Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
This is an excerpt from The Breakdown newspaper. To read the full article, register.
“You have to enjoy failure.”
—James Dyson
In a 2003 research paper, Google experts he shared their new approach to building “reliable computers from clusters of unreliable PCs.”
This was a big change. Instead of paying for high-end Sun servers designed for high reliability, Google was embracing cheaper, more expensive hardware with free Linux software. They then developed additional software to deal with frequent hardware failures.
They found that volume is cheaper than accuracy through technology.
This is the model for today’s AI data center, too: You can’t have a three-month training crash because one GPU dies out of 20,000.
It is not, however, a GPU-only model.
In fact, semiconductor manufacturers are built to produce GPUs with very low errors, which require near perfection in every critical process – unlike Google’s idea of a data center.
But this is not a technical ambition. It is a financial necessity.
“Why do we need to have 99.9999% reliability at every step of the process?” Ben Thompson he asks. “It’s all about the cost of capital equipment.”
That’s mainly because the $200 million ASML machines that each plant uses to make advanced semiconductors “are some of the most beautiful things in the world,” Thompson explains.
This creates assumptions around cost and perfection: Manufacturing plants must be near-perfect to be profitable because ASML machines are expensive – and ASML machines are expensive because they must be perfect.
As a result, it now costs $20 billion to manufacture a single semiconductor fabric.
Thompson compares the process of manufacturing semiconductors and hyper-complex ASML machines to the process of making rockets at SpaceX, where Elon Musk chooses to “blow up a bunch of Starships because you don’t want to waste anything on a problem that isn’t really important.”
Repeatedly firing rockets helps SpaceX find “the exact line where you want to be,” as Thompson puts it – the line between complexity and utility that allows them to avoid the upfront costs of spaceflight that are needed perfectly.
In other words, failure may be part, not the fault, of complex systems.
It’s about the journey…
Failure is a big part of crypto systems, too: every blockchain is designed with the assumption that some part of its nodes will be malicious or faulty.
This makes blockchains resilient – like Google’s data center, they continue to create valid blocks even if they fail. But the error tolerance does not extend to the running code on blockchain.
Smart contract code, written by criminals, will contain errors or design flaws. This can be expensive, because unlike traditional finance, there is no reversal of a crypto transaction in the event of misuse of a contract or protocol.
The code removes the middleman, but at the cost of zero tolerance.
This makes code-is-law crypto a system in itself wisely they agree to be tolerant of mistakes but structurally it requires perfection.
There is no way to make your way around it.
In traditional finance, where practices are regulated and regulated, banks may strive to achieve greater efficiency by enforcing stricter KYC/AML regulations and hiring more officers.
In crypto finance, where systems are unlicensed and transactions cannot be changed, we can only seek to learn from our mistakes.
There have been many: from the DAO use that nearly sank Ethereum in 2016 to the Bybit hack for $1.5 billion in 2025.
However, luckily, this will be seen in retrospect as Elon Musk’s exploding rockets. Each exploit exposes an attack vector — replays, oracle changes, protocol design flaws — that can be implemented in the future.
Like the explosion of the Starship, the failures show what needed to be hardened against what was not necessary to care – each an opportunity to find the right balance of difficulties and benefits in the new type of economy.
Crypto is often judged by its failures, which is appropriate, given its frequency and size.
But failures are not bugs in a plan to build a new financial system. They and process.
To enjoy crypto, you must enjoy failure, too.
Happy New Year.
Get news in your inbox. Check out the Blockworks blog post:
