Ethereum co-founder Vitalik Buterin has weighed in on the debate over AI data centers, arguing that slowing the growth of hyperscale AI compute could give society extra time to prepare for cyber and biological risks.
What Buterin said
In a post on X on November 8, 2026, Buterin defended recent movements opposing new data center construction, focusing on the warehouse-sized facilities that train and run the largest AI models.
His core claim centers on algorithmic efficiency: work that would require more data centers two years from now could potentially be handled by existing infrastructure within four years, and likely sooner.
Buterin also addressed the cost of a slowdown, indicating that an expected one-year slowdown in AI scaling would cause only minimal delays to scientific progress. He reasoned that many factors holding science back are not compute problems at all—running experiments, gathering data, and testing results take time regardless of how many GPUs are available.
As he frames it, the trade is lopsided: society gives up a small amount of speed and gains a meaningful window to strengthen cybersecurity and reduce biological risks.
A pragmatic case
Buterin’s remarks responded to an emotional backlash against anti-data-center movements, and he chose not to lean on more controversial arguments about artificial general intelligence posing existential risk. Instead, he made a narrower, practical pitch: one does not need to believe a superintelligence is coming for humanity to think defenses should be ready before the technology outpaces them.
The commentary got most of its traction in crypto-focused outlets, though the underlying debate is about AI infrastructure policy and how fast the industry should build—reaching well beyond blockchain circles.
Connecting to recent warnings
In October 2026, Buterin warned that advances in AI could significantly weaken lattice-based cryptography within two years, recommending larger parameters in cryptographic practice. Lattice-based schemes are a family of math puzzles used to protect data, widely discussed as candidates for security against future computing threats. Larger parameters make those puzzles bigger and harder, trading some efficiency for a fatter safety margin.
A month earlier, in September 2026, Buterin highlighted AI’s potential to strengthen cybersecurity by enabling formal verification of entire programs, rather than simply hunting for individual bugs. Formal verification is the difference between checking a bridge for visible cracks and mathematically proving it can bear a given load. Finding vulnerabilities one at a time is useful; proving that whole programs behave as intended is a far stronger guarantee, and it has historically been expensive and slow to do.
Put the three posts together and a clear picture forms: AI may threaten cryptography soon, AI may also help build much stronger defenses, and a slower compute buildout, in Buterin’s view, gives the defensive side time to mature.
What this means
For the crypto industry, the most concrete takeaway sits in the October warning rather than the November one. If AI really could weaken lattice-based cryptography within two years, protocol developers and wallet builders have a short runway to revisit their assumptions and parameter choices. Buterin’s argument for slowing data center growth is, in part, an argument for giving those developers that runway.
His emphasis on formal verification is also worth watching. Smart contracts are programs that hold real money, and bugs in them have long been a source of losses across the sector. Tools that could prove whole contracts correct, rather than just patching discovered flaws, would address one of crypto’s most persistent weaknesses.
There is no evidence in Buterin’s remarks of an immediate market reaction, and he did not tie his comments to any specific project or token.
On the policy side, Buterin’s intervention adds a recognizable voice to the anti-data-center camp without the usual existential baggage. Critics of rapid AI buildout often get dismissed as alarmists; a tech-native figure arguing that the cost of waiting is small and the benefit is real makes that harder to do.
The weak point in his case is the forecast itself. The claim that current infrastructure could handle future workloads within four years depends on algorithmic progress continuing at a healthy clip. If efficiency gains stall, a slowdown would cost more than he expects. Likewise, the idea that a one-year delay would barely dent science rests on his read that non-compute bottlenecks dominate.
Still, the question Buterin is posing is not whether AI is good or bad. It is whether the defenses are being built as fast as the capabilities, and what a modest delay might be worth if they are not.
