The article claims, if I understood it correctly, that GCP will not be profitable unless it reaches the scale of being at least #2. The implication is that econ

The article claims, if I understood it correctly, that GCP will not be profitable unless it reaches the scale of being at least #2. The implication is that economies of scale are necessary for profitability. For example, in 2014, James Hamilton spoke about how every day that year, AWS was adding enough new server capacity to support all of Amazon's global infrastructure when it was a $7B annual revenue enterprise (2004). That scale of build-out, server purchasing, etc. presumably justifies and enables significant cost reductions, such as buying equipment from vendors in bulk with steep discounts, or buying or building large facilities at once, etc. AWS is custom designing many parts of its stack including networking chips and software, ARM-based CPUs, and more; that kind of investment is only worthwhile at a certain large scale. This scale presumably motivated AWS's purchase of chipmaker Annapurna Labs in 2015. Some operator of a few colocation facilities, by comparison, could not justify designing custom silicon since their scale would not allow them to recoup the R&D costs. Before AWS reached whatever scale made these kinds of investments worthwhile, they were purchasing commodity hardware. To pick another example, a company like SpaceX can only be profitable if it launches a certain minimum number of rockets per year. Without enough rocket launches, and demand from customers, you can't recoup the R&D budget needed to design them. Google has a head start in cloud because their company's existing facilities are already huge. However, public infrastructure has different demands than internal infrastructure, and it sounds like the necessary investments in public have been costly.