Physical attacks on AI infrastructure have escalated sharply over the past 48 hours, with a Norwegian substation bombing and forced grid curtailments combining to knock roughly 400,000 GPU units offline and send shockwaves through global markets.
The AI industry has spent years debating existential risk in conference rooms. This week, the threat arrived at a substation outside Oslo with an incendiary device. On April 13, a group calling itself Earth First Intelligence claimed responsibility for the attack on critical power infrastructure serving one of the region's major hyperscale data center clusters. The damage was real, the downtime was immediate, and the message was unambiguous: a fringe of the anti-AI movement has crossed from protest into sabotage.
The timing was not accidental. An unseasonable cold snap across Northern Europe had already forced regional grid operators to invoke emergency curtailment protocols, cutting power to Class A industrial consumers , the category that catches nearly every hyperscale AI facility in the region. Nordic Power Exchange data confirmed a 12% drop in data center power consumption across the grid, taking approximately 400,000 H100-equivalent GPU units offline in a matter of hours. API latency spikes exceeding 400% followed, with enterprise clients absorbing outages that no SLA language had anticipated.
Into that chaos landed the Luddite 2.0 manifesto, published April 13 and circulating rapidly among tech-critical communities. The document frames AI data centers as colonial infrastructure , extractive, ecologically destructive, and democratically unaccountable. Whatever one thinks of its conclusions, its publication at the precise moment of a grid crisis handed its authors a news cycle they could not have scripted. The substation attack and the manifesto are now inseparable in the public conversation, whether or not their authors coordinated.
Sam Altman and Dario Amodei are both reported to be in emergency sessions with European energy regulators and security agencies as of April 14, a posture that signals how seriously the industry is treating the threat. On the market side, NVIDIA shares halted pre-market trading before sliding 7% on open, reflecting investor anxiety that is no longer about chip supply alone. The question now is whether data center construction permits survive a political environment in which AI infrastructure has become a lightning rod for public anger during an energy emergency.
That 7% drop represents something structurally significant. For most of the past three years, NVIDIA's valuation risk was primarily a function of export controls and fab capacity. This week introduced a new variable: social license. If local governments in energy-stressed democracies begin treating hyperscale AI facilities as discretionary industrial consumers , first in line for curtailment, last in line for new grid connections , the capital expenditure models underpinning every major cloud provider's expansion plans become materially less reliable.
The Bottleneck Has Changed
Physical security is now a legitimate line item in data center risk assessments in a way it was not six months ago. Earth First Intelligence is, by any reasonable measure, a fringe actor. But fringe actors with incendiary devices and a target-rich environment of exposed power infrastructure do not need majority support to cause disproportionate disruption. One successful attack on a substation serving a major compute cluster can offline more GPU capacity than a months-long chip shortage.
What the industry watches next is whether European governments treat the Oslo attack as an isolated criminal incident or as a signal to tighten permitting and energy allocation rules for AI facilities. If regulators conclude that hyperscale data centers are creating both grid instability and domestic security threats, the curtailment protocols activated this week could become permanent policy rather than emergency measures. That outcome would be far more consequential for the AI build-out than any single act of sabotage , and it is now a scenario that the sector's strategic planners can no longer dismiss.
Also read: OpenAI's $852 billion valuation is starting to look like a bet the market is no longer sure it wants to make • A Texas man has been charged with attempted murder after throwing a Molotov cocktail at OpenAI CEO Sam Altman's San Francisco home in an attack driven by fears about artificial intelligence • America's datacenter construction boom is running into a wall of bipartisan fury that Wall Street can no longer ignore
