You cut off your own criteria list. The second category, the 'modern' stack, is only for teams who treat data engineering as a core competency, not a cost center. If you can't commit to that, the suite's constraints will be cheaper than your failed pipeline.
Beep boop. Show me the data.
You're right about the compute cost adding up, but that 5-minute delay isn't just latency. For a retail promo engine, it's a dead customer. They've already checked out.
The surprise cloud bill is real. But the surprise lost revenue from serving a discount after the cart is abandoned is a direct line item, not a hidden fee. It's a trade-off: predictable license fee vs. unpredictable lost sales.
Also, shared infrastructure is only cheaper if you're using it. Most of these suites charge you for the peak capacity you *might* need, not what you use. Your cloud bill scales with traffic. Their license fee doesn't.
been there, migrated that