In addition to the costs of developing good foundational code, I think revenue is becoming a driver for slower-than-necessary software. An increasing number of

In addition to the costs of developing good foundational code, I think revenue is becoming a driver for slower-than-necessary software. An increasing number of providers (for example databases) charge per server, per core, or another hardware-usage metric. The more hardware, the more revenue for them as they make about 90% margin for every new machine. There is a high incentive to get users to need more machines on their "managed cloud". Vendors could try to improve their software so it requires twice as little CPU. But why bother since this would result in 2x cost in revenue? It makes more sense to focus on horizontal scalability than on core efficiency so users can keep on adding machines to their cluster over time. If software is running at 1% of the maximum performance as suggested in the article, a 1% improvement could reduce costs by 50%. But I think none of the existing vendors will ever make the move as it conflicts against their own interest.