The Token Weapon: Why Alphabet Could Pull the Trigger on AI's Trillion-Dollar Private Market

On 10 June 2026, the founders of Google Ventures sat down with the All-In hosts and laid out a scenario that, if it plays out, would redraw the economics of the artificial-intelligence industry and the venture funds that have spent the last three years underwriting it. The argument: Alphabet has the cash, the distribution, and the product parity to weaponise token pricing against OpenAI and Anthropic, and the rational move for the company is to do exactly that.

The implications run well beyond two frontier-model competitors. The same pricing move would revalue the trillion dollars in private AI marks that pension funds, endowments, and — via S&P 500 exception rules — retirement savers are now sitting on. It would also vindicate a contrarian thesis Bill Maris has been refining since he left Google in 2016: that the venture industry's centre of gravity has shifted decisively toward smaller funds, and that the mega-funds printing the most spectacular paper returns are structurally incapable of converting them into cash.

A war-chest scenario

Maris did not present the token-pricing gambit as speculation. He framed it as the obvious move for a company that built its moat on distribution and now watches two challengers extract rent from that distribution through APIs. "If I were Google, that's what I'd do," he told the All-In hosts. "Tokens as a weapon. Grab market share. Grab an install base on consumer and enterprise."

The mechanism is simple enough to describe in one paragraph. OpenAI and Anthropic sell access to their frontier models by the token — a unit of compute billed per million inputs and outputs. Gemini, Alphabet's competing model, is, by Maris's characterisation, "a basically identical product" at the API layer. If Alphabet cuts its token prices by 80% — a figure he flagged as illustrative — the gross margins on competing model providers compress toward what he called a "super critical" level, while Alphabet's enterprise sales force uses the price gap to push Gemini into the same contracts.

The playbook is not novel. Uber ran it for a decade in ride-hailing. Amazon Web Services ran a milder version of it in the mid-2010s. What is novel is the absolute scale of the AI revenue base the move would distort. The All-In hosts placed current AI revenue at roughly $60 billion against aggregate infrastructure and capital commitments approaching $1 trillion — a ratio Maris described as "unsustainable" without either a demand surprise or a supply-side shakeout.

The small-fund rebellion

The pricing war, in Maris's telling, is downstream of a deeper structural problem in the venture industry. He cited DPI data — distributions to paid-in capital, the only metric he considers honest — showing that funds below $750 million in size have averaged 4.76x DPI, while funds above $1 billion have averaged 2.42x. Ninety-five per cent of top-decile performers fall into the smaller cohort. The compression, he emphasised, is discontinuous: returns do not degrade gracefully with fund size, they collapse.

The arithmetic is unforgiving. A $500 million fund needs $5 billion of exit value to return capital and $15 billion to hit 3x. A $7 billion fund needs $210 billion — a figure Maris noted "exceeds the total venture-backed M&A and IPO exit value in most years." The math explains why the mega-funds that have raised the most capital in the current cycle are also the most exposed to a liquidity event in which token pricing resets the underlying market.

Maris's own firm, Section 32, sits firmly in the small-fund cohort. The first fund was raised at $150 million; the six funds that have followed average roughly $400 million, and Maris said all six are performing in the top decile. The portfolio includes CrowdStrike, Cohere, and Coinbase. The track record, he implied, is not a personal boast but a validation of a thesis the data already supports.

The bag-holder problem

The third leg of the argument is the one most likely to land on retail readers. Companies are staying private longer, raising at valuations that exceed what public-market investors can underwrite through discounted cash flow, and then entering the S&P 500 through index exception rules that force passive funds — and the 401(k) accounts that hold them — to buy the stock at whatever price the late-stage private rounds established.

Maris did not soften the framing. "Don't say you're doing this for the benefit of humanity and do the other thing," he said, addressing the rhetoric around AI's public-benefit mission. "Make the public's retirement accounts the bag holders."

The mechanism is structural, not conspiratorial. Index inclusion rules were written for a market in which companies went public at saner valuations. The exception clauses allow private companies to enter the S&P 500 at the mark established in their last funding round, even when that mark implies revenue multiples that cannot be defended on cash-flow grounds. Pension funds and target-date retirement accounts must buy because the index tells them to. Early private investors — the endowments, the family offices, the sovereign wealth funds that got in at a fraction of the late-stage price — exit into that demand.

The asymmetry is stark. Maris noted that 75% of venture funds lose money, and that the general-partner incentive structure actively rewards the outcome. A $5 billion fund returning 1.01x — the 75th percentile — generates more carry for the GP than a $500 million fund returning 3x. The mega-fund is paid for existing; the small fund is paid for performing. The LPs who anchor mega-funds — chiefly endowments — are effectively subsidising the GP carry structure with the assumption that the cycle will eventually produce liquidity.

The infrastructure wager

For investors still allocating to the AI theme, Maris drew an analogy to the gaming industry's history. "I think we're at the Atari command line stage of AI," he said, "and we're going to get to the PlayStation 10 stage in the next 5 years." The implication: today's frontier models are the Zork-equivalent text adventures of a medium that will, within a half-decade, be photorealistic and ubiquitous. The venture returns, he argued, will accrue to the infrastructure layer — physics engines, controllers, GPU fabrics, platforms — rather than to the foundation models themselves.

This is where the token-pricing scenario and the venture thesis converge. If Alphabet compresses model-layer margins, the foundation-model cohort revalues downward. The infrastructure picks — the chips, the data centres, the model-serving platforms, the agent frameworks — retain their pricing power because they sit below the commoditisation line. Section 32's portfolio tilt toward Cohere, the model-serving layer, and away from foundation-model pure-plays is consistent with that read.

The talent drain

The final beat of Maris's appearance touched on biotech and the long-term cost of US science policy. He argued that the gutting of the CDC and NIH, combined with H-1B friction, is driving top scientific talent to China — where, he said, recruiting from Europe and India is proceeding at a pace the US is not matching, and where experiments in cloning and other frontier biology are happening that American funding will not support.

For investors with Section 32–style exposure to computational biology — Maris founded Google's Calico and holds a stake in New Limit, the longevity company co-founded by Brian Armstrong — the talent drain is a competitive moat for non-US labs and a warning sign for the domestic pipeline. The same dynamic that makes Chinese battery and solar manufacturers formidable is now being seeded in biotech.

Stakes

The scenario Maris sketched is not inevitable. Alphabet may choose margin preservation over market-share aggression; OpenAI and Anthropic may have distribution moats he underweights; the $1 trillion in private AI marks may eventually be validated by revenue growth that today's ratio of capital deployed to dollars booked does not yet imply. The All-In hosts pushed back, arguing that a few mega-funds will produce "unbelievably excessive" returns from the current cycle. Maris's reply was that those paper returns must eventually find a buyer who can underwrite them through discounted future cash flows — a buyer who has not yet been validated for the trillion-dollar private cohort.

For 401(k) holders, the question is whether the S&P 500 exception rules that bring overvalued private companies into the index will be reformed before the next major mark-down. For LPs in mega-funds, the question is whether the 1.01x-at-75th-percentile reality will, at last, redirect allocations toward the smaller managers whose DPI math actually works. For OpenAI and Anthropic, the question is whether Gemini's product parity will translate into a price war Alphabet decides it can afford to fight.

The scenario Maris described is, in the end, a question about incentives. The venture industry's GP economics reward the wrong behaviour. The index industry's rules reward the wrong entry prices. The frontier-model industry's economics depend on a margin structure that one well-capitalised incumbent can collapse at will. Each of those incentive problems is, separately, addressable. Together, they describe a market structure in which a single pricing decision by Alphabet could revalue trillions of dollars of paper wealth and rebalance the venture industry's centre of gravity in a single quarter. The question is no longer whether Alphabet can do it. It is whether the rest of the stack has built enough of a defence to make it not worth doing.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://www.youtube.com/watch?v=0umrMuUClC4