Congratulations 🎉 to the winners of the 2026 Optica awards and medals! Find the full list of recipients in the OPN June issue: bit.ly/4urGoGn #honors #awards #optics #photonics

Micron crossed $1 trillion in market cap and it is still undervalued (Save this). Every AI chip ever built needs high bandwidth memory stacked directly on the die, physically bonded to the processor itself at the package level, because modern AI compute has gotten so fast that the only way to feed it data is to fuse the memory directly to the chip. You cannot run a GB300, Vera Rubin, Google TPU or an Amazon Trainium without HBM sitting on top of it. And there are only three companies on Earth that can make it at scale, Samsung, SK Hynix, and Micron. Right now, every unit of HBM Micron will produce in 2026 is already contracted. HBM4 for 2026 is sold out too. Micron's own management stated publicly that the memory crunch will outlast 2026 and push deep into 2027, with demand continuing to outpace every capacity addition they can bring online. Elon Musk laid out the exact structural reason for this in a recent interview, and his framing is worth looking at. He said the limiting factor in AI is not the model, not the software, not even the GPU, it is the ability to manufacture chips, specifically logic, memory, and packaging, at the volumes the AI buildout demands. He then noted something that most people do not know: there is not a single high-volume computer memory fab operating in America right now. The first Micron fab in Boise, Idaho, part of a $50 billion expansion will not reach volume production until 2027 while the second Boise fab comes online in 2028. And the New York megafab is 2029 and 2030. And Elon's conclusion is that even if you take every memory maker's most optimistic production assumptions and stack them together, it still does not come close to meeting the demand that AI infrastructure is projected to generate across that same window. That is why he said Micron's market cap going to $1.2 trillion or higher is the just the start. William Blair estimates Micron's HBM revenue grows 164% in 2026 and another 40% in 2027, as AI servers continue consuming more memory per rack with every new generation of hardware. Despite that trajectory, the stock currently trades at just 7.2x projected 2027 earnings. A company growing at triple digits, with its entire production sold out for two consecutive years, sitting on the only American controlled fab pipeline capable of addressing the supply gap over the next decade and it is trading at 7x forward earnings. That is the undervaluation, we are buying more Micron here. Come join Milk Road Pro and get our full breakdown on where Micron sits in the AI infrastructure stack, including how we are thinking about HBM pricing through the supply crunch and all of our AI thesis. Link below.

For AI data centers, the key question right now is time-to-power. Cost and emissions matter, but the first question is simpler: when can power actually reach the site where the GPUs are being installed? Grid interconnection can take years. Large gas turbines now come with long lead times as well. That gap is where fuel cells have become one of the more realistic alternatives. But the important point is that not all fuel cells are the same bet. SOFC, PEM, and MCFC use different fuels, serve different roles, and carry very different economics. So this market is less about a broad “fuel cell theme” and more about who can bring megawatts to the site first. This one is free, so feel free to read and share it around.

What caught my attention is the 3D-printed copper heat sink structure. The intricate lattice geometry would be extremely difficult—if not impossible—to manufacture using conventional machining processes. With metal additive manufacturing, engineers can create highly complex internal structures that significantly increase surface area while maintaining mechanical strength and reducing weight. More surface area means more efficient heat exchange and better cooling performance. As AI chips continue pushing toward kilowatt-level power densities, thermal management is becoming a critical differentiator. The future of cooling may not just be about better liquids or larger pumps—it may also depend on entirely new heat sink architectures enabled by advanced manufacturing technologies. The AI race is no longer just about faster chips. It’s increasingly becoming a race to remove heat more efficiently.

This Nvidia GPU farm sits in a spare room and prints $18,000 a month Ten cards running 24 hours a day, seven days a week. The setup cost $120,000 to build but it paid for itself in seven months. It does not mine crypto. It rents compute to AI companies that need processing power right now. Companies pay by the hour and the demand never stops. At full capacity the farm pulls $18,000 a month after electricity costs. The owner does not touch it. It just runs. Nvidia GPUs are the most in-demand piece of hardware on the planet right now. The companies that figured this out two years ago are already sitting on serious passive income. The barrier to entry is high but the people inside are not leaving. Follow if you want to understand where the real AI money is actually going.