Using Fixed Effects Versus First Difference When T = 2 Based on Need for HAC Standard Errors

Question

When T = 2, the fixed effects and first-difference models yield equivalent estimates. However, let's assume you have heteroskedasticity and serial correlation in your errors. With the fixed effects estimator you would typically use heteroskedasticity and autocorrelation consistent (HAC) standard errors. With the first difference estimator, there is no need to correct for serial correlation because you only have 1 observation per unit. Since HAC standard errors generally give more conservative (i.e. larger) standard errors, this seems like a reason to just use a first-difference model with heteroskedasticity robust standard errors. So when T = 2, is there any reason why the more conservative standard errors from a fixed effects model with HAC standard errors is preferable to first-difference model with heteroskedasticity-robust standard errors?