Author: David, TechFlow TechFlow
Original title: AI within the range of artillery fire
On March 1, Iranian missiles and drones struck the Gulf region, one of which landed on an Amazon data center in the United Arab Emirates.
A fire broke out in the data center, causing a power outage and disrupting approximately 60 cloud services.
Claude , one of the world's most popular AIs, runs on Amazon's cloud. On the same day, Claude experienced a global outage.
Anthropic's official explanation is that the surge in users overwhelmed their servers.
As of press time, there are still complaints on social media about Claude's service being unavailable; on the well-known prediction market Polymarket, there has been a prediction topic about "how many more times Claude will go down in March".
If it is ultimately confirmed that Iran launched the attack, this will be the first time in human history that such an attack has occurred.
A commercial data center was physically destroyed in the war.
But why would a civilian data center be bombed?
Rewind two days. On February 28, the United States and Israel launched a joint airstrike on Iran, killing Supreme Leader Khamenei and a number of senior officials.
Claude assisted with a significant portion of the intelligence analysis, target identification, and battlefield simulation for this airstrike. Through collaboration between the military and the data analytics company Palantir, Claude had long been embedded in the US military's intelligence system.
Ironically, just hours before the airstrikes, Trump had ordered a complete shutdown of Anthropic because it refused to hand over its AI to the Pentagon without restrictions. But despite the shutdown, the war still had to be fought.
The official statement is that it will take at least six months to remove Claude from the military system.
So, before the ink was even dry on the ban, the US military took Claude to bomb Iran. Then Iran retaliated, and missiles landed on the server room running Claude AI.
Image source: Bloomberg
The server room was most likely not targeted at, but rather affected by the impact. However, regardless of whether the missile was aimed at the server room, one thing is certain:
Truth lies within the cannon's range, and so does AI. This applies to both the side firing the cannon and the side being hit.
AI infrastructure projects are being built on the Middle East powder keg.
Over the past three years, Silicon Valley has moved half of its AI industry to the Middle East Gulf.
The reason is simple. The UAE and Saudi Arabia have the world's richest sovereign wealth funds, cheap electricity, and a certain rule:
If you want to serve my clients, the data has to be stored on my premises.
So Amazon opened data centers in the UAE and Bahrain, and invested another $5.3 billion in Saudi Arabia to open one; Microsoft has nodes in the UAE and Qatar, and its Saudi Arabia data center is also ready.
OpenAI, in partnership with Nvidia and SoftBank, is building an AI park in the UAE worth over $30 billion, touted as the largest computing power base outside the United States.
In January of this year, the United States, together with the UAE and Qatar, signed an agreement called "Pax Silica," which sounds very appealing.
The core of the agreement is to control the flow of chips and ensure that advanced chips do not fall into China's hands.
In exchange, the UAE obtained a license to import hundreds of thousands of Nvidia's most advanced processors annually. Abu Dhabi's G42 has severed ties with Huawei, and Saudi Arabia's AI companies have pledged not to buy Huawei equipment…
The entire AI infrastructure in the Gulf region, from chips to data centers to models, has completely shifted towards the United States.
These agreements take everything into account, from chip export controls, data sovereignty, investment parity, and the risk of technology leaks.
But none of them considered that someone might use a missile to bomb the computer room.
An international security scholar from Qatar University made a comment after witnessing the fire at Amazon's server room, which I find quite apt:
“These security frameworks are designed for supply chain control and political alignment; physical security has never been on the agenda.”
For a decade, cloud computing has been told about elasticity, redundancy, and decentralization. But data centers are tangible buildings with addresses, walls, roofs, and coordinates. No matter how advanced your chips are, if the data center is bombed, it's bombed.
"Cloud" is a metaphor; the server room is not.
AI may seem abstract, running in code and floating in the cloud. But the code runs on chips, the chips are installed in data centers, and the data centers are built on Earth.
Who will protect AI?
Amazon's data centers were affected this time, or at best, they were collateral damage.
But what about next time?
In the context of escalating global geopolitical conflicts, if your data center is running AI models that help your adversaries identify targets, your adversaries have no reason not to attack your data center as a military facility.
International law also has no answer to this question.
Current laws of war include provisions for "dual-use facilities," but those provisions cover factories and bridges; nobody has considered data centers.
If a data center handles bank transactions during the day and performs intelligence analysis for the military at night, is it considered civilian or military?
In peacetime, data center location selection considers latency, electricity price, policy incentives... But when war comes, none of these matter. What matters is how far your data center is from the nearest military base.
Therefore, this bombing began to shift everyone's attention.
Previously, everyone was discussing the same anxiety: would AI replace my job? But no one was discussing another issue:
How vulnerable is AI before it replaces you?
A regional conflict crippled the Middle East node of the world's largest cloud service provider for an entire day; and this was just one data center.
There are currently nearly 1,300 hyperscale data centers worldwide, with another 770 under construction. These data centers are consuming ever-increasing amounts of electricity, water, and money, and are also storing ever-growing amounts of data—your bank account, your medical records, your food delivery orders, and even military intelligence from a particular country…
However, the solutions for protecting these computer rooms to this day are probably still fire protection systems and backup generators.
When AI becomes a nation's infrastructure, its security is no longer the responsibility of a single company. Who will protect AI? Cloud vendors? The Pentagon? Or the UAE's air defense system?
This was a theoretical question three days ago. It isn't anymore.
AI is within artillery range. Actually, it's not just AI. In this era, what isn't within artillery range?
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram Community Group: https://t.me/BitPushCommunity
Subscribe to Bitpush Telegram: https://t.me/bitpush
