"We believe POET’s optical engines will open the possibility of creating new products where electronics photonics convergence is successfully achieved at an advanced level of integration between InP and Si-based interposers, which take us and our valued customers into the next generation of data networking for AI and hyperscale data centres,” Yasuhiro Yamauchi, general manager of the optical device department at Mitsubishi Electric, said in a statement. “This promises to be a momentous technological achievement when it is unveiled.”

Let's look at all the Hyperscalers that are not Mitsubishi customers, do you think they will sit on their hands and let their competitors run away with less cost per bit on the throughput? Less energy consumed per calculation (picojoules). Nobody has a 3.2 transceiver in the same form factor as a 100g transceiver. Sure the infrastructure will have to catch up, but the driving forces will be the cost (much cheaper as you scale up) and less energy used for much more horsepower. Everyone is racing with this goal in mind already.

Mitsubishi wants to use their fastest lasers on Poet's interposer because they've seen something in the performance metrics or minimal power usage (I think it's both as Suresh alluded to the power consumption in the pr) that blows them away. Hence the momentous technical achievement comment from the GM.

My real point in all this is I think we have something that's world class and leading edge but more importantly, the big companies (Microsoft, Google, Apple, Amazon, Facebook) will not let Mitsubishis's customers run too far ahead with these solutions as it's tips the AI compute advantages too much in their favour.