“De-commoditize protocols”: that was the line leaked from internal Microsoft memos in the late 90s. The core strategy behind Embrace, Extend, Extinguish.
I was working on developer platforms at Microsoft right in the thick of our $20M “Get the Facts” campaign. We were running ads trying to prove Linux was more expensive than Windows. The UK regulators actually made us pull our flagship ad because we had compared Linux running on a massive IBM mainframe to Windows on a cheap Intel box. It was wild. We fought like hell to defend the Windows monopoly, terrified that if the OS went open source, we’d lose everything.
Watching the open vs closed AI war play out today rhymes oh so beautifully.
Open weights are the new .docx.
I doubt many of you have heard of OpenOffice. Because it never beat Word on desktop share. And its mere existence forced Microsoft to open the file format to ISO standards. The lock-in moat died, but Word and Office survived and have continued to thrive.
That’s exactly what models like DeepSeek, Qwen & GLM are doing today. By offering ~90% of frontier performance at a fraction of the cost, they cap what any closed API can charge. There is no need for them to win the absolute frontier. They just need to exist to break monopoly pricing.
But here’s the real plot twist, and perhaps the ultimate lesson for today's AI labs: Microsoft didn't just survive. Microsoft won the next era by pulling off the greatest strategic pivot in tech history.
Microsoft stopped fighting the gravity of open source. Windows kept the desktop, but Linux took the cloud. Instead of dying on that hill, we built Azure, embraced Linux natively, and turned the open ecosystem into our biggest growth engine.
Today, the majority of Azure VMs run Linux.
Microsoft won by moving up-stack and selling the layer above the commodity. The AI model layer is commoditizing right on schedule. The real money is moving to compute, energy, data, and applications.
But, this is no eulogy for the closed labs.
For every Linux, there has to be a Microsoft.
Someone has to fund the bleeding-edge frontier, set the standard, and give the enterprise something to trust. Closed labs push the boundary forward. Open weights spread it and keep the pricing honest. The system needs both.