I think Go + sum types could be good. Maybe. But, honestly, it's hard to say for sure. First-order effects are great: we have sum types and can use them to model problems. Second-order effects get muddy: We have sum types and they are the ideal solution for a bunch of problems, but without other features they're awkward to use. For example... now you can do a Result type... but if you want to return multiple values in a Result, you need a set/tuple type too. If you do that, is Go's multiple return values concept still a good idea? I could probably go on but hopefully this illustrates the point.
I think a lot of people don't acknowledge why Go is truly so successful. Well OK, first elephant in the room, it's obviously successful because it's backed by Google, a company who can and did throw immense resources at making the implementation and standard library pretty damn good as well as helping to push it, but that alone wouldn't have propelled it to where it is today (I mean, compare it to Dart.)
But beyond that, Go is successful because Go code is very simple to write and leaves the programmer with relatively few things to be concerned about, and yet in spite of that, Go code generally runs reasonably fast, uses relatively small amounts of memory and is generally not very crash-prone. (Some people would happily debate that, but I trust software like restic, esbuild, rclone and Syncthing every day without fail, among other smaller Go programs. I'm OK with making that assertion.)
If you put in the effort to make really good Rust code, the effort is not wasted, but it is a lot of effort when the stupid Go code often does the trick. Consider Discord's presence service: they switched from Go to Rust and massively cut costs and improved latency. Rust wins, Rust better? But... most services will never scale to the point where marginal improvements in latency, throughput or RAM are going to be worth a lot of man-hours worth of programming and ongoing maintenance. Usually throwing a couple more VMs at the problem is just going to be the path of lesser resistance. This was always the case when comparing lower-level to higher-level, but Go amplifies it because the performance gap isn't as big, but the complexity gap remains very large or maybe gets larger.
Is writing Rust code really that hard? No, not at all, it's more that writing good Go code is so relatively easy, the language is simple and the standard library is loaded with useful tools. Go's CLI flag parser is pretty unimpressive, but so many projects just need a very basic flag parser and it works totally fine for that, you just don't need to overthink it 99.99% of the time. Same for net/http, the built-in image and compression codecs, TLS stack, and more. Meanwhile, designing and maintaining good high-quality Rust crates is just relatively hard. You've got to worry about async vs sync, various cfg options like nostd and noalloc, and dealing with a lot of wrinkles in the language. Want to use a dyn trait? You'll need to make sure the trait doesn't have any functions with generic parameters or async; makes perfect sense, but adds tension; you want to avoid unnecessary runtime costs in the ideal case but still have the flexibility to use a dyn trait in other cases. Not to mention the borrow checker and how it interacts with a lot of these design choices. Go code is much dumber, but often times it's sufficient.
And that's the crux of it. Go is simple in a stupid way, rather than an elegant way, but it really makes the most of it. If you want to supplant Go at what it does best, trying to make a better programming language overall may be a bit misguided. It's probably not that hard to come up with a programming language design that is "better" than Go. What's hard, IMO, is coming up with a programming language where the cognitive load increase relative to Go is met with a pay-off that people using Go would consider genuinely worth it. Something like guaranteed data race safety is definitely good enough if it's something someone needs or at least strongly wants for a given use case. Sum types, on the other hand, are very nice thing to have that make modelling data easier and less error-prone, but not having them isn't exactly the end of the world... In Go, people sometimes emulate them with interfaces and type-switches, and it's not fantastic, but it's usually sufficient.
Ocaml probably could/should be more successful, but I'm not sure it competes with Go, I think they're in entirely different wheelhouses. Ocaml feels like it doesn't want to compete with Go, but then again, I only have relatively surface-level knowledge of Ocaml, so I can't really say for sure.