Microsoft (MSFT) is not broken. It’s just… exposed.
Let’s start with the good. The productivity suite is extraordinary. Excel, Office, Teams, they’re deeply embedded in global workflows. And yes, aligning with Claude and layering AI on top of Excel is genuinely powerful. That stack can scale.
But here’s the tension.
Copilot has been, at best, unevenly received. Enterprises are experimenting, but pricing power isn’t obvious. In a world where AI tools are rapidly commoditizing and becoming deflationary, it’s getting harder to charge premium add-ons for things that can be recreated cheaply or circumvented entirely.
The deeper issue is structural.
Microsoft is the backbone of software. But agentic AI increasingly sits on top of the backbone.
You can already:
- Overlay Windows with smarter automation
- Replace workflows with custom AI agents
- Build low-cost tools that replicate high-margin software features
That means Microsoft becomes infrastructure, but less differentiated infrastructure.
And hyperscaler dynamics have shifted slightly. The ability to build custom AI stacks cheaply reduces dependency on bundled enterprise software. Endless builders are figuring out how to pay less, do more, and avoid lock-in.
The operating system moat? Over time, it may matter less in an AI-layered world.
Microsoft isn’t going away. But in a deflationary AI regime, software incumbents face pricing compression risk. That’s not a collapse thesis. It’s a margin pressure thesis.
One to Love: Coherent Corp. (COHR)
Coherent Corp. (COHR) is the play. They’re a vertically integrated photonics company building the optical transceivers, lasers, and interconnects that AI data centers need to move data at scale. Their Data Center and Communications segment grew 26 percent year over year last quarter. They’re ramping 800G and 1.6T transceivers. They’re shipping optical circuit switches to hyperscalers. And they’re converting to six-inch indium phosphide wafers, which cuts costs and improves yields. When GPUs are choking on electrical I/O and power budgets are forcing adoption of co-packaged optics, Coherent is the company with the parts.
This isn’t hype. It’s physics.
We’re building a world of infinite compute. But compute without data movement is useless. You need to feed the beast, and you need to feed it fast. Copper is good. It’s gotten us this far. But good isn’t great, and great is what AI scale demands. That’s why we need transmission at the speed of light.
Bandwidth. Heat. Power. Signal loss.
Copper doesn’t scale at AI cluster speeds. Not rack-to-rack. Not campus-wide. Not for next-gen inference.
The shift is already happening:
- Electrical to optical
- Telecom to inside the data center
- Edge to massive scale-across architectures
There’s no credible non-photonics alternative at AI scale.
Wireless inside data centers? Physics problem. Copper evolution? Dead end at speed. Architectural tricks? They reduce traffic, they don’t eliminate it.
And co-packaged optics isn’t “next year” vaporware anymore. Power budgets are forcing adoption. GPUs are choking on electrical I/O. Hyperscalers are moving.
The winners will rotate, margins will compress over time, and timing will be lumpy. But directionally, the demand curve is locked in.
If Microsoft is exposed to deflationary AI economics…
Photonics benefits from the opposite force.
More AI = more data movement. More data movement = more optical density. More optical density = structural demand.
This is the picks-and-shovels layer of AI 2.0. The part you can drop on your toe.
Trade smart. Stay hedged.
Hans