Cruise ship to scrap heap

How DEC's overnight collapse previews AI's creative destruction

Back in 1988 Digital Equipment Corporation (or DEC) was riding high. They were the second largest computer company in the world. Only IBM was bigger and DEC was catching up fast. The annual company conference, DEC World, was a lavish affair. Held in Cannes, DEC charted a cruise ship to accommodate all the attendees. DEC was doing well and flush with cash - back in the US they even operated a fleet of helicopters to take staff between sites.

But it wasn’t to last. By 1990 everything had changed; DEC was struggling to survive. Two years later the founder and CEO, Ken Olsen, was ousted. But removing Olsen only accelerated the decline, and, in 1998, the remains of DEC were sold to Compaq - a simple PC manufacturer.

What had gone wrong? And how could it have happened so quickly?

Shrinking transistors

The answer lay with the ever shrinking size of the transistor. Smaller transistors meant faster, cheaper hardware. Transistor densities increased forty fold between 1980 and 1990. Suddenly everyone could afford a computer with an easy to use graphical user interface (such as Microsoft Windows). DEC’s market was eaten by young upstarts such as Sun and SGI who offered better products for less money.

And it wasn’t just DEC who were struggling. IBM nearly went under during this period. Industry names - WordPerfect, Lotus 1-2-3 - disappeared. Many companies failed to navigate the rate of change.

It is said Ken Olsen didn’t believe transistors could shrink so rapidly. And that fed into his decision making. During the late 1980s DEC developed their next super-computer - the VAX 9000. It was physically massive - multiple racks full of complex logic - big transistors. It was powerful. But not powerful enough. Within a year, DEC had a single chip - smaller than a fingernail - which matched the VAX 9000. Olsen had poured millions into the wrong technology.

DEC could have survived. They drove the development of highly available computing, helped create ethernet, built the VMS operating system (which heavily influenced today’s Windows 11), built the world’s fastest CPU during the mid 1990s, pioneered one of the first MP3 players and created what was arguably the first widespread Internet search engines (AltaVista). Failed leadership killed DEC. Ken Olsen built DEC into a technical powerhouse. And then destroyed it.

As DEC imploded, transistors continued to shrink. By the late 90s they enabled the explosive growth of the internet. And a second wave of change was unleashed. Sun and SGI were casualties this time round. Microsoft was wounded. And a new round of companies appeared - Amazon, Facebook, Google.

Since then, the computing industry has been remarkably stable. There have been some innovations - phones, cloud, containers. But those have been at best point releases; most are little more than patch releases. We haven’t had a major version upgrade since the early 2000s.

Talk to many in the industry and they don’t remember the upheaval of the 1990s. It’s too long ago now. Back then DEC, Compaq, Wordperfect, Lotus, Sun - those were well known names. Today? It’s a different set of names.

But that looks set to change. New names are appearing. OpenAI, Anthropic, Character.AI. Some of the old guard will inevitably stumble. The FAANGs are no longer the young, nimble, energetic companies they once were. Bureaucracy has thickened their arteries; they have become set in their ways.

As DEC discovered change can happen very suddenly. A few days ago OpenAI launched ChatGPT Pulse - a push feed of information curated to your interests. OpenAI have an app installed on 10% of mobile handsets. An app with a text and voice interface. It’s not hard to see where this might go. Just when should Facebook and X start to worry?

And it’s not just companies that will be affected. Last week OpenAI shared an update on their GPDval benchmark, which compares AI performance against humans on a range of roles. AIs and humans complete tasks, and then a human 'expert' evaluates who completed the task best. Last spring GPT-4o had a 12% win rate. Today Opus gets 47%.

I’ve seen that all too well with my Codex experience. Earlier this year I used Claude to build a kloc counting tool in Rust. Not because I needed a kloc counting tool. But because it was a challenge for the models back then. And because it is easy to test. That first version was fiddly - there was a lot of back and forth. Claude 3.5 struggled to write Rust. I got there, but I only had time to implement support for a subset of languages.

Today?

I ran Codex over a large codebase and compared the output with Tokei (a well respected Rust kloc counter). The results differed, so I gave both sets of output to Codex. Codex noticed the missing language support immediately. Then it added support for all the missing languages (plus some others), found and fixed several other bugs and wrote fifty unit tests.

The final results from the tool are as close to Tokei as makes no difference. I hate to admit it but it would have taken me at least a week to do what Codex did in a few hours (and about 10 minutes of my time).

And so?

Change is already here. The rate of progress over the past year has been astonishing. Nor does it show any signs of slowing down.

Some are destined to repeat Ken Olsen’s mistake by refusing to believe. Others will suffer the fate of WordPerfect and Lotus and watch helplessly as new competitors reimage their products with new interfaces and features. Organizational bureaucracy and a failure to adapt will kill others.

And some will start to build the companies that will become the names of the future.

As DEC employees snuggled up in their bunks on the cruise ship back in 1988, how many imagined the company would soon be lost at sea and sunk within a decade? How many of us are set to repeat the mistakes of the past?