What are the "fake" ones, exactly? Is it some other chipset altogether? Not 10GbE?
I discussed the rather wide range of possibilities in
this post.
I believe most of the 'counterfeits' use a real chipset or one that didn't pass QC, and then the other components are equivalents or not as stringent as the original.
I'm not sure about the "didn't pass QC", for a couple of reasons. First, Intel must have some testing done at the component level once the chip is bonded and the package sealed (in addition to the many tests done during the wafer process) and the rejects would never be shipped. If there were enough finished board duds that they made up a good portion of the many non-genuine cards, Intel would look at the whole process - final test at chip packaging, transport damage on the way to the board assembly site, errors at the assembly site, and design problems resulting in a marginal product. And once they had a pile of duds, they would probably make the decision to not rework the boards to change out the actual chip just to re-use the dud on a counterfeit board. Aside from likely being the most expensive component on the board, rework plays havoc with production - a regular assembly line has a constant flow of product through various stages, with those stages designed so nothing "backs up" previous stages. So there might be 3 Step C lines feeding 2 Step D lines, if D takes only 2/3 the time of C.
But rework is incredibly variable depending on the number of duds (one thing no factory wants is a steady supply of duds) and is a much more manual process than regular assembly, even if the fault has been conclusively proven to be in a specific component. Yes, labor is cheap there, but a fast, predictable process still takes precedence. And you have to reball the pads on the chip without any guarantee of success using it in an automated production system for the fakes. Far easier to take completed boards that flunked QA and scrap them. Particular at this price point for a single card - if it was a $10,000 Cisco router that would be another issue entirely).
This may involve dishonesty on the part of the assembler ("yessir, we scrapped those! <wink>") or a contracted recycling company who was supposed to scrap them but instead sold them. With the tangled chains of ownership of many of these companies, it becomes almost impossible to determine who is ultimately responsible for the diversion of the supposedly-scrapped boards. Intel may have found a problem with either the chip fabrication process or the mask (for example, the Sandy Bridge SATA issue) and declared a large pile of chips bad, but that wouldn't explain the ever-increasing number of fakes in the market, nor how all of the duds made it somewhere they could be put on boards (most of the SATA bug chips never made it that far, IIRC).
While I haven't visited the Asian factories involved, I
have been involved in rework at various levels (from being the "change-the-chips guy" to upper management) for both factory and 3rd-party rework operations, and I still have a complete, functional SMD rework system at home. So I'm very familiar with the process.