Actually, getting to strong AI through math (rather than through mimicking humans) sounds more probable to me. We already have formalisms that can compute the most accurate possible prediction and the most efficient possible way to optimize a a utility function (inferring the right physical laws in the process) if given tons of computing power, for example look up Solomonoff induction or Marcus Hutter's AIXI. These count as superintelligences, or at least superweapons that can destroy the world. Stross's argument does not demonstrate the unlikelihood of someone implementing a fast approximation to AIXI tomorrow.
While that'd be a nice thing to have, it doesn't seem anything like an AI in the usual sense... i.e. the sense in which AIs have personhood and their own thoughts and feelings and goals and so forth.