That and buzzword compliance. Why are MongoDB, Scala, and machine-learning on the list? What customer-visible advantage do they give your startup that your existing Rails/Django app on MySQL/Postgres doesn't have?
If you want to use machine-learning effectively anyway, you need to have a bunch of other technologies, like crawling & unstructured data extraction, or custom log processing & millions of users, or some other way to get the raw data that your machine-learning algorithms can process. Without data, machine-learning is useless. But that's a problem for an established company, not a startup, and established companies have long wanted specialists anyway.
If I were to shoot for the most charitable reading of the article, I would abstract a way from particular brands and replace Amazon with "cloud hosting(AWS, azure, GC, ...), mongo with "scale out database", scala with "modern language", and machine learning with "data science".
With such replacements the article looks fairly reasonable - larger scale demands newer technologies, proliferation of client platforms demands more suppor for them, and growth in computing power and science makes users expect some smarts they did not demand a decade ago.
Except that within the context of startups, it's still not reasonable. You don't need to scale when you're first founding your startup: your biggest challenge is building a product that one person finds useful, let alone one million. You can run such a service off your favorite shared webhost, or a single EC2 micro instance for like $10/month, or just use the free Heroku plan.
If and when your service catches on, then all the article's points become valid. But then you'll have resources available to hire specialists - and the full stack developers you hired in the early phase become full-stack integrators managing teams of specialists, assuming they have the ability & inclination to grow with the company. You're going to have to rewrite your code and throw out your early technology choices anyway: you might as well take advantage of the full stack developers you can find to get to the point where you can afford the specialists.
If you want to use machine-learning effectively anyway, you need to have a bunch of other technologies, like crawling & unstructured data extraction, or custom log processing & millions of users, or some other way to get the raw data that your machine-learning algorithms can process. Without data, machine-learning is useless. But that's a problem for an established company, not a startup, and established companies have long wanted specialists anyway.