Case Study 2: Waymo's Long Road to ROI — When AI Value Takes a Decade

Case Study 2: Waymo's Long Road to ROI — When AI Value Takes a Decade

The Bet

In 2009, Google launched a project that would have seemed absurd to any traditional ROI analysis. The goal: build a fully autonomous vehicle — a car that could drive itself on public roads, in real traffic, without human intervention. The project was initially called "Chauffeur" and was housed within Google X, the company's semi-secret research lab dedicated to "moonshot" technologies.

The project had no revenue model. No clear path to commercialization. No precedent in the market. The technology required breakthroughs in computer vision, sensor fusion, robotics, machine learning, and real-time decision-making under uncertainty — each of which was, at the time, an active and unsolved research problem.

The person leading the project was Sebastian Thrun, a Stanford professor who had won the DARPA Grand Challenge — a competition to build a self-driving vehicle that could navigate a desert course. Thrun's desert vehicle was a proof of concept. Turning it into a car that could navigate a busy intersection in downtown San Francisco was an entirely different challenge.

Google's leadership — Larry Page and Sergey Brin in particular — backed the project with the conviction that autonomous driving would eventually transform transportation, logistics, and urban planning. The specific business model could be figured out later. The technology came first.

This is the purest form of a moonshot AI investment: enormous upfront cost, profound technical uncertainty, no clear timeline, and transformative potential if successful.

The Investment

The precise amount Google (and later Alphabet, its parent company) has invested in autonomous driving is difficult to pin down, because the spending has been reported under different entities and accounting categories over the years. However, external estimates, analyst reports, and Alphabet's own financial disclosures provide a rough picture.

Phase 1: Google X (2009-2016)

During this phase, the project was a research initiative within Google X. Spending was not disclosed separately but was embedded in Google's overall R&D budget. External estimates suggest cumulative investment of $1 billion to $1.5 billion during this period, covering salaries for a growing team of engineers (which grew from approximately 20 people in 2009 to over 300 by 2016), hardware costs (custom LIDAR sensors, computing platforms, prototype vehicles), and operational costs (test tracks, safety drivers, regulatory engagement).

The project achieved remarkable technical milestones: by 2015, Waymo's vehicles had driven over 1 million miles on public roads, primarily in Mountain View, California, and Austin, Texas. But there was no revenue. Zero.

Phase 2: Waymo LLC (2016-2020)

In December 2016, Google spun the project out as a separate company called Waymo, a subsidiary of Alphabet. This structural change had two purposes: to provide greater operational independence and to make the investment more visible (and accountable) to investors.

During this period, Waymo's spending accelerated. The company hired aggressively (growing to over 1,500 employees), built a testing facility in a former military base in California, expanded testing to more cities (including Phoenix, Arizona, where it launched its first limited public service), and developed its fifth-generation hardware platform.

Cumulative investment estimates for this period range from $2 billion to $3 billion, bringing the total to approximately $3.5 billion to $4.5 billion. In 2020, Waymo raised $2.25 billion from external investors in its first-ever funding round — a signal that Alphabet wanted to share the financial burden and validate the project's value through market pricing.

Phase 3: Commercialization Attempt (2020-Present)

From 2020 onward, Waymo shifted focus from pure research to commercial deployment. The company launched Waymo One, a robotaxi service in Phoenix (initially with safety drivers, then fully driverless in a limited service area) and expanded to San Francisco, Los Angeles, and Austin. By 2024, Waymo One was serving over 100,000 paid trips per week — a significant milestone, though still a tiny fraction of the ride-hailing market.

Alphabet's financial disclosures show that its "Other Bets" segment (which includes Waymo along with other ventures) has consistently reported operating losses. In 2023, Other Bets reported an operating loss of approximately $4.8 billion on revenue of $1.5 billion — with Waymo accounting for a significant but undisclosed portion of both the losses and the revenue.

By 2025, conservative estimates place Waymo's cumulative investment at $5.7 billion to $8 billion, with some analysts citing figures as high as $10 billion when fully burdened corporate costs are included. Waymo's annual revenue run rate, while growing, remained well below the level needed for profitability.

The ROI Question

If you applied a standard five-year NPV analysis to Waymo in 2014 — five years into the project — the calculation would have been devastating. Investment: approximately $1 billion. Revenue: zero. NPV at any positive discount rate: deeply negative. IRR: negative. Payback period: undefined.

If you applied the same analysis in 2019 — ten years in — the picture would have been only marginally better. Investment: approximately $4 billion. Revenue: negligible (from a limited Phoenix pilot). NPV: catastrophically negative.

And yet, Alphabet continued to invest. Why?

The Strategic Optionality Argument

Alphabet's investment in Waymo cannot be understood through traditional ROI analysis. It must be understood through the lens of strategic optionality — the fourth pillar of AI value from Section 34.2.

Waymo's potential addressable market is enormous. The global ride-hailing market alone was valued at approximately $130 billion in 2024, with the broader autonomous vehicle market (including trucking, logistics, last-mile delivery, and personal vehicles) estimated at over $1 trillion by 2035. If Waymo captures even a modest share of this market, the return on its cumulative investment could be astronomical.

But "could be" is the operative phrase. The option has value precisely because there is a chance — not a certainty — that it will pay off enormously. And the option's value is enhanced by the fact that very few organizations in the world have the financial resources, technical talent, and institutional patience to make this bet.

The Knowledge Asset Argument

Waymo's 15+ years of autonomous driving research have produced a knowledge asset that is extraordinarily difficult to replicate. By 2024, Waymo's vehicles had driven over 20 million miles on public roads and tens of billions of miles in simulation. This data — annotated, processed, and used to train increasingly sophisticated models — represents a competitive moat that no new entrant can quickly match.

The knowledge asset extends beyond data. Waymo has developed proprietary sensor hardware, custom machine learning architectures for perception and prediction, a full-stack autonomous driving software system, and deep expertise in safety engineering for autonomous systems. These capabilities have applications beyond ride-hailing — in trucking (Waymo Via), logistics, and potentially defense and industrial automation.

The Competitive Preemption Argument

Google invested in autonomous driving partly to prevent competitors from establishing an unassailable lead. If autonomous driving is transformative — and most industry analysts believe it will be — then not investing carries its own risk: the risk of being left behind in a trillion-dollar market transition.

This is the dark side of strategic optionality. Sometimes the value of an investment is not the return it generates but the catastrophic loss it prevents. If Google had not invested in autonomous driving and a competitor had achieved full autonomy first, the impact on Google's core business (advertising, which depends on people using their phones and computers — not sitting in self-driving cars consuming alternative content) could have been severe.

Lessons for AI Portfolio Management

The Waymo case offers several lessons that connect directly to the AI ROI framework in this chapter.

Lesson 1: Moonshots Require Different ROI Metrics

Applying a standard NPV analysis to a moonshot project is not just unhelpful — it is misleading. Moonshots have payoff structures that resemble venture capital investments or financial options: high probability of partial or total loss, small probability of enormous return. The expected value calculation must account for this asymmetry.

The venture capital approach is more appropriate: invest a fixed amount (bounded downside), accept a high failure rate (most bets lose), and design the portfolio so that the winners more than compensate for the losers. This is the portfolio management logic from Section 34.8 — and it is why moonshots should represent no more than 10 to 15 percent of an AI budget.

Lesson 2: Time-to-Value Can Exceed Any Reasonable Forecast

In 2009, optimistic forecasts predicted fully autonomous vehicles on public roads by 2020. In 2025, Waymo operates in a handful of cities under constrained conditions. The time-to-value for autonomous driving has exceeded even the most conservative estimates.

This is not unusual for breakthrough AI technologies. AI has a long history of underestimating the difficulty of "last mile" problems — the gap between a technology that works in controlled conditions and a technology that works reliably in the real world. This gap is where most AI value is created or destroyed, and it is consistently underestimated.

Business Insight: When evaluating AI time-to-value, apply the "Hofstadter's Law" heuristic: it always takes longer than you expect, even when you take Hofstadter's Law into account. For moonshot projects, multiply your initial time estimate by 2-3x. For transformative projects (not moonshots, but projects that require significant organizational change), multiply by 1.5-2x. Budget your patience accordingly.

Lesson 3: The Difference Between Research Value and Commercial ROI

Waymo has generated enormous research value — advances in computer vision, sensor fusion, reinforcement learning, simulation, and safety engineering that have influenced the broader AI field and benefited Alphabet's other products. Google's self-driving car research contributed to advances in Google Maps, Google Cloud AI services, and Alphabet's robotics initiatives.

But research value is not commercial ROI. A CFO cannot pay dividends with research citations. The gap between research value and commercial value is one of the most important distinctions in AI ROI analysis — and one of the most commonly blurred, especially by AI teams seeking continued funding.

For Ravi's portfolio, this distinction is clear: the research value of the killed visual merchandising project is acknowledged (the team learned about computer vision for retail applications), but research value alone does not justify continued investment. The project needed commercial value — a clear business process it improved and a measurable financial impact — and it did not have one.

Lesson 4: Who Can Afford Moonshots?

Waymo's investment was possible because Alphabet generates approximately $300 billion in annual revenue and $80 billion in annual operating profit (2024 figures). The $5.7 billion+ invested in Waymo over fifteen years represents less than one year's operating profit. Alphabet can absorb this investment without existential risk.

Most organizations cannot. A company with $500 million in revenue cannot afford to spend $50 million on a moonshot AI project with a decade-long payback horizon and a high probability of failure. The financial cushion simply does not exist.

This has a practical implication for AI portfolio management: the appropriate allocation to moonshots depends on the organization's financial capacity to absorb losses. Large, profitable companies can allocate 10 to 15 percent of their AI budget to moonshots. Small and mid-size companies should limit moonshot allocation to 5 percent or less — or consider partnering with research institutions rather than building in-house.

Lesson 5: Patience Has Limits

Even for Alphabet, patience has limits. The external funding round in 2020 ($2.25 billion) was partly a signal that Alphabet wanted external validation of Waymo's value — and external investors to share the risk. Subsequent rounds brought the total external investment to over $5 billion, diluting Alphabet's ownership but also distributing the financial burden.

The lesson: even moonshot investors need an exit ramp. The question is not "will you be patient?" but "how much patience can you afford, and what milestones will tell you whether patience is being rewarded?"

The Competitive Landscape

Waymo's story is inseparable from its competitive context. Other companies have made enormous autonomous driving investments:

Cruise (General Motors): Approximately $10 billion invested before suspending operations in 2023 following a safety incident in San Francisco. GM subsequently restructured the effort as an "in-house" project focused on personal vehicles rather than robotaxis.
Uber ATG: Uber invested approximately $2.5 billion in autonomous driving before selling the unit to Aurora Innovation in 2020 — essentially writing off the investment.
Tesla: Has invested heavily in its "Full Self-Driving" system, which uses a camera-only approach (no LIDAR). Tesla's approach is controversial in the industry and, as of 2025, has not achieved full autonomy.
Motional (Hyundai/Aptiv joint venture): Approximately $4 billion invested, with commercial robotaxi service launched in Las Vegas.
Zoox (Amazon): Acquired for $1.2 billion in 2020, with continued investment estimated at $1-2 billion.

The competitive landscape illustrates a key feature of moonshot investments: many companies invest, but only a few will capture significant returns. This is the power-law distribution at the project level, mirroring the power-law distribution at the organizational level described in Case Study 1.

What Waymo Means for Your ROI Analysis

Professor Okonkwo uses the Waymo case to close the lecture on a nuanced note.

"Waymo is the extreme case," she says. "Most of you will never manage an investment of that scale or that time horizon. But the principles are universal."

She lists them:

Different project types require different ROI frameworks. You would not evaluate a quick-win automation project with the same framework you use for a multi-year platform investment. Know which type of project you are evaluating and choose the appropriate methodology.
Strategic optionality is real value — but it is not a blank check. The option value of AI investments justifies a premium over traditional ROI hurdles, but it does not justify unlimited spending without accountability. Even Waymo has investors, milestones, and governance.
The hardest discipline is knowing when patience becomes stubbornness. Pre-committed kill criteria, quarterly reviews, and honest assessment of progress are not bureaucratic obstacles — they are the mechanisms that distinguish disciplined investment from wishful thinking.
Portfolio balance is the ultimate risk management strategy. Quick wins fund the portfolio and build credibility. Strategic bets drive growth. Moonshots provide optionality. A portfolio composed entirely of any single category is fragile.

"The question Waymo forces us to ask," Professor Okonkwo concludes, "is not 'was it a good investment?' — we will not know that for years. The question is: 'Given the information available at each decision point, was the process for evaluating and continuing the investment sound?' That is the discipline of AI ROI measurement. It does not guarantee good outcomes. It guarantees good decisions."

Discussion Questions

At what point, if any, should Alphabet have killed the Waymo project? Apply the kill criteria from Section 34.7. If none of the criteria apply, explain why moonshots require different governance mechanisms than standard AI projects.
A startup founder pitches you the following: "We're building autonomous delivery robots. We need $50 million over five years. The ROI is uncertain, but the addressable market is $200 billion." Using the lessons from Waymo, what questions would you ask before investing? What portfolio allocation framework would you apply?
Compare Waymo (continued investment, approaching commercialization) with Cruise (operations suspended) and Uber ATG (sold at a loss). What factors explain the different outcomes? Is it possible that Cruise and Uber made the right decision to exit even if Waymo eventually succeeds?
Waymo's cumulative data asset (20+ million road miles, billions of simulation miles) is often cited as a competitive moat. Under what circumstances could this data advantage be eroded or made irrelevant? Consider technological shifts (e.g., Tesla's camera-only approach), regulatory changes, or data-sharing mandates.
Apply the simplified option value formula from Section 34.5 to Waymo. Estimate the probability of Waymo achieving profitability within five years, the potential annual revenue at scale, and the cost of replicating Waymo's capability from scratch. Is the option value calculation consistent with Waymo's valuation?

This case study is based on publicly available information from Alphabet's SEC filings, Waymo's public disclosures, industry analyst reports (Morgan Stanley, UBS, ARK Invest), and reporting from The Information, Bloomberg, and the Financial Times. All financial figures are estimates based on publicly available data and should be treated as approximations.