Let's lay out the problematic letters A, M, and T first and then make sure they get separated.
Without restriction, there are ${6\choose2,2,2}={6!\over2!2!2!}=90$ ways to arrange $2$ A's, $2$ M's, and $2$ T's. Among these $6$ arrangements have all three letters doubled up (e.g., AAMMTT), $18$ have two letters doubled up (e.g., TAATMM), $36$ have one letter doubled up (e.g., TMAAMT), and $30$ have no letters doubled up (e.g., AMTMAT). Each of these counts requires a bit of thought, but I'll take them as given.
We now need to wedge in the remaining five letters, C, E, H, I, and S, in a way that busts up any doubled-up letters. Let's do one example carefully and then summarize the rest.
Suppose we have MMTTAA. Our first three wedge-in will bust up the pairs by attaching something immediately to the right of the first occurrence of each letter, i.e., (Mx)M(Ty)T(Az)A. This can be done in $5\times4\times3$ ways. The remaining two letters must then be wedged in among these $6$ groupings, and this can be done in $7\times8$ ways, for a total of $(5\times4\times3)\times(7\times8)$ arrangements.
If there are two doubled letters, the same approach gives of total of $(5\times4)\times(7\times8\times9)$ arrangements, and if there is only one doubled letter, it's $5\times(7\times8\times9\times10)$. Finally, if there are no doubled letters, the remaining $5$ letters can be wedged in in ($7\times8\times9\times10\times11)$ ways.
To wrap things up, we just have to multiply these counts by their occurrences and add everything up:
$$6\times(5\times4\times3)\times(7\times8)+18\times(5\times4)\times(7\times8\times9)+36\times5\times(7\times8\times9\times10)\\+30\times(7\times8\times9\times10\times11)$$
If I've done all the arithmetic correctly, the final answer is $2{,}772{,}000$.
Remark: N. F. Taussig's inclusion-exclusion approach leads to the same final answer, so I am somewhat more confident in my arithmetic. It's a bit curious that $2{,}772{,}000=11\times10\times5\times7!$. In particular, the answer is divisible by $11$, the total number of letters, even though neither approach makes it obvious that that should be the case. So I wonder if that's purely coincidental, or if there is some approach to the count that does make it obvious that the answer is divisible by the number of letters involved.