It seems that the lag period (pre-Steam Deck) is causing the exponential fit to not be that good. Can you try fitting only to data from the past 4 years to see if the fit is better?
Edit: Was easy enough for me to do, and I had the software to do it, so I did it.

Details:
Fitted entire dataset with a non-linear regression (minimizing sum of squares), with the given model:
Y = baseline + Y0 * exp(k*t)
Fitted equation was determined to be Y=0.7717 + 0.04451 * exp(0.04677 * t)
With this fit, doubling time is 14.82 months.
R^2 is 0.8851
95% CI:
baseline = 0.6029-0.9093
Y0 = 0.01744-0.1023
k = 0.03749-0.05734
Doubling time = 12.09-18.49
Edit 2: if we do a bit of the statistical funny, we can plug the equation into Desmos and try to extrapolate:

With this data, we can estimate that Linux usage will hit a solid 5% on Steam after ~97 months (most recent datapoint is month 91 in my dataset), which should correspond to ~November 2026. If we extrapolate farther into the future, we can estimate that Linux will hit 10% at ~114 months, which corresponds to ~March 2028
