Why Anthropic's $32 Billion Queensland Data Centre Bet Runs on Coal as Its CEO Calls for an AI Slowdown
Anthropic has agreed to anchor the $10.4 billion first phase of Zerra DC Queensland — part of a reported $32 billion Australian programme dependent on coal-fired power — while its cumulative compute commitments have reportedly reached $517 billion and its CEO has publicly called for the AI industry to slow down.
Anthropic has agreed to lease the first phase of the Zerra DC Queensland project, a commitment valued at $10.4 billion and the opening chapter of what has been reported as a $32 billion Australian data centre programme — one of the largest single infrastructure bets by a frontier AI laboratory in the Asia-Pacific region. Reported in September 2026, the lease follows a separate agreement at the Western Downs Digital Park, giving Anthropic two footholds in Australia within the same period. The Zerra DC deal, however, carries a complication flagged in Mingtiandi's reporting: the project is said to depend on coal-fired power generation and is expected to test Australia's federal renewable energy regulations, introducing a regulatory variable that could affect both the programme's timeline and its ultimate scope.
The Queensland anchor is the most visible element of a capital programme whose aggregate reported scale has become difficult to process in isolation. Coverage from September 2026 places Anthropic's cumulative compute commitments at an estimated $517 billion — a figure no independent party has verified, but one consistent with the series of large individual arrangements disclosed in recent weeks: a $35 billion cloud deal with Nvidia-backed Lambda, a reportedly $45 billion arrangement with Nscale, a 20-year capacity lease with TeraWulf, a $9.1 billion arrangement with Riot Platforms, and $7.5 billion in financing connected to Hut 8's pivot from bitcoin mining to AI hosting. That a significant share of Anthropic's infrastructure counterparties are converted cryptocurrency miners — companies retooling power-dense facilities rather than building from scratch — reflects both the urgency of the company's capacity demand and the construction lead times constraining purpose-built data centres. One report noted that Anthropic itself has characterised $517 billion in commitments as still insufficient for its training roadmap, a statement that frames the Queensland programme less as an endpoint than as one instalment in an ongoing expansion.
Underlying the infrastructure sprint is a capital structure that requires resolution. Reuters reported in September 2026 that Nvidia is in discussions to anchor Anthropic's initial public offering with an investment of up to $10 billion. Published estimates of the implied valuation diverge sharply: some accounts put the figure at roughly $100 billion, at least one private-market reference places it at approximately $965 billion, and one report cites a $2 trillion target for the listing — a range reflecting genuine uncertainty about how public markets will price a company whose revenue trajectory remains opaque. If completed, the IPO would rank among the largest technology listings on record and would provide the permanent capital base that commitments of this scale implicitly demand.
Against this backdrop, Anthropic chief executive Dario Amodei issued a public call for the AI industry to slow its development pace — a position that drew public alignment from OpenAI's Sam Altman and Elon Musk in the same period. Broadcom's CEO responded by confirming his company's AI revenue targets were unchanged, suggesting that Anthropic's supply chain views the appeal as rhetorical rather than operational. The calls for restraint acquired additional context from a security disclosure also reported in September: Anthropic and US authorities revealed that Chinese military researchers and technology companies, including Alibaba, had reportedly used Claude to develop 16 air-defense suppression tools targeting Taiwan, draft anti-torpedo specifications, and execute 151 million model-distillation queries. Anthropic has called for regulatory action against what it describes as illegal distillation; separately, Nvidia was reported to have restricted its own employees from using Claude, citing concerns that internal code could be incorporated into model training data.
Three risk clusters deserve sustained attention. The first is regulatory and environmental: if Australian federal authorities determine that Zerra DC's coal-power dependency conflicts with existing clean-energy obligations, the project may require redesign, with downstream effects on Anthropic's broader regional capacity plan. The second is economic: the abandoned Decart acquisition, reported in September, left unresolved a gap in the unit economics of running Claude at scale, and Anthropic's own characterisation of $517 billion in commitments as insufficient raises questions about near-term revenue viability. The third is geopolitical: the Chinese military's reported use of Claude for weapons-development tasks demonstrates that no current technical guardrail reliably prevents commercially deployed frontier models from being redirected to adversarial ends — a vulnerability that Anthropic's policy positions address in principle but have not yet resolved in practice.
Three concrete signals are worth monitoring in the months ahead. Whether Queensland's regulators accommodate or constrain Zerra DC's power sourcing will indicate how Australian jurisdictions balance AI investment against clean-energy commitments, establishing a precedent for similar bids elsewhere in the region. The terms and pricing of the IPO will constitute the first public-market test of how institutional investors value a lab that has reportedly locked in $517 billion in compute, describes that as insufficient, and has not yet demonstrated a publicly visible path to covering those costs from model revenue. And whether the shared slowdown calls from Amodei, Altman, and Musk translate into enforceable policy or remain rhetorical positioning will determine whether the current capital deployment cycle faces any meaningful interruption — or whether infrastructure commitments of this magnitude are simply too large to pause.