SpaceX, OpenAI and Anthropic IPOs Risk Entrenching Founder Control

By Gill Whitehead — Visiting policy fellow, Oxford Internet Institute; member of the advisory council of Frontier Economics

After weeks of waiting, investors finally have sight of the prospectus for the initial public offering of SpaceX. OpenAI and Anthropic could join Elon Musk’s firm in floating shares later this year. The three prospective IPOs show how the race to control frontier AI is defined not only by vision and scale, but by the concentration of corporate control.

All three firms have sought to govern cutting-edge technologies with custom-made rules, departing from contemporary corporate governance designed to check unfettered control by an individual through effective board oversight, risk-based incentive structures and robust succession plans.

Why founder control is back in fashion

The logic is not without reason. The global financial crisis revealed that most failing banks actually met the criteria for good governance on paper. The recent backlash against heavy compliance costs and short-term market pressures — seen as a drag on innovation — has prompted a move towards founder-centric models.

The speed and scale of experimentation in corporate governance is, however, unprecedented, writes Gill Whitehead. And given the high degree of uncertainty as to the nature of AI risks, the tools of regulation, legislation, ethics and governance will be needed to manage the impact of these companies.

Three ways the IPO designs break with standard governance

1. Super-voting equity that silences public investors

The first departure is the entrenchment of super-voting equity. Adopting the founder-control model pioneered by Alphabet and Meta, dual-class share structures bypass the “one share, one vote” principle. Founders retain voting control despite owning a minority of the equity, in effect silencing public investors.

Provisions in SpaceX’s IPO filings cement a structure granting Mr Musk and other insiders shares with ten times the voting power of ordinary shares, securing him a clear voting majority. This will be combined with the use of SEC-permitted “controlled company” exemptions that forgo an independent board or compensation committee.

2. Jurisdiction-shopping and a fast track into the Nasdaq-100

The second departure involves jurisdiction-shopping and regulatory arbitrage. Because the scale of these listings is historic, American states and stock exchanges are jostling to secure them. States are competing by offering increasingly robust statutory protections to directors. Stock indices, meanwhile, have increasingly skewed their listing standards to accommodate tech giants.

SpaceX is using this competition to its advantage. Following a Delaware court decision to void Mr Musk’s Tesla pay package on the grounds of weak board independence — later overturned — he moved SpaceX to Texas and its director-friendly Business Court, later merging xAI into the rocket firm.

Furthermore, SpaceX plans to list on Nasdaq, which would enable it to benefit from the exchange’s new “fast entry” rule that allows newly listed large firms to enter the NASDAQ-100 after just 15 trading days, rather than three months. As a result, index funds are legally compelled to buy stock almost immediately, further weakening the leverage of active fund managers to extract governance changes.

3. Hybrid structures meant to insulate the mission

The third departure is the use of hybrid structures. The anticipated IPOs of OpenAI and Anthropic may yet enshrine their novel corporate architectures, designed to insulate their technical missions from standard venture-capital incentives.

In October OpenAI completed its restructuring into a for-profit Public Benefit Corporation (PBC) in which the original non-profit foundation now has a 26% stake. Although the foundation retains the legal authority to appoint and remove the PBC directors — and thus the chief executive — the intentional significant overlap in foundation and PBC board members, many of whom were installed after the reinstatement of Sam Altman as chief executive, indicates that operational power lies increasingly with the CEO.

Anthropic has made perhaps the most creative undertaking to enshrine its safety-first mission into its corporate governance. Its Long-Term Benefit Trust enables independent experts to elect a majority of directors over time. However, even under this model trustees serve only one-year terms and must consult the CEO on appointments. Crucially, a supermajority of shareholders retains the power to dissolve the trust without trustee consent. The trust provides an advisory guardrail, but the founders and their main backers — Amazon and Google — retain ultimate authority.

Why the experiments matter now

There is much to admire in these experiments; corporate governance must always be evolving. But they are untested in situations with such high stakes for humanity. The irony is that as industries scramble to engineer digital AI guardrails, many of their corporate-governance equivalents have just been dismantled.

The Musk v Altman trial spent weeks interrogating the trustworthiness of AI’s elite, only to be dismissed this week on procedural grounds. Yet it exposed a raw truth: these historic IPOs and the humanity-impacting potential of their technologies now rely on the precarious scaffolding of a few founders’ good intentions.

What investors and regulators are weighing

Investors have noticed. Institutional groups, led by public pension funds, are lobbying for governance changes such as clauses to limit the length of time for which dual-class structures apply. In the absence of change, responsible investors must weigh the immense financial upside of heavily oversubscribed IPOs against the lack of independent board oversight and accountability.

Governments and regulators should take notice, too. The hubris is not in believing transformative technology can be built. It is in believing that individual judgment is the appropriate final line of defence. Anthropic’s Mythos preview last month showed that regulators are already struggling to keep up. Future AI rules must be designed with these corporate structures in mind. Backstops must not be removed precisely when they are most needed. ■

Gill Whitehead is a visiting policy fellow at the Oxford Internet Institute and a member of the advisory council of Frontier Economics.

Report a correction