Is Waymo Actually Profitable? The Real Cost of the Robotaxi Revolution

Ben Lorica

Robotaxi Economics, Data Center Rebellion, and the Battle for AI Infrastructure.

Subscribe: Apple • Spotify • Overcast • Pocket Casts • AntennaPod • Podcast Addict • Amazon •  RSS.

Ben Lorica and Evangelos Simoudis discuss two critical technology infrastructure topics: the evolving economics of autonomous vehicles and growing local opposition to AI data centers. Simoudis explains how “end-to-end AI” is transforming robotaxi viability, comparing Waymo’s multi-sensor approach to Tesla’s camera-only strategy, while both explore the challenges of profitability and safety. They then examine “The Data Center Rebellion” — grassroots resistance from local communities facing electricity demands, water shortages, and noise pollution from AI infrastructure, set against the backdrop of U.S.-China technological competition.

Interview highlights – key sections from the video version:

Jump to transcript

Related content:

Support our work by subscribing to our newsletter📩

Transcript

Below is a polished and edited transcript.

Ben Lorica: We are back with Evangelos Simoudis. His blog can be found at corporateinnovation.co. We are recording this on the morning of February 6, 2026. Episode notes, including links to all the resources we discuss, are available at thedataexchange.media. If you’re watching on YouTube, please hit the subscribe button, or subscribe wherever you get your podcasts.

Let’s jump in. Our first topic is autonomous vehicles, a field Evangelos is deeply involved in. Our goal is to provide a quick update on the current state of the industry, starting with economics and financial viability. In San Francisco, Waymo and Zoox vehicles are everywhere. Evangelos, what is the current economic situation for these robotaxis?

Evangelos Simoudis: To understand how we “cracked the nut” here, we have to look at how prominent robotaxis have become. We see three major companies operating in China, and obviously, they are a fixture here in San Francisco. The real game-changer has been the shift to “end-to-end AI.” This is a different model for processing sensor input to control vehicle movement. This approach, now being adopted by everyone including Waymo, makes vehicles more capable of operating in larger geofences or “operating design domains.”

From an economic perspective, we now have to calculate the cost per mile versus the price charged per mile, as well as the cost per vehicle. We are seeing companies order increasingly larger fleets. For example, Uber has ordered roughly 30,000 vehicles from Lucid, while Waymo is ordering thousands of vehicles from Zeekr and Hyundai. Similar fleet expansions are happening in China.

Ben Lorica: Just to clarify for our listeners: many people still associate Uber primarily with human drivers.

Evangelos Simoudis: Uber is adopting a hybrid model. This is central to their financial viability. In their recent quarterly results, Uber noted that in certain cities like San Francisco, they will maintain a hybrid fleet of both autonomous robotaxis and human-driven vehicles. In other cities, they will stick exclusively to human drivers. This has significant economic implications for both Uber and its drivers.

Ben Lorica: Is Uber still just experimenting, or are they confident that autonomous vehicles will pay for themselves in specific locations?

Evangelos Simoudis: Uber is experimenting with several models. In cities like Atlanta and Austin, they are already offering rides using Waymo’s autonomous vehicles. However, a major unknown in the economic equation is Tesla. Tesla is currently testing robotaxis in the Bay Area and Austin—notably operating in Austin without a safety driver. I recently saw an Uber robotaxi in Menlo Park—the specialized gold vehicle Musk announced about a year ago.

As these enter production and deployment, the economic model will shift again because Tesla has a significant manufacturing edge. Furthermore, Tesla vehicles do not use the expensive array of sensors found in Waymo or Pony.ai vehicles. Because they eschew Lidar, the cost of a Tesla robotaxi is significantly lower, which will impact the broader economics of the industry.

Ben Lorica: You mentioned that Tesla doesn’t use Lidar, which raises safety concerns, but let’s set that aside for a moment. Many people may not realize that Waymo still uses human operators. While we don’t know the exact ratio—perhaps one operator for every five to ten vehicles—is the plan to increase that ratio to improve economics? And does Tesla use human operators?

Evangelos Simoudis: There are several drivers for these economics: tele-operators (human operators), their vehicle-to-human ratio, and insurance. Recently, the insurance company Lemonade agreed to use Tesla’s data to reduce premiums, which is a major factor for companies like Waymo and Uber. Regarding tele-operators, they will remain a necessity for the foreseeable future.

Ben Lorica: But has Tesla publicly admitted to using tele-operators?

Evangelos Simoudis: They haven’t admitted it, but I believe they are. There is almost no way they aren’t using them. Ultimately, whether you use Tesla, Uber, or Waymo, it comes down to trust. Some people use Tesla’s Full Self-Driving (FSD) daily as if it were a Level 4 system. Many experts in the autonomous ecosystem argue it isn’t there yet, but roughly 12% to 20% of Tesla owners have licensed FSD and use it daily, which is a remarkable number.

Ben Lorica: What do we know about Waymo’s current economics in a restricted area like San Francisco? They’re still losing money, correct?

Evangelos Simoudis: They are still losing money. Earlier this week, they raised $16 billion, with $13 billion coming from Alphabet. This suggests they need capital to accelerate expansion, both domestically and internationally. They are already testing in London and Tokyo. This is a competition not just with Uber, but with Chinese companies that are very active in the Middle East, Southeast Asia, and Europe.

Ben Lorica: Is this a “blitzscaling” strategy? Waymo might lose money in San Francisco, but as they drive more, the service improves, and they expand rapidly to other locations to create a flywheel effect—much like Uber did in its early days.

Evangelos Simoudis: I think the difference between “AV 1.0” and “AV 2.0” is that Waymo has realized their model is now robust enough that moving to a new city no longer requires the same level of extensive testing and high-definition mapping.

Ben Lorica: So, when Waymo launches in a new city now, there is less “reconnaissance” required?

Evangelos Simoudis: They still do some mapping, but not to the extent they used to. The same applies to Tesla. FSD versions 14.4 and 14.5 have improved dramatically due to the sheer volume of data collected and the sophistication of their AI models. That is the big leap from 1.0 to 2.0.

Ben Lorica: However, companies like Waymo and Zoox use multiple sensors, including Lidar, which provides a richer data stream. This allows them to build a more comprehensive “world model” that functions in rain, fog, and other difficult conditions. If you only use cameras and computer vision…

Evangelos Simoudis: And radar. The big question in the industry is whether Lidar provides enough additional safety to justify the cost. For Tesla, using fewer types of sensors reduces the cost per vehicle and the complexity of the “stack.” That impacts the bottom line.

Ben Lorica: Does multimodal data give you an edge in poor conditions? Elon Musk’s original objection to Lidar was its high cost, but those prices have dropped precipitously.

Evangelos Simoudis: They have dropped, but adding data streams also increases computational costs. I’ve said publicly that the extra safety and peace of mind Lidar provides to passengers is important, but it isn’t a panacea. We saw an accident in Santa Monica recently where a Waymo vehicle struck a child who ran into the road. Waymo maintains the vehicle stopped much faster than a human could have, but it proves that even with Lidar, no system is 100% accident-proof. This brings us back to the “trolley problem” conundrum.

Ben Lorica: Before we move on, let me pin you down on the economics. Are you bullish?

Evangelos Simoudis: I would say we are improving. We know which levers to pull regarding the cost of the technology stacks and the vehicles themselves. If Tesla succeeds, it will significantly disrupt the economics. I’m optimistic. As people become more comfortable with longer rides—like trips to the airport, which Uber says account for 15% of rides—the model becomes more viable. You can now take a Waymo to SFO or San Jose via the freeway, which is a major step forward.

Ben Lorica: To wrap up this topic, the three main players in the U.S. are Waymo, Tesla, and Zoox.

Evangelos Simoudis: Waymo is the only one truly deployed in multiple cities. Tesla is in Austin and running pilots in San Francisco. Zoox is in a very limited area in Las Vegas and testing in San Francisco. Today, Waymo is the only real multi-city game in town.

Ben Lorica: Moving to our next topic, I recently wrote an article titled “The Data Center Rebellion is Here.” I use the term “rebellion” because while we in San Francisco live and breathe AI, we often think of data centers as an abstract concept.

My in-laws live in Sioux Falls, South Dakota, and I noticed local opposition to a potential data center there. When I researched the topic, I found that many municipalities are pushing back against these developments for three main reasons: electricity, noise, and water.

Interestingly, a decentralized network of local groups is forming to share lessons on how to stand up to developers. The era of “backroom deals,” where data centers are quietly added to a local board agenda, seems to be ending. Municipalities are also growing wary of tax subsidies. Developers used to promise jobs in exchange for tax breaks—similar to how sports stadiums were pitched—but cities have realized these projects rarely pay for themselves.

Evangelos Simoudis: We’re seeing a shift here similar to the one in autonomous vehicles. When Bitcoin mining was at its peak, we saw similar complaints, but Bitcoin never reached the scale of AI. The capital investments now are staggering: Amazon recently announced a $200 billion investment plan, Meta is at $75 billion, and Alphabet is somewhere in between. These numbers make the local problems even larger.

Ben Lorica: Exactly. People have realized that the promise of “lots of jobs” is a myth. Once a data center is built, thousands of servers are maintained by maybe 20 people.

Evangelos Simoudis: I often talk about “dark spaces”—dark assembly lines, dark industrial kitchens, and dark warehouses. Data centers are the ultimate dark spaces. They require minuscule staffing relative to their physical size and power consumption. Communities and even countries need to understand this before they try to entice companies like Microsoft or Amazon to build on their land.

Ben Lorica: Some countries think hosting a data center makes them part of the AI revolution.

Evangelos Simoudis: In some ways it does, but they may also be taken advantage of. This opposition isn’t just a U.S. phenomenon.

Ben Lorica: No, people everywhere are realizing the consequences. Utility prices go up, and water rights become a major issue. AI-specific data centers are much denser than “vanilla” enterprise data centers. They require significantly more power, and their GPUs often require water cooling rather than air cooling.

This is a problem in places already facing water scarcity due to climate change or industrial agriculture. Athens recently reported they have only one year of water left; Johannesburg is in a similar situation. In my article, I mentioned Florida, where the aquifer is being depleted much faster than projected. Adding AI data centers on top of that only exacerbates the crisis.

Evangelos Simoudis: There is no free lunch. You either use water cooling, which leads to shortages, or air cooling, which strains the electrical grid. Then there is the looming question of overcapacity. I’ve cited utility analysts who suggest that grid operators are adding capacity so aggressively that we might end up with “ghost infrastructure,” much like we saw after the Bitcoin boom.

Ben Lorica: If you were advising a local municipality, what would you say?

Evangelos Simoudis: You must clearly understand what you are giving up versus what you are gaining. Transparency is essential. Citizens need to know what they are getting into. In the Bay Area, you’d never convince the residents of Woodside or Palo Alto to put a data center in their backyard.

We also need to consider the “half-life” of AI chips. If we have to replace and upgrade these chips every two or three years, what does that mean for energy needs and environmental impact? It calls for a broader conversation about the pace of construction. It’s not just about whether we can supply the power—Google, for instance, is looking into small nuclear reactors.

Ben Lorica: As some utility analysts have noted, we likely have enough power, and chips are becoming more energy-efficient. Some of the current dire models might be off-base.

Evangelos Simoudis: The question is how quickly providers will replace old chips with efficient ones and how that supply will meet demand. We haven’t reached an equilibrium yet.

Ben Lorica: One thing that surprised me in my research was the issue of noise pollution. I assumed there was technology to muffle it, but apparently, it’s a major problem for residents.

Evangelos Simoudis: When you have massive fans and auxiliary mechanical generators running constantly, it’s very difficult to control the noise.

Ben Lorica: Finally, there is the geopolitical race with China. There is a psychological pressure in the U.S. to keep building to match China’s command-and-control economy, which can mobilize infrastructure and electricity more quickly.

Evangelos Simoudis: The competition—or “war”—is real. We are also seeing the European Union, the Middle East, and Southeast Asia waking up to this. Singapore is thinking very seriously about how to compete, and Malaysia is building data centers rapidly.

The issue of “sovereign AI” isn’t going away. Countries want to ensure they have the hardware, infrastructure, and power to be self-sufficient. These are massive questions that our firm is dealing with, and we’ll see much more of this throughout the year.

Ben Lorica: We’ll catch up with you again next month. Listeners can visit thedataexchange.media for links to the articles we discussed.