Ooh, broken disks are AR$1200 (US$8) for a ten-pack. $120 (80¢) (or US$1 in the EU in 02013) gets you a voice-coil actuator built around two pyrophoric neodymium magnets plated in nickel mounted on permalloy or mu-metal brackets, a 5400rpm long-life bearing, and a BLDC motor with a controller, plus some extremely flat (Ra 120 pm) first-surface glass-ceramic mirrors with 80% reflectivity and a low thermal coefficient of expansion (7.4 ppm/° for TS-10) and high modulus (100 GPa), about 400g of castable aluminum between possible platters and the case (maybe A380 or ADC12 or 6061 or 5052) some Torx screws, a SATA power connector, some machined 6061-T6 aluminum spacers, and some jumper blocks. Oh and maybe an accelerometer and temperature and pressure sensors. As scrap metal this totals about US$1.66, mostly from copper and gold from the PCB, but obviously you can’t get voice-coil actuators or bearings that cheap.
Oh, and an ARM core accessible over JTAG, using external RAM and Flash. And it talks to the controller that controls the spindle and head over SPI. The Flash format has been largely reversed, but the chipmakers don’t publish datasheets.
The whole precisely balanced platter assembly with the platters, motor, and bearings is nowadays almost invariably 5400rpm or 7200rpm; 10krpm disks are exotic rarities, and 3600rpm disks are antique. A 7200-rpm 3.5" desktop drive has a rim speed of 5.3 m/s, while a 5400-rpm 2.5" laptop drive is 2.9 m/s. Either of these is a fairly respectable speed on its own for things like fine grinding, but also you can also overclock them quite a bit. They’re normally only operated at about 3 watts, so you probably can’t get more than 30 watts out of them no matter how much cooling you add.
The motor might be suitable as a hand pullstring generator.
Desktop disks still use aluminum platters, typically 635 μm thick. These are also 6061.
For small machinery the cobalt alloy used for the magnetic medium might be worth extracting and refining. I’m guessing it’s on the order of 10 mg per disk, assuming 25 nm thickness, 3 platters, and 3.5", and ignoring the center hole; there are also two layers on top of it and three underneath.
Reputedly the head arm bearing is also very high precision. Micah Elizabeth Scott built a 4kpps laser projector in 02008 using two voice-coil actuators from hard disks.
Not sure how to use the actual GMR sensors themselves; presumably they can detect small magnetic fields reliably.