From DES to PQC: A Timeline of Cryptographic Migration and Its Bills

Archi4 min read

cryptographyhistorysecurity

Contents
  1. 1977: DES ships, and the government picks up the tab
  2. The cost nobody counted
  3. 1993: Clipper and the $500 million that never shipped
  4. 1997 to 2001: AES, a competition with a budget
  5. What AES taught buyers
  6. 2004 to 2012: MD5 and SHA-1 collisions turn academic
  7. The embedded lag
  8. 2008 to 2015: The Heartbleed era and the maintenance bill
  9. 2016 to 2020: TLS 1.3 and the cost of doing nothing
  10. 2022 to 2024: NIST picks post-quantum algorithms
  11. The inventory problem
  12. 2024 onward: Pilots, deadlines, and who writes the check

1977: DES ships, and the government picks up the tab#

The Data Encryption Standard became a federal standard in 1977. The National Bureau of Standards ran the competition, and IBM, which submitted the Lucifer cipher that became DES, absorbed most of the research cost. The public debate that followed centered on key length. A 56-bit key was small even then, but the standard was free to implement, and that price decided the market. Banks, terminals, and government systems built on it because nobody had to pay a license fee.

The cost nobody counted#

Cheap standards create long tails. By the late 1990s, DES protected ATM networks, point-of-sale terminals, and federal payroll systems. The eventual replacement bill would land on whoever owned those systems, and most of those owners had no line item for it.

1993: Clipper and the $500 million that never shipped#

The Clipper chip proposed government key escrow for voice encryption. The Clinton administration pushed it as a voluntary standard, and AT&T and others built prototypes. Industry resistance was loud, and the program died without a mass deployment. The visible spending was modest, but the episode set a precedent: cryptographic standards that require hardware replacement are expensive to abandon, and vendors remember the write-off.

1997 to 2001: AES, a competition with a budget#

NIST ran an open competition to replace DES. Fifteen candidate algorithms went through public cryptanalysis, and the winner was Rijndael, standardized as AES in 2001. The competition cost a fraction of what a procurement program would have, and the payoff was a royalty-free algorithm that ran well in software and hardware. Chipmakers added AES instructions to processors within a decade. The migration itself was the expensive part: every smart card, VPN appliance, and disk encryption product needed new silicon or new firmware.

What AES taught buyers#

Free publication of the algorithm compressed implementation costs. The rework costs stayed, and they fell on customers who had to replace hardware on a schedule they did not control.

2004 to 2012: MD5 and SHA-1 collisions turn academic#

Researchers showed practical collisions in MD5 in 2004. SHA-1 collisions became feasible enough that browser vendors announced deprecation of SHA-1 certificates, and the shift to SHA-256 took years. Certificate authorities, server operators, and embedded device makers all had to reissue or rebuild. The bill was spread across millions of small operators rather than concentrated in one budget, which is why it took until roughly 2017 for SHA-1 to disappear from public certificate chains.

The embedded lag#

Firmware in industrial controllers and medical devices outlived the deprecation deadline. Replacing a certificate on a device with no update path means a truck roll, sometimes a full unit swap. Those costs rarely appear in the security press release.

2008 to 2015: The Heartbleed era and the maintenance bill#

OpenSSL's Heartbleed bug in 2014 forced a global certificate and key rotation. The direct fix was small. The operational cost was not: password resets, incident response, and reissued certificates across a large share of the web. The episode pushed companies to fund open source cryptography more seriously, and the Linux Foundation's Core Infrastructure Initiative followed. That funding was in the low millions per year, a rounding error against the remediation spend it was meant to prevent.

2016 to 2020: TLS 1.3 and the cost of doing nothing#

TLS 1.3 removed legacy ciphers and simplified handshakes. Cloud providers and CDNs absorbed most of the server-side work, which kept the transition cheap for small sites. The remaining cost sat with old clients and middleboxes that broke on the new handshake. Enterprises discovered they were paying for appliances that could not be upgraded, and some bought replacements rather than wait for firmware.

2022 to 2024: NIST picks post-quantum algorithms#

NIST selected CRYSTALS-Kyber for key encapsulation and Dilithium and related schemes for signatures, with final standards published in 2024. The algorithms are free. The migration is not. Post-quantum keys and signatures are larger, which means more bandwidth, more storage, and in some cases new hardware security modules. Hybrid deployments let organizations run classical and post-quantum together, at roughly double the handshake size.

The inventory problem#

Cryptography is embedded in places that do not announce themselves: firmware signing keys, code signing pipelines, VPN concentrators, and smart meters. Nobody has a reliable count of them. Consulting firms now sell crypto inventories, and the first deliverable is usually a spreadsheet that shows the problem is bigger than the buyer expected.

2024 onward: Pilots, deadlines, and who writes the check#

Government agencies have set migration targets for the early 2030s. Browser and cloud vendors are testing post-quantum key exchange on live traffic. The spending pattern is familiar from earlier transitions: a small number of large vendors fund the standards work, the biggest customers pay for the pilots, and everyone else discovers the deadline late and pays a premium for urgent replacement.

The money question is not whether post-quantum cryptography is worth adopting. It is who budgets for the long tail of devices that will still be running old primitives when the deadline arrives. History suggests the answer is the operators who deferred the work, and that the bill arrives with interest.

More articles