15 years from now:

The desktop computer will be still there, same thing as before. Console market share will decrease, but they'll be still there. Smartphone games will become increasingly more complex, and some games will be made with phone joysticks in mind.

A bunch of new input methods will be released; they'll be seen as gimmicks, most will fade away. One or two will stick for longer.

There'll be a bigger crash than in 1983, and a lot of AAA studios will fill bankruptcy. It'll be like a forest fire - as it burns down the old and big trees, it leaves space for smaller plants to thrive.

You'll at least one game featuring a multiplayer version of any given current single-player genre. Multiplayer Vampire bullet heaven, multiplayer colony simulator, multiplayer dating sim...

Speaking on bullet heaven (I mean things like Vampire Survivors), you'll see elements of the genre creeping into other genres. Much like RPGs did in the past.

Nethack version 3.8.4 will be released. It'll have changes like

• fixed exploit where players would #name a corpse Vladsbane and use it to kill Vlad, then use the output message to know if the corpse was fresh or tainted.
• fixed critical bug: if you #wish for a statue of the fourth rider and cast stone to flesh on it, hitting it with a rubber chicken used to end your run. Now it correctly reverts the rider back into a statue.

LLM presence in games will be subtle. For example, irrelevant NPCs might say things generated by a LLM, instead of pre-scripted ones. This won't become the standard because they tend to output hallucinations that confuse players (e.g. refer to some item that doesn't exist), but some games will make good use of it.

"Evolving" enemies on the other hand will be the next big hit. They won't use any current AI model, but they'll run local (and way simpler) neural networks. For example, if you kill lots of enemies by luring them into a pit, you might notice newer enemies avoiding to hang around pits.