Waymo's 250,000-Ride Week Is Building a Closed Ledger on Public Roads

There was a specific moment when the story stopped being about driving. On a morning in May 2025, Alphabet's self-driving unit announced it would begin charging passengers across three cities at once: Denver, San Diego, and Tampa. The press release called it "operational scale" – the numbers deserved a sharper word. Ten months prior, Waymo was recording roughly 25,000 paid rides per week. By March 2025, that number had crossed 250,000. For a physical-world infrastructure company, one whose service depends on cars, asphalt, weather, and hundreds of millions of dollars of capital per city, a tenfold increase in adoption within a single year is not growth. It is a phase transition. The detail that stopped me cold was the geography. Denver sits at exactly 5,280 feet, with winter snow that arrives without warning. San Diego is Pacific fog crawling over rolling hills. Tampa lives in the Atlantic hurricane corridor, where summer thunderstorms drop two inches of rain in forty-five minutes. Every one of those environments stresses an autonomous perception system differently. Waymo chose all three simultaneously. That is not a marketing decision. It is a proof-then-expand statement aimed squarely at everyone who argued the technology could never generalize. I have spent most of the past decade thinking about trust. I audited DeFi contracts in my university years. I traced NFT provenance to centralized servers during the 2021 frenzy. I eventually co-authored a manifesto about cryptographic identity in the age of synthetic media – a thesis I called “The Proof of Soul.” So when I look at this expansion, I do not see a mobility story. I see the construction of the largest centralized ledger ever written. Not a ledger of money. A ledger of the physical world itself, written by a single corporate actor. Let me establish the baseline facts, because context matters. Waymo is Alphabet's self-driving arm – a sixteen-year project that emerged from Google X, accumulated over 20 million real-world test miles, and in late 2024 closed a $6 billion funding round that included Alphabet, Andreessen Horowitz, and Fidelity. On any reasonable technical assessment, Waymo operates the most commercially mature Level 4 system on earth. The debate over capability ended the moment its weekly ride count went vertical. The relevant question is no longer whether these vehicles can drive. They can. The relevant question is what happens to the trust architecture around them as they scale – and whether anyone outside the company will ever be able to independently verify what they actually do on public roads. A word about my own bias before I go deeper. During the 2020 DeFi Summer, I worked as a community liaison for a lending protocol that genuinely empowered people locked out of the traditional banking system. That experience gave me an abiding respect for what permissionless systems can do. But it also showed me the pathology of the space – wash trading, predatory algorithms, the quiet moral exhaustion of watching freedom become speculation. I retreated to a cabin in the Alps for two weeks to process the dissonance. What I brought back was a conviction I still hold: any meaningful analysis of decentralized technology must be forensic about centralized power and honest about decentralized failure. This essay tries to do both. Here is the uncomfortable core of the matter. While the blockchain community spent the last decade arguing about fee markets, finality layers, and the semantics of true decentralization, Alphabet built a system that makes a high-stakes trust decision millions of times per week. Nobody outside the company can verify any of those decisions. The vehicle's perception, planning, and control stack is a closed box. Its safety statistics are published in self-authored reports. And now the company is replicating that closed architecture across multiple cities simultaneously. I first encountered the danger of unverifiable claims in 2018, while volunteering to audit the smart contracts of etherTrust, a fledgling DeFi prototype. I found a reentrancy vulnerability in its donation logic – one that would have allowed an attacker to drain roughly $200,000 in a single transaction. The fix was trivial. The lesson was not. It was my first glimpse of what happens when trust rests on the claim that code is safe, made by people who simply have not looked hard enough. Code that has been read holds a different promise than code that hasn't. The same is true for miles driven. Consider what Waymo vehicles actually produce. Every second, each vehicle generates between one and ten gigabytes of sensor data – LiDAR point clouds, radar returns, camera frames at thirty frames per second. A single operating vehicle in a twelve-hour day accumulates tens of terabytes. Now scale that across a fleet of several hundred vehicles per city, across three cities that are deliberately ecologically distinct. The dataset assembled is not a testing corpus. It is a high-resolution, continuously updated memory of the physical infrastructure of American urban life. Every traffic-light cycle, every pothole, every pedestrian's habitual crossing, every fire hydrant, every queue at a drive-thru: all of it archived and owned by a single corporate entity. In crypto, we call a single source of truth an architectural flaw. Waymo calls it a high-definition map and treats it as a competitive moat. The parallel to my NFT investigation is too exact to ignore. In 2021, amid the generative-art mania, I traced the metadata of a prominent project called CryptoSculptures to a centralized server. The community believed they owned their art on-chain. What I found was that the pointer to the actual art lived in a server controlled by the project team; if that server disappeared, every “owner” held a token referencing a void. My analysis triggered significant backlash – I was accused of killing the culture. But a small group of developers reached out to thank me for clarity. Truth isolates before it liberates. Now I am asking the same question about Waymo's maps, its safety logs, its perception bottle-neck decisions: where does the ground truth live? The answer is Google Cloud, guarded by Alphabet's corporate security policy, summarized in self-published safety reports. The road is public infrastructure. The ledger of the road is a private asset. This is the provenance problem applied to the territory of everyday life – and the industry narrative treats it as a feature rather than a defect. There is a quieter monetary debate hiding inside the robotaxi expansion. An autonomous vehicle must settle transactions constantly: per-minute parking meters, time-of-use electricity for charging sessions, per-mile tolls, congestion surcharges, micro-insurance premiums for every mile it covers. Conservative estimates suggest a mid-sized fleet in a city like Denver initiates tens of thousands of micro-transactions per hour during peak operations. This is the machine-to-machine payment economy that cryptocurrency was supposed to own. But Waymo will not settle on a permissionless rail. It will use Alphabet's existing stack – dollars, banking integrations, centralized identity, Google-scale latency. The machine economy of the near future will not boot on public blockchains by default. It will boot on permissioned rails, and the window for permissionless settlement in the physical world may close before most crypto builders realize it was ever open. I have to be honest about why that is. Permissionless payment rails have failed for a decade to capture the high-frequency, low-latency, micro-value settlement space. Lightning Network is the clearest case; seven years after its deployment, channel management overhead, liquidity constraints, and routing failure rates have confined it to a niche used by enthusiasts. The centralized settlement layer that Waymo will use faces none of these problems because it owns the counterparties inside its own perimeter. The uncomfortable reality is that for machine-speed settlement, permissionlessness has not solved the routing problem – and the routing problem, in the end, is not an engineering bug but a social coordination cost that the decentralization community has not yet found a way to amortize. Yet the choice of settlement rails is a fork in the blockchain sense. If the machine economy hardens on closed rails within the next three years, the network effects will be locked inside a corporate perimeter for a generation. If there is a future in which any robotaxi can pay any charging station regardless of fleet operator, where a delivery bot can settle an insurance claim without leaking its route to the customer, where a fleet can cryptographically prove it paid its congestion fees without revealing its passengers – that is a protocol question, not a software product question. And it is a question that becomes much harder to answer once the centralized system reaches critical mass and defines the default interface of the physical world. The dimension of this expansion that worries me most is the one the coverage barely touched: safety asymmetry. The industry tells a clean story – Waymo vehicles are involved in fewer incidents per million miles than human drivers. The claim may be true. But “fewer collisions than humans” is not the same as safe. It is the same logical error as saying a smart contract is secure because all its external calls are validated before state updates. In my etherTrust audit, the reentrancy vulnerability was not in a component anyone wrote carelessly. It was in the interaction between a donation function and a withdrawal path – two individually correct pieces of code that formed a recursive failure when composed. The same failure class applies to autonomous driving. We are composing perception, planning, and control systems; any one of them may be correct in isolation. But the threat lies precisely in the interaction between them – a perception model that correctly identifies a pedestrian in rain, a planning module that correctly selects a braking trajectory, a control system that correctly applies the brakes, until the recursive composition of those correct behaviors produces an outcome that is incorrect. We have seen what a single instance of that failure does to the industry. In October 2023, a Cruise vehicle dragged a pedestrian who had already been struck by another car. The subsequent regulatory freeze stalled autonomous operations across the United States for months and erased a material share of public confidence in the entire category. Because here is the asymmetry: when a human driver makes a mistake, it is an individual event. When an autonomous system makes a mistake, it is treated as evidence against the technology itself. The acceptable accident rate for a deployed autonomous system is not “better than human average.” It is “better than the most forgiving human context for any single incident” – and that bar is set by public perception, not by statistical modeling. Crucial to Waymo's continued expansion is the fact that it operates in a state of single-point-of-failure exposure: one severe crash in one new city could suspend the entire narrative across all cities, much as a compromised smart contract can drain liquidity from every protocol built on the same underlying framework. There is also the human layer, which the expansion announcement omitted almost entirely. Rideshare drivers in the United States number between 1.5 and 2 million. Waymo will not lay them off. It will simply lower the clearing price of their labor, block by block. No severance, no strike, no public acknowledgment – just a gradual substitution of a variable human cost with a fixed machine cost, in cities where the service is available. When I taught blockchain fundamentals to underserved teenagers in Milan during the depths of the 2022 bear market, I had to ground my enthusiasm in something real. I came to believe that a technology's worth is measured by who it includes. Autonomous mobility has a genuinely liberating potential – for elderly riders, disabled individuals, and residents of transit deserts. But I notice the expansion narrative does not mention the people whose livelihood it quietly displaces. That silence is a choice. Now the contrarian turn. The instinct of the decentralization community, my community, is to denounce Waymo as a surveillance machine and warn everyone to stay clear. The data does not support that posture. As of mid-2025, Waymo performs 250,000 paid trips per week. No decentralized ride network has achieved even a thousandth of that volume. The open-coordination stack has no current answer to the capital intensity of autonomous fleets, the data requirements for validating perception in snow, or the legal accountability that any crash demands. If a robotaxi injures someone, the public needs an entity that can be sued into a settlement the public accepts. A DAO does not fit that role. The Lightning parallel is uncomfortable, and I will name it plainly. Centralized systems solved machine-speed settlement problems that permissionless systems failed to crack. Google's payment stack has no routing failures because it owns the counterparties within its boundary. This is sobering for the decentralized thesis. Efficiency is the most reliable centralizing force in technological history; permissionless coordination pays off only after the initial coordination costs are paid, and our community repeatedly fails to confront the pattern of building a tool and then waiting for someone to need it badly enough to absorb those costs. Waymo did not wait. Alphabet wrote checks for sixteen years and watched the network effects compound. But centralized architecture has an upper bound. If autonomous fleets scale to ten million vehicles across the United States, the volume of micro-disputes – toll rebellions, parking violations, insurance disagreements, charging disputes – will exceed anything a centralized settlement desk or a traditional court system can absorb. At that scale, the marginal cost of resolving a contested decision must approach zero. That is not a services problem. That is a protocol problem. The first entity to encounter that limit will be the first to discover the structural ceiling of centralized settlement – and that discovery will create space for an open rail that can provide verifiable, low-cost resolution without a trusted intermediary. The road is becoming the largest common ledger humanity has ever written. Every mile driven is a transaction in a global consent economy – the public grants access to shared infrastructure, and in exchange, the autonomous system writes the physical world into corporate memory. The question of who writes that ledger, who can read it, and who can independently verify its entries is not a question about autonomous vehicles. It is the foundational political question of the coming decade, identical in structure to the one we should have resolved for monetary networks before we let the window close on payment rails. If the ledger of the physical world is written by a single corporation and sealed from public audit, then we have engineered the most sophisticated centralized trust architecture in history — a system that asks us to trust not because it is verifiable, but because the alternative is too complex to investigate. That is a failed definition of trust. In my etherTrust audit, the truth was available to anyone who looked at the code. In my CryptoSculptures investigation, the provenance was discoverable – buried, but findable. In the autonomous fleet, the ledger is closed by design. And so the only honest role for permissionless technology in this story is not to build a competitor ride network that will never match Waymo's capital, nor to promise machine payments that cannot yet sustain the throughput. It is to build the verification layer: an open, auditable record of machine decisions, machine identities, and incident provenance that does not belong to any single corporate actor. The Proof of Soul applies to machines as much as to humans. When a machine drives, it makes moral decisions on our behalf. We deserve a public record of those decisions that we can actually audit. The road is a proof – of movement, of judgment, of accountability. The best proof is the one you don't have to take on faith.