By Du Jie

Edited by Sun Congying

Dario Amodei, former VP of Research at OpenAI, co-founded Anthropic, which has risen to global AI leadership with its Claude large model.

While news of Anthropic's IPO draws attention to its revenue growth, model capabilities, and trillion-dollar valuation expectations, few notice its governance structure—a departure from Silicon Valley's conventional model.

This is unsurprising. Silicon Valley tech IPOs typically follow a fixed template for control: dual-class shares (AB stock) concentrating voting rights with founders.

Anthropic, however, eschews tradition, opting for a Public Benefit Corporation (PBC) superposition long-term independent trust (LTBT) structure.

Here, power rests not with founders or investors but with external trustees—non-shareholders without profit-sharing rights. Former Federal Reserve Chair Ben Bernanke is among these trustees.

This design challenges capital market 's (capital markets') long-held consensus: public companies must prioritize shareholder returns.

But when AI safety and public missions supersede capital gains, how should secondary markets price this?

Do investor equities secretly depreciate? This is no abstract legal question but a tangible test for Anthropic's IPO and Wall Street.

Its trajectory will set a precedent for all large-model firms eyeing public listings.

Dual-Layer Framework: Trust Holds Real Power

In U.S. capital markets, most tech firms adopt traditional C Corp structures.

For such companies, legal mandates require management and boards to prioritize shareholder profits in major decisions. If management forgoes short-term gains for safety or social impact, shareholders can sue for damages.

Public Benefit Corporations (PBCs) are also for-profit entities. They can raise capital, issue equity, and eventually IPO, competing like ordinary corporations.

Their core innovation? Redefining board legal liabilities.

Anthropic's governance comprises two layers: a PBC as the legal entity and a long-term interest trust (LTBT) wielding actual control. Together, they form a unique constraint mechanism.

Under PBC rules, management need not prioritize shareholder returns. Public safety and social impact can legally factor into decisions.

Divergent Governance Paths in AI: Anthropic's Middle Ground

This corporate form isn't Anthropic's invention. Outdoor brand Patagonia, crowdfunding platform Kickstarter, and ice cream maker Ben & Jerry’s adopted PBC structures early.

Among global AI leaders, three distinct governance paths have emerged, exemplified by OpenAI, Anthropic, and xAI.

OpenAI, whose commercial arm is registered as a PBC, uses a hybrid model: a non-profit foundation at the top retains control, with a for-profit subsidiary handling commercialization to curb capital excesses.

Anthropic takes a more balanced approach. While both aim to limit capital's influence, their structures differ fundamentally.

OpenAI's apex is a non-profit foundation; Anthropic remains a for-profit PBC.

Anthropic merely strips control from itself, vesting it in external trustees via an independent trust. OpenAI's foundation holds equity; Anthropic's trust holds only voting control.

In 2023, Elon Musk founded xAI, initially mirroring OpenAI as a PBC before abandoning the structure for a standard C Corp.

xAI sheds mission-based constraints, prioritizing commercial growth with flexible capital operations and unrestricted expansion. Safety relies on corporate self-regulation, free from legal or trust-based checks.

Anthropic straddles both worlds. It retains market financing channels without OpenAI's non-profit shackles while using independent trusts to hardwire safety missions, reducing the likelihood of short-term capital impulses overriding long-term risk management.

Anthropic seeks to reconcile two incompatible goals: commercial freedom and AI's public responsibilities.

This divergence reflects a deeper industry split: Are super-large models ordinary businesses or infrastructure-like technologies with inherent public risks? Different judgments yield vastly different corporate systems.

Why Do Large-Model Firms Prioritize Corporate Responsibility?

Large-model technical risks have strong spillover effects. Blind commercialization risks societal harm.

Under traditional corporate governance, capital could pressure management to loosen model controls or launch high-risk ventures for short-term financial gains. PBCs legally empower boards to prioritize AI safety over profits.

But Anthropic argues PBCs alone are insufficient. They legally protect boards but can't prevent shareholders or founders from pivoting under commercial temptation. Hence, the LTBT long-term independent trust adds a second layer of constraint.

The PBC is merely the legal vehicle; true power lies with the LTBT trust.

Trustees, external experts with no Anthropic equity or dividends, receive compensation covering only their time.

They wield two core powers: (1) replacing most board members, controlling personnel decisions; (2) requiring five days' advance notice for major transactions (e.g., large debt issuance, selling >50% equity, transferring core AI models above thresholds).

Legal scholars note this five-day window isn't mere notification but a soft veto right.

If the trust deems a transaction unsafe, it can block it via personnel changes or sustained pressure.

This differs fundamentally from AB stock models: AB stock concentrates voting power with founders, keeping control internal. Anthropic externalizes this power to third parties independent of shareholders and founders.

Long-Term Moats and Inevitable Valuation Discounts

Anthropic's structure cedes some shareholder rights upfront.

This sparks polarized views, especially between primary and secondary markets.

The bull case is clear: the trust, as a third-party gatekeeper, avoids Silicon Valley's typical control battles and founder-investor conflicts, ensuring strategic consistency.

Amid tightening global AI regulations, a governance structure with built-in safety constraints serves as a compliance moat. Capital can't pressure the team into technical risks for quarterly results, securing long-term resources for R&D and safety evaluations.

Primary market investors favor this long-term narrative. Secondary market concerns are equally real and represent the biggest pricing uncertainty.

The core issue is structural dilution of shareholder rights. Investors buy shares but lack influence over transactions that could reshape the company.

Major decisions rest with trustees unaccountable to shareholders. If the trust vetoes a lucrative merger or licensing deal on safety grounds, shareholders have no recourse.

Traditional due diligence can model revenues, margins, and competition but not trustees' subjective judgments.

This creates a paradox: the company has top-tier technology and hypergrowth—all traits of a high-potential stock—but its equity lacks complete rights. Markets will discount for this deficiency.

Primary markets can absorb this discount for long-term upside; secondary markets, with their shorter investment horizons, view such governance risks more cautiously.

This Poor cognition (cognitive gap) will persist in Anthropic's pricing dynamics long after its IPO.

An Industry-Wide Governance Experiment: More Tests Ahead

In the long run, this transcends Anthropic's IPO success.

Short-term, AI hype may overshadow governance concerns, ensuring smooth issuance and strong demand.

Mid-term, post-IPO volatility and major transactions will reignite debates over the trust's overreach and opaque rules. If trustee decisions clash with investor interests, shareholder lawsuits, institutional rebalancing, and valuation revisions will follow.

Long-term, its demonstration effect is significant. As global AI safety regulations tighten, reliance on corporate self-regulation and founder discretion will wane.

If Anthropic's model proves stable, balancing commercial expansion and risk control, it could become a blueprint for leading AI firms.

Capital markets, built on shareholder primacy for centuries, have rarely faced fundamental governance challenges from technological waves. Internet and semiconductor booms didn't alter this logic.

AI is different. Its risk spillovers and societal impact dwarf those of prior tech sectors. Old capital rules may not fit this new technology.

We're witnessing a slow institutional adaptation. Should capital cede power for public safety? How to weigh commercial gains against long-term risks? The industry lacks answers.

Anthropic's IPO is an industry-wide trial. Its outcome will shape not just the company's stock story but also how future AI giants enter public markets.