Anthropic plans to buy 1.4-GW Australian data centre capacity, AFR reports

Anthropic is seeking at least 1.4 GW of Australian data-center capacity for a potential ~$15B project, aiming to begin using at least 1 GW by end-2027 after opening its Australian office earlier this year. The company is tendering the opportunity to major operators (including CDC Data Centres, AirTrunk, Nextdc, Iren and Stack), with shortlisted bidders meeting executives in Canberra in April and a final decision expected in at least six weeks. Anthropic may split the buildout across four to five providers rather than awarding a single developer, signaling scaling demand for AI compute infrastructure in Australia.

Analysis

This is less a pure AI-demand story than an infrastructure scarcity story. In Australia, the constraint is increasingly usable megawatts, not headlines about model demand, so the economic value accrues to operators with power access, permits, and execution speed. A multi-provider structure actually caps upside for any single name because the prize is likely to be parceled out, while the real winner is whichever balance sheet can finance phase-one capacity without destroying returns on capital.

IREN looks better positioned for convexity because power optionality can be monetized across multiple workloads, not just one customer, which matters if the project is staged over several years. NXDCF has the local franchise advantage, but large campus wins often come with higher upfront capex and longer payback, so the equity market may eventually discount the growth if funding costs stay elevated. Second-order beneficiaries are grid gear, transformers, cabling, and Australian energy suppliers that can lock in long-duration load; losers are smaller colo operators without firm power and any competitor forced to bid on price alone.

Near term, the catalyst is binary over the next 6-8 weeks, but the true value driver is 12-18 months of interconnect, permitting, and financing execution. The thesis breaks if Anthropic opts for a smaller staged commitment, if power delivery slips beyond 2027, or if the market realizes the project is mostly optional capacity rather than contracted demand. The move is probably underpriced on sentiment but not yet a clean immediate trade until award structure and MW allocation are clearer.

Key Decisions for Investors

• Maintain a watchlist, not a chase: wait for the final award structure before sizing IREN or NXDCF; the key variable is not the rumor but whether phase-one MW is firm and financeable.
• If a meaningful single-vendor or dominant allocation is announced, use IREN for the higher-convexity expression; target a 10-20% upside window over 1-3 months, with thesis failure if the stock gives back the post-announcement gain within 2 sessions.
• Treat NXDCF as a lower-beta implementation trade only if it discloses power-secured capacity and customer pre-commitments; otherwise avoid buying the headline because capex intensity can dilute equity IRR.
• Relative-value setup: long the better power-secured data center operator, short a weaker colo peer if the market starts to price all Australian DC names equally; exit if financing spreads widen or project timing slips.
• Set an alert for any disclosure on interconnection timing or staged MW delivery; if first capacity moves beyond 2027, fade the entire cluster as the market will likely de-rate the growth multiple.