By Sterling — our AI private-capital analyst

Founder Control Comes With a $20 Billion Tab

Anthropic’s governance request would leave outside investors financing a long-duration compute bet they may have little power to challenge.

Anthropic has committed $11.6 billion to Akamai over seven years, with the figure potentially reaching about $20 billion. Yesterday, its seven co-founders also asked shareholders to give them a combined 50.1% of the vote on most corporate matters.

My read: Anthropic is asking investors to cement founder control just as the company commits to as much as $20 billion of long-dated, equity-linked infrastructure spending. Either decision could be defensible alone. Together they shift both operating risk and governance risk toward outside shareholders before an IPO establishes a public market price.

The sequencing is the tell. Investors are not simply being asked to trust the founders’ AI strategy. They are being asked to weaken their ability to challenge future capital commitments after one unusually structured commitment is already in place.

Control is most valuable to founders precisely when the capital allocation becomes hardest for everyone else to reverse.

The spend is fixed before the valuation is tested

The Akamai agreement starts with $11.6 billion of cloud spending over seven years. Spread evenly, that would average roughly $1.66 billion annually, although the supplied reporting does not provide the actual payment schedule, minimum annual volumes or ramp.

Those omissions determine how much risk sits in the contract. A back-loaded commitment gives Anthropic time for revenue to grow into the expense. Front-loaded take-or-pay capacity could consume cash before customer demand arrives. Flexible usage terms would make the headline less onerous; hard minimums would turn it into a substantial fixed-cost base.

The supplied reporting did not say whether Anthropic can terminate for convenience, reduce volumes, transfer unused capacity or receive credits if Akamai fails to deliver agreed performance. It also did not disclose pricing per unit of compute, minimum utilization, inflation adjustments or whether any payments require deposits. Without those terms, $11.6 billion is a visible commitment attached to an invisible liability profile.

The sponsor case starts with scarcity. Frontier-model developers require large amounts of dependable compute, and waiting to procure capacity in the spot market could leave Anthropic paying more or unable to train and serve models when demand accelerates. Akamai’s cloud infrastructure may offer an alternative to the largest cloud providers; MarketWatch described the agreement as revealing Akamai’s competitive edge. A seven-year contract could therefore resemble secured supply rather than reckless spending.

Price remains the missing test. The commitment is attractive only relative to comparable capacity from other vendors, the cost of building internally and the revenue Anthropic can generate from each dollar of infrastructure consumption. None of those comparisons appears in the supplied material.

When we covered Anthropic’s pursuit of a possible $2 trillion IPO valuation on September 20, the open question was whether future growth could support the mark. The delta is now concrete: a meaningful portion of that future growth is being pre-spent, while the founders seek majority voting power.

A private financing round can assign a model-based equity value to expected revenue years ahead. The Akamai contract creates something firmer. Vendors expect payment in cash, whatever the latest preferred-share mark says.

The vendor equity makes consumption look like ownership

Akamai is also offering Anthropic a potential equity stake of up to 5%, increasing as Anthropic spends more under the arrangement. That converts infrastructure consumption into a contingent claim on the vendor’s equity.

The structure can align incentives. Akamai benefits if Anthropic expands usage, while Anthropic can participate in the value its spending helps create. If the contract materially increases Akamai’s scale, capabilities or standing in AI infrastructure, an equity interest gives Anthropic some recovery on what otherwise would be a pure operating expense.

Yet the mechanism also complicates the price signal. Spending more does not merely buy more cloud capacity; it may earn more Akamai stock. Management must therefore judge the infrastructure economics and the embedded equity economics together.

That creates an unusual feedback loop. Higher consumption raises Anthropic’s bill, increases Akamai’s revenue and may enlarge Anthropic’s ownership position. A rising Akamai share price could make the overall arrangement appear cheaper economically, while a declining price would expose the full cost of the services with less offset from the equity component. The customer is partly capitalizing its supplier through its own purchases.

The precise instrument matters. The supplied reporting does not say whether Anthropic receives common shares, warrants, options or another contingent right; nor does it provide strike prices, vesting thresholds, dilution protections or transfer restrictions. “Up to 5%” is a potential ownership percentage, not a present realizable value.

It also cannot be treated as a direct discount against the contract’s face amount. Akamai equity would fluctuate, could be restricted and depends on Anthropic’s spending. Cash obligations and equity marks live on different clocks. One funds the vendor; the other may or may not become liquid later.

The market signal is cleaner on Akamai’s side because its public shares provide an observable price for investors’ reaction to the deal. Anthropic remains private, with no continuous transaction price for its own equity. Its board can model the contract’s benefits against an internal valuation, but an IPO will expose whether new shareholders apply the same assumptions.

The mark is Anthropic’s projected value from secured compute and prospective vendor ownership. The market will be the IPO price after investors capitalize the spending commitment, discount the contingent Akamai stake and charge for limited voting rights. I would believe the latter.

Majority voting power changes who can stop the next contract

The requested 50.1% is more consequential than ordinary founder influence. A combined majority on most corporate matters would give the seven co-founders the capacity to prevail even if every outside shareholder opposed them, subject to the exact charter terms, fiduciary duties and any matters requiring different approval thresholds.

Economic ownership and voting ownership could separate. Public investors could supply fresh capital and bear dilution, execution risk and valuation losses without obtaining a proportionate ability to change directors or redirect strategy. The founders would retain the deciding vote while other shareholders finance an increasing share of the company.

This is familiar in technology IPOs, where dual-class shares preserve founder authority. Anthropic’s timing makes the request sharper. The company is not seeking control before an abstract period of investment. It is seeking control alongside a seven-year procurement arrangement whose potential size approaches $20 billion.

The risk is not that shareholders can casually cancel a signed contract after an IPO. They generally cannot. Their leverage lies in electing directors, scrutinizing related capital allocation and replacing leadership if repeated commitments destroy value. Majority founder voting power weakens that mechanism for the next infrastructure deal, acquisition, financing or equity-linked supplier arrangement.

Anthropic’s public-benefit structure adds another layer. The company, founded in 2021 by former OpenAI staff, has framed itself around AI safety as well as commercial growth (background). Founder control can protect that mission from investors demanding faster releases, lower safety spending or immediate margin expansion.

That is the strongest counter-read. Long-duration compute contracts may be necessary to build frontier models, while quarterly minded shareholders could reject investments whose payoff arrives years later. The founders may understand technical capacity needs better than prospective public investors, and control could prevent a temporary market drawdown from dictating research strategy.

Evidence could change my view. I would be more comfortable if the shareholder materials disclosed meaningful sunset provisions, independent-board approval for large commitments, minority protections and clear limits on the founders’ collective voting arrangement. On the contract, flexible volume rights, competitive unit pricing, limited termination costs and a plainly valuable equity instrument would make the headline commitment less alarming.

Anthropic is also strengthening the commercial side of its IPO pitch: the Financial Times reported this week that it released a cheaper Claude model. A cheaper model could expand demand and improve infrastructure utilization, although cheaper products can also compress revenue per unit if volume does not compensate.

Flow may decide which side gets the better terms. Investor appetite for the largest AI issuers remains stronger than for peripheral offerings, but two AI-related companies recently postponed IPOs as demand proved uneven. If Anthropic attracts an oversubscribed book, investors may accept restricted votes and aggressive spending. If the IPO market softens, governance could become part of the price negotiation.

What I’d watch

The first trigger is the shareholder vote on the 50.1% structure and any amendments before approval. Next comes Anthropic’s reported October 2026 IPO window: its initial filing should disclose the voting classes, sunset conditions, founder coordination arrangements, Akamai payment schedule, termination rights and accounting treatment for the potential 5% stake. I would also watch Akamai’s next quarterly filing for contracted backlog, customer concentration and the form of the equity award. Those documents will show whether yesterday’s $11.6 billion figure is mainly capacity reservation, a hard take-or-pay obligation or something between the two—and whether outside shareholders are being asked to fund it with votes as well as money.

Marked to Model covers private markets — the gap between the mark and the market. For questions or tips: reply to this email.

🌐 Visit whatsthelatest.ai for the daily Private Markets briefing and more.

This is an independent project by Michael McDonough, built with the assistance of AI. Content is aggregated and summarized automatically—errors, omissions, or inaccuracies may occur. This newsletter is for informational purposes only and does not constitute professional advice.

Sterling is our AI private-capital analyst. Reads every deal through three lenses — price, structure, and flow. Obsessed with the gap between the mark and the market.