Anthropic’s $2T IPO, OpenAI’s Astra 6.1 cancelation, and Sydney’s data-center backlash—what’s really changing in AI power and policy?

Anthropic’s long-awaited IPO is moving closer to a potential $2 trillion valuation as new filings and prospectus details surface. MarketWatch reports that Anthropic’s revenue grew more than tenfold in 2025, but the cost of training and serving its models is also rising fast. The Financial Times adds that Anthropic warns of “existential risks to humanity” in its IPO prospectus while also disclosing a dramatic loss—about $8 billion—on roughly $4.6 billion in revenue. In parallel, Reuters reports that Anthropic leaders plan to retain control of the lab through a structure dubbed “Founder LLC,” framing it as prioritizing public good over pure market forces. Strategically, the cluster points to a new phase of AI governance where corporate structure, safety claims, and infrastructure permitting are converging. Anthropic’s “existential risks” language and Founder LLC control mechanism suggest an attempt to pre-empt political backlash and shape regulation before it hardens into binding constraints. OpenAI’s reported cancellation of the “Astra 6.1” model release—citing safety concerns raised by researchers during internal testing—reinforces that safety gating is becoming a competitive and reputational battleground, not just a technical one. Meanwhile, Australia’s Goodman Group withdrawing a planned 24-hour data center on the Sydney north shore after public fury shows that even when demand is clear, social license and federal/state policy shifts can abruptly reshape AI compute expansion. For markets, the immediate read-through is that AI capex intensity is rising while uncertainty around deployment timelines is increasing. Anthropic’s disclosed losses despite rapid revenue growth imply continued pressure on AI infrastructure providers and on investors’ expectations for unit economics, which can weigh on AI-adjacent equities even as the sector’s narrative remains bullish. Data-center permitting risk in Australia can tighten near-term supply of hyperscale capacity, potentially supporting pricing for colocation and power-linked infrastructure, and it can also affect regional cloud and GPU leasing sentiment. On the model side, cancellations like Astra 6.1 can shift attention toward safer releases and compliance-ready roadmaps, influencing sentiment around AI software platforms and enterprise adoption schedules. What to watch next is whether these governance and safety signals translate into concrete regulatory outcomes and investment pacing. For Anthropic, key triggers include how investors react to the combination of existential-risk framing, large disclosed losses, and the Founder LLC control terms, and whether underwriters or regulators demand changes to governance disclosures. For OpenAI, the next signal is whether Astra-class models are reworked, delayed, or replaced with safer variants, and whether internal safety findings become a public standard. For Australia, the next indicators are any revised data-center applications, state/federal policy clarifications on 24-hour operations, and whether public opposition spreads to other planned sites—each of which could alter compute supply expectations over the next 6–18 months.

Geopolitical Implications

• 01

  AI governance is becoming a strategic asset: corporate control structures (e.g., Founder LLC) may influence how quickly firms can comply with emerging regulation.

• 02

  Safety narratives are likely to be leveraged in policy negotiations, potentially affecting cross-border AI standards and procurement decisions.

• 03

  Infrastructure permitting in key markets (like Australia) can become a chokepoint for regional AI compute capacity, altering competitive dynamics with US-based hyperscalers.

• 04

  Public backlash and federal/state policy divergence can slow compute buildouts, shifting investment toward jurisdictions with clearer permitting pathways.

Key Signals

• —Any revisions to Anthropic’s IPO governance terms or prospectus risk language following regulator/investor feedback.
• —Whether OpenAI reintroduces Astra-class models under a revised safety framework or replaces them with different release milestones.
• —New Australian data-center applications, approvals, or rejections tied to 24-hour operations and power/environmental constraints.
• —Changes in AI infrastructure pricing signals (colocation rates, power availability) in Australia and comparable markets.