It is correct that there are ways to reduce the keyspace you need to search - and that future research may reduce this a little more. And it may turn out that in the future, there will be a breakthrough that reduces the search keyspace significantly - but there is no indication of that at the moment.
So even with 128-bit AES, the cheapest and most reliable way to break the key is to use one of the two traditional methods - the three B's technique (bribery, burglary, blackmail) or rubber hose cryptoanalysis. And it looks likely to remain that way for a long time yet.
The energy argument is a good point. There are theoretical limits to information storage density, and to the minimum amount of energy for calculations.
As far as I know, the theoretical minimum energy for switching one line is kT, where k is the Boltzmann constant and T is the temperature (in K). That's 4e-21 J at room temperature. If we assume that testing an n-bit key takes 1000n switches (an absurdly low estimate), then it takes 5e-16 J per test, and thus 1.75e23 J total to do a brute-force crack of a 128-bit key.
The earth's current energy consumption is about 150 PWh per year, or 5.4e17 J per year.
That means it would take 300000 years to power the calculation to break the 128-bit key, assuming the same power generation of the earth, assuming absolute theoretical minimal switching energies, and assuming ridiculously small numbers of switches per test.
Call me naive, but I don't think the NSA has a secret AES-128 cracking lab...
According to current research on cryptoanalisys strenght of AES-256 is comparable with bruteforce key size about 220-bits. So there is no sense to use pure bruteforce. Of course 2^220 is still very big number, but future technics may still lower number. Beside of that, there are additional tricks that could be used against particular AES implementation (in hardware or software), that may lead to recovering the value of the key. This is the most current problem.
People who talk about this stuff often ignore 3 factors that make any code more secure. I was reminded of this by the speed comment.
- Symmetric encryption is usually used to securely open a conversation. The initial part of the conversation trades information that is then used to encrypt the rest of the conversation with a much simpler asymmetric encryption.
- Because it is primarily a conversation opener, you can say that you have to have 100 msec between attempts. At this point the power of your computer doesn't matter. You can only try 10 times per second no matter how fast your computer is.
- The response to the last one is that they don't have to retry. They can try repeatedly to decrypt the same recorded message. However, that only works if they have a way to assess that they have successfully decrypted it. Information exchanges as given above can be in the form of a chain of completely random numbers with no encrypted checksum. That makes it impossible to know if you have cracked it without cracking the whole bloody conversation. What I am saying is that the messages can be designed in such a way that the code breaker has no way of knowing when he got it right. It does him no good to guess the correct key if he doesn't know he got it.
Thanks for the great article on data security based on brute force attacks. I am looking at quantum algorithms and the use of quantum computers (D-Wave from Canada). My sense is that quantum is the biggest threat to data security in the medium term. In the short term attacks based on side channel leakage and active attacks are the main problem.
Sorry to have forgotten to add this the way the system would work is to start with DES being used 4 times on the four input blocks then those four input blocks are fed into AES256 and finally into 4 OUTPUT blocks via the DES to complete one cycle of encryption also the system can be used in reverse.