds minor. Together they were not. The tighter layout and the aluminum trapped more heat, which raised the risk of overheating. And because the per board controller was gone, a fault on one hashboard could take down the entire unit instead of just one section. The company that flagged it said it might have simply gotten a bad batch, and it noted a rule that seasoned operators live by. You do not buy the very first production run of a new model. You let someone else find the flaws first.
On a spreadsheet the machines look the same. On the floor they run hot and short.
Here is the part that hurts the most. A defect that started as a tiny flaw on a wafer does not stay small. It rides the whole supply chain and lands at scale. When you deploy hundreds of units from the same run into the same containers, they share the same weakness, and they tend to fail together.
The path is simple. A shared wafer defect or design change gets built into one production run, ships as one large order, gets racked in the same containers, and then fails in clusters that hurt your uptime all at once.
A cluster failure is worse than random failures spread out over time. It hits your uptime all at once, it floods your repair queue, and it can trip questions about who is responsible, the maker, the host, or the buyer. That last question is where these stories often end up in a dispute.
You cannot stop the physics of manufacturing, but you can protect yourself from it. This is the routine we run so a bad batch gets caught in week one and not month six.
Never buy the first production run of a brand new model. Let the early flaws surface on someone else's floor.
Run your own burn-in before full deployment, and log the failures instead of trusting the factory sheet.
Track hashboard failure rates, PSU swaps, and fan replacements by batch, not just by site total.
Compare pool reported hashrate against nameplate hashrate, unit by unit, to spot low-bin silicon.
Watch inlet temperature and voltage. Weak bins show up as extra heat and higher power draw.
Keep the serial numbers and purchase records clean, so a warranty or legal claim has real evidence behind it.
The bottom line. Bad batches are not a rumor and they are not always someone cheating you. They are a normal result of making chips at massive scale on the hardest nodes in the world. The operators who win are the ones who expect it, test for it, and document it.
When a bad batch turns into a lawsuit, the fight is almost always about the same thing. Did the machines fail because of the hardware, or because of how they were run? A theoretical expert can describe how a chip works. That is not the same as knowing what a real failure pattern looks like across hundreds of units in a live facility, or being able to read the daily operational data and tell the story it holds.
I have reviewed damage models built on best case assumptions, with perfect uptime and no attrition, that fell apart the moment real operational facts were applied. A bad batch case is won or lost on facts like burn-in records, batch level failure rates, and honest uptime data. Those facts come from the floor, not from a slide deck.
Aviran Vargas is the Director of Operations and Infrastructure at EZ Blockchain, a Chicago based Bitcoin hosting company running mining infrastructure across multiple U.S. sites. He has more than 15 years in data center and critical infrastructure operations, manages 60 MW of mining capacity, and serves as a crypto hosting expert witness. If you are facing a mining hardware or hosting dispute, you can reach him through BTCExpertWitness.com. He works both plaintiff and defense sides.