The PS5 has been in development since 2013, and if Horizon started development in 2017, I think its fair to assume that Guerilla had more insight and quite possibly input into how the PS5 hardware would eventually shape up. They had more knowledge about what the system could potentially do as time went on from the different design phases, and until the specs were finalized, and thus, shaped their engine accordingly.
These processors, GPU and SSD didn't exist in 2017.
In respect to Unreal Engine 5, not all current gen games will be using it. As amazing as the engine is, some developers decide to develop their own engines, or even use older Unreal Engines for various reasons, so I don't think that this an accurate conclusion of when big AAA exclusive games will start dropping. I mean Returnal and Ratchet and Clank: Rift Apart are good examples of this.
The UE5 is just an example. It's the first engine that shown a demo of taking advantage of next gen stuff but they are still developing it. So until later this year devs won't be able to start making UE games taking full advantage of its new stuff, and if we add some years of development, then we won't see AAA games taking advantage of that until a few years in the future.
This is with UE5, the most popular engine. So the timing will be the same, or even longer for most other game engines. Returnal and Ratchet could have been done for PS4 with no problem, with the only difference of longer loading times, no RT if they use them and no DualSense features (plus less native resolution or framerate etc). They could have been crossgen if desired, but for marketing terms they wanted to have some next gen only games.
Regarding Horizon 3, I don't expect we will have to wait until 2026 for it at all. The franchise has only grown in popularity, and the studio has grown respectively. I expect Horizon 3 at least early 2025. We'll see how that plays out.
They have grown but they have been working on at least 3 games: Horizon 2, Horizon VR and their multiplayer game. And even if the team for the mainline Horizon games grow, every generation the work required for AAA games, so it also grows the amount of people or time needed to make them because specially games get more detailed, more dense and more varied in terms of content which basically means a lot of more work specially on art. So even that team grew, they still would need at least the same time they needed to develop Horizon 2, around 5 years. Being optimistic, 4.
We're in 2022, so approx. around 2026 being optimistic. And this is very optimistic considering that it would be in normal conditions, only with the same but only throwing more polygons, some better textures, maybe a few characters more on screen, higher native resolution and stuff like that.
But isn't the case, the potential of what this next gen offers with that SSD+ I/O system combined with the horsepower they have means a big paradigm change in terms of game developent processes, tools and engines. Changing how their game designers, programmers and specially artists worked since they started to make 3D games. With also a big impact on level design and how they calculate the budgets of what they can put in the game "at the same time" in memory and many things more. They'll need time to prototype, test and implement these things plus adapt how they work to the new methods and then to design, test, prototype and iterate the new stuff they will be capable of doing now.
They basically will be able to make stuff almost photorealistic, better than the stuff shown in the Matrix demo. With worlds that could be more dense and detailed and with a level design not constrained by the the low streaming speeds that HDD had or by the fake baked lighting and shadowing they had. Example: with current/past lighting it was problematic to make some environments destructible because if changed the tricks made to fake the lighting would be visible. And level design was made in a way to hide pop-in and mask/hide level streaming and also to hide enemies/npcs/animals being spawned only a few meters away from you.
There are many possibilities open allowing insane detail, density, variety with more open, less linear, more realistic and dynamic worlds than they are right now. Stuff that implemented obviously will require way more work. Which would mean way more dev time, way bigger dev teams and pretty likely to stop making the games bigger and bigger if they want to do all that on a reasonable budget. Plus finding new revenue sources to pay that extra budget (so guess why the big publishers, even the 1st parties are more and more interested on GaaS and multiplatform).