Maligned #9 - The Platform Lock-In Trap

Short one this week.

Platform lock-in is the AI story nobody wants to tell

I keep having the same conversation with CTOs. They chose a cloud provider’s AI stack for speed, built their entire pipeline around provider-specific tooling, and now they’re stuck. Switching models means rewriting integrations. Switching providers means rebuilding infrastructure. This is the classic enterprise platform trap, and the AI industry is recreating it at speed. The companies that are doing this well are building abstraction layers from day one, keeping their core logic provider-agnostic even if they deploy on a specific platform. It’s more work upfront but it’s the only sane approach if you think you’ll be doing this for more than two years.

The AI startup graveyard is filling up

Multiple AI startups quietly shut down or pivoted this week. The pattern is consistent: raised a seed round on a compelling demo, couldn’t find product-market fit, burned through cash trying to compete on a feature that the model providers then shipped natively. The lesson keeps not being learned. Building a product that depends entirely on a capability gap that the platform owner can close at any time is not a business, it’s a bet. Some bets pay off, but most don’t.

Hardware innovation beyond NVIDIA

While NVIDIA dominates the AI chip conversation, there’s meaningful innovation happening elsewhere. AMD’s MI300X is gaining real traction for inference workloads. Intel’s Gaudi3 is finding a niche in cost-sensitive deployments. And a handful of startups, Groq, Cerebras, and others, are building specialized architectures that outperform GPUs for specific workload types. The GPU monoculture is unhealthy for the industry, and any diversification is welcome. Competition drives down prices, and AI compute costs need to come down a lot more before the industry’s unit economics actually work.

That’s it for this week.

Maligned - AI news by Mal