As someone who makes a living writing scientific software it is not that the market is small, but that scientists find it very hard to pay for software. They are happy to spend hundreds of thousands of dollars on some piece of hardware, but in most scientist's mind software should be free.
Disclaimer: I'm in the scientific equipment and software business. There are some issues to overcome. These may just be excuses from people who don't want to change their game, but nonetheless:
1. Sometimes, wages and equipment / software come from different pots of money. No matter how much sense it makes to replace labor with automation in the grand scheme of things, if you can't move money from one pot to the other, then you're stuck with the status quo.
2. I think that people sometimes underestimate the degree of customization and effort required to automate a specific process. Or sometimes overestimate, as in, "our process is so special that no commercial tools will work for it."
3. If anybody is going to work on automation, it will be the students themselves, as it's a way to learn a valuable skill. They may be doing it with minimal fanfare, and there is a strong movement towards open source tools. Students realize that the generous budgets and site licenses will vanish when they leave the academe, and are interested in preparing themselves for freelance work, startups, etc. This may also favor general purpose tools, rather than those designed specifically for science.
My anecdote, from 25+ years ago: I taught myself electronics and programming by automating my student projects, culminating in my thesis experiment.
I was really lucky that I took to programming pretty easily while still in high school. (graduated '82). Likewise with math. As a result, I was able to integrating computing into my work with minimal guidance.
But it's my view that the teaching of math and science should involve computation, starting as early as possible. It still amazes me that a kid can go through high school without learning about something that has had so much impact on our society. In my utopian world, there would be at least one question in each physics homework assignment with the instructions: Solve this with computation. And it wouldn't be a big huge deal.
The people capable of the automation have financial incentives to move to industries with broader appeal. There are a lot of people doing great scientific computing work in academia (e.g. Sean Eddy, Titus Brown), but they are notable exceptions.
This of course is stereotyping, but for me it feels like there is truth to it:
It has more prestige to have a bunch of Phd's doing busywork in the lab than to pay for software which could replace them. Expensive equipment at least looks cool and impressive if the bosses go round with guests and want to show off where the money went, software can at the very best make pretty pictures on a screen somewhere.
Academic research is not necessarily oriented on profit on that level: It cares about getting grant money, but not always about using that money as efficiently as possible.
And report graphs. Lots of report graphs. Oh, look, a real-time dashboard. Do the metrics its tracking mean anything to anybody that is reading it? Who cares?
This seems to be one of the main ways salesmen and execs communicate over enterprise software deals.
Amusing aside: Whenever I see anybody mention a "dashboard" in business, I'm reminded of the kiddie toy for long car trips, that has a brightly colored steering wheel, several interesting levers and knobs, a mirror, etc. The idea is that the tykes can pretend that they're driving.
Not always. There has been commercial molecular modeling code since the 70-80's, but the number of players has noticeably shrunk over the years. The whole market might be 50-100mil/year and as pg said there's no exit strategy for something that small. It's hindered by a perceived lack of impact and various scientific problems yet to be solved. There's also a disconnect between data held proprietary by the customers vs. commercial and academic developers and researchers.
And bioinformatics is (or at least was) different in that there were good, open-source alternatives which made it financially impossible to to sustain a commercial effort. There are players but I don't think they've done better than molecular modeling. Hard to compete with free and good.
The business needs some real scientific breakthroughs to jump forward. Currently, everything can be coded well but the tools are too blunt.
It is far more profitable to just write mainstream software.