The problem with being second to the party, is that the end use clients have already invested heavily in customization, vetting (testing), negotiated prices, staffed up and trained to adopt the new product, let's say Celestil AI's photonic fabric. They may be pleased and have issued POs for lots of future product.

Along comes something cool and new. Is it twice as good? Will CAI be 5X in it's next gen? It is really hard to come in second. 

The reality is that there are lots and lots of smaller data centers. There will be many, many AI agents running in these data centers, even on desktops. I know of one that points his AI product at only the data on his desktop. This will become common. My point is that there will be lots of entry points for stitching together xPUs in various types of photonic packages. These may compete with CAI or Lighmatter too. There will be a need at various price points, form factors, and device friendlyness. POET can dive in ha many of these battles, by merely supplying light engines and OI's and modules.

It is hard to improve upon the small size of the OI, and the experience POET is getting by working with the industry giants. Foxconn and Mitsubishi serve a diverse client base that may need POET's chips in surprising applications that we have not even imagined. I clipped from a video for fun, just to visualize an OI chip. Remember, it is half the size of your pinky fingernail.

https://youtube.com/clip/UgkxlGphWyY88nJDtg8n3Mi2ERD7y10lvqbA?si=FRH2vXXDm8n276rk

Don't pay attention to the stock investor guy in the background :)