After thinking about it for a bit, I wonder where an 1170 Teensy would/could exist in the market. The performance of even the T4 seems to be bounded more by it's IO than anything else. The 1170 even more so. Not having an integrated radio like the Espressif offerings make it difficult to compete there. General CV/NN/ML performance will always be outperformed by SB-GPUs such as the Jetson Nano, or ASICs like the Kendryte K210 or the Google Edge TPU. Linux SBCs are like sand at the beach. So, what else is left? What is the right kind of workload for a 1Ghz MCU with a co-processor?
Instead of trying to answer that question, I think a 1170 Teensy would benefit from just providing different flavors of high bandwidth IO. I'm sure there are a number of applications requiring a much higher IO bandwidth than what is practically possible with an existing MCU. High speed Ethernet interconnects are currently the realm of Linux SBCs, but I think there are a number of applications that are power sensitive, latency sensitive, or would simply benefit from more robust, deterministic behavior.
As a specific example (and pet project), I've always found it silly that a guitar, with it's multiple strings, typically has an analog, monophonic output, highly susceptible to EMI, whose sound is permanently "colored" by a combination of pickups, wiring, and amplifiers. What I also find silly is if you do indeed have
multiple pickups, is sending the analog signals individually over
a thick cable to a breakout box. This is where a robust, low latency, multi-channel audio CODEC, Ethernet with POE, and an AES67 implementation would be the DIY alternative to Audinate's expensive Dante stuff. A Raspberry Pi just wouldn't cut it here.
So, to summarize, if planing a Teensy based on the 1170, prioritize breaking out the high bandwidth interfaces such as CSI/DSI, I2S, and Ethernet, perhaps even at the expense of a couple of the extra lower speed interfaces (UART, I2C, CAN, etc.).