Just as there are already “enough” physical cars, there is also already enough physical restaurant infrastructure, so today it’s more a question of efficient utilization. What Uber Eats is doing in this space is really smart: Keeping track of local food search keywords, and reaching out to local restaurants saying “Hey, if you provided such-and-such food items at such-and-such time of day in your area, we think you’d do really well. You can sell these items on Uber Eats under a different brand if they don’t fit in your existing cuisine niche.”

Longer-term you’re probably right that centralization could be more efficient in some cases, but looking too far into the future often makes less economic sense today. See: Webvan.