$UBER Stock is one of my core positions in my portfolio.
In this Deep Dive I will be going over my thesis on $UBER from its Moat, Financials, Management and Growth Runway.
🧵In this thread I will break down why I hold uber as a core position
In this Deep Dive I will be going over my thesis on $UBER from its Moat, Financials, Management and Growth Runway.
🧵In this thread I will break down why I hold uber as a core position
1⃣ The Financials:
FY 2025 Results (Latest)
Revenue $49.61 Billion (+18% YoY)
Gross Margin: 34.15%
Net Income: 16.64 Billion (Inflated/onetime tax benefit)
Free Cash Flow: 8.6 Billion
Cash on Hand: 9.5 Billion
Debt: 10.6 Billion
FY 2025 Results (Latest)
Revenue $49.61 Billion (+18% YoY)
Gross Margin: 34.15%
Net Income: 16.64 Billion (Inflated/onetime tax benefit)
Free Cash Flow: 8.6 Billion
Cash on Hand: 9.5 Billion
Debt: 10.6 Billion
2⃣ Moat
Layer 1. H3 Hexagonal Indexing/ Data Moat
The Concept: Imagine you are playing hide and seek in a giant park. Usually, you’d have to look at every single person in the park to find people which takes a long time and makes your brain tired. What Uber does is they’ve covered the entire world in giant, invisible hexagon stickers (hexagons). Instead of searching the whole park, Uber just asks: "Which sticker are you standing on?" If you are on sticker #5, it only looks for cars on sticker #5 or the six stickers right next to it. It’s like having a secret code for every spot-on Earth, so Uber never has to search it just "knows" exactly where to look instantly. This is how they are so efficient at hailing rides for customers.
Uber’s foundational technical moat is H3, a hexagonal hierarchical spatial index. Traditional maps use latitude/longitude, which requires complex spherical trigonometry computationally difficult math to calculate distances.
Each H3 hexagon is represented by a 64-bit integer. To "zoom out" from a street-level hexagon to a city-level view, the system simply truncates the bits (a bit-shift).
H3 uses a hierarchical structure where each parent hexagon contains seven child hexagons at the next resolution level. This allows Uber to aggregate data (like demand or weather) at 16 different scales (Resolution 0 to 15) without re-calculating the underlying data points.
This is why Uber can calculate surge pricing for every hexagon in a city every few seconds. Competitors using traditional latitude/longitude math often face "latency lag," where their prices are based on data that is already 2 minutes old.
If you are on "Hexagon A," Uber doesn't have to look through the whole book to find you a car. It only looks at "Hexagon A" and the stickers touching it.
If Uber wants to see the whole city, it doesn't need new stickers; it just looks at the "Big Stickers" that the little ones are sitting on. It’s like a puzzle that is already put together.
Uber processed 3.5 billion trips with near-zero matching latency. A competitor using standard GPS-radius math would see their servers melt under that much data. Uber has turned global geography into a simple, high-speed lookup table.
Most legacy taxi apps and smaller regional players (like Bolt in Europe or DiDi in its earlier stages) relied on traditional Latitude and Longitude point-matching or Geohashing.
The Problem: If a rider is at 40.7128, -74.0060, the server has to search for every driver's coordinate and calculate the distance using the Haversine formula.
Doing this math for 100,000 people simultaneously causes "Search Latency." This is why, on smaller apps, you often see a "Searching for drivers" spinner for 10–30 seconds. Uber’s H3 system eliminates this wait because it doesn't do "math" it does a "lookup."
The proof of Ubers moat is that competitors switched from their systems to their H3 system
DoorDash explicitly moved away from its old system to H3. They realized that in dense cities like NYC, their old "Coordinate-to-Store" matching was causing their servers to crash during dinner rushes.
Even though DoorDash uses H3, they are 5+ years behind Uber in data "training." Uber has 15 years of "Hexagon-level data" (knowing how traffic flows specifically in Hexagon #882 on a rainy Tuesday). DoorDash has the tool, but Uber has the History.
Layer 2. The Cross-Platform Flywheel
One of Ubers most valuable economic weapons is their ability to acquire customers for free.
Zero-Cost Acquisition: Most apps pay Google or Meta $15–$30 to find one new customer. Likely 20-30% of new Uber Eats users in 2025 were acquired directly from the Mobility (ride) app.
The Uber One Lock-in: Uber’s membership program has hit 36 million members as of late 2025. These members are the "whales" of the ecosystem: they spend 3x more than non-members and have a 35% higher retention rate.
The Economic Moat: Once a user pays for Uber One, they are "psychologically locked." If they want a pizza, they won't even check DoorDash because they've already paid for "free" delivery on Uber. This creates a "sunk cost" moat that makes it nearly impossible for competitors who just do rides or only food to compete.
Cross Seling:
Approximately 20–25% of Uber’s consumers are now "multi-platform," meaning they use both Mobility (Rides) and Delivery (Eats).
Data from Q3 2025 shows that these cross-platform customers spend 3x more than customers who only use one service.
Uber's AI now uses "Intent-Based Cross-Selling." For example, if you book a ride to work at 8 AM, the app automatically offers you a Starbucks order to be waiting for you at your destination. This isn't a random ad; it's a contextual utility that drives conversion rates 10% higher than standard promotions.
Cross Selling to drivers:
Because Uber already owns the "Rider" relationship, they can acquire a "Delivery" customer for $0. Competitors like DoorDash have to spend an estimated $15–$30 in marketing to acquire that same customer. This "CAC Advantage" is a moat that starves competitors of high-intent traffic.
Layer 3. AV Orchestrator
Uber wants to become the Marketplace for AV's
The Partnership Moat: In October 2025, Uber partnered with Stellantis and NVIDIA to deploy 5,000 Level 4 vehicles using the NVIDIA DRIVE AGX Hyperion 10 architecture. Uber isn't building the cars; it's building the software marketplace they run on. This is part of Uber's broader goal to scale a global fleet to 100,000 autonomous vehicles starting in 2027.
The "Data Factory": Uber and NVIDIA are currently running a "Robotaxi Data Factory" powered by the NVIDIA Cosmos platform. Uber is feeding 3 million+ hours of human driving "edge cases" how a human driver handles a sudden parade or a broken fire hydrant into the AI.
The Hybrid Advantage: Pure AV companies like Waymo hit a wall during demand spikes or bad weather. Uber’s moat is its Hybrid Network: if a robotaxi gets stuck in a snowstorm, the app instantly dispatches a human driver. No standalone robotaxi company can match Uber’s 100% reliability.
Layer 4. Advertising
Although ride-hailing operates on slim margins (around a 30% take rate), advertising delivers hefty profits (exceeding 70% gross margins). During 2025, Uber's ad business reached a $1.5 billion annual run rate.
Uber channels these lucrative ad revenues to subsidize discounts for users. Essentially, Restaurants shell out for promotions on Uber, and the company redirects those funds to knock a few dollars off your trip back home.
Competitors like Lyft, missing a robust delivery and advertising operation, struggle to offer Uber's cut-rate fares without incurring losses. In effect, Uber imposes a "tax" on restaurants to accumulate resources that fortify its control over the transportation landscape.
Layer 5. The Last 100 Meters
Uber’s proprietary machine learning model, DeepETA, processes up to 500,000 requests per second. It uses an encoder-decoder architecture with self-attention to predict arrival times with millisecond precision, accounting for specific "drop-off" vs. "pick-up" behaviors.
After 15 years and 40+ billion trips, Uber has the world’s most precise map of Points of Interest (POI). They know exactly which "hidden" airport pillar or stadium side-entrance is the most efficient.
By eliminating just 45 seconds off every pickup through better data, Uber effectively increases its "fleet capacity" by 10% without adding a single new car. Competitors without this 15-year data history can't duplicate this.
Layer 6. Regulatory
Uber has spent over a decade fighting "Worker Classification" (Independent Contractor vs. Employee) legal battles. In late 2024, the California Supreme Court upheld Prop 22, solidifying Uber’s model.
Any new startup trying to start a company today would face "misclassification" lawsuits and large insurance requirements on Day 1. Uber has the legal experience and lobbying infrastructure to work with these laws in 70+ countries.
Uber has 9.5 Billion in cash and over 8 billion in free cash flow. Uber can simply "out-wait" any legal or competitive threat. They have turned legal compliance into a fixed cost that smaller rivals simply cannot afford.
Network Effects
One thing I didn't talk about much was network effects meaning riders attract more drivers and riders attract more drivers. This is a Powerful flywheel, but I figure it is obvious and talked about a lot.
Layer 1. H3 Hexagonal Indexing/ Data Moat
The Concept: Imagine you are playing hide and seek in a giant park. Usually, you’d have to look at every single person in the park to find people which takes a long time and makes your brain tired. What Uber does is they’ve covered the entire world in giant, invisible hexagon stickers (hexagons). Instead of searching the whole park, Uber just asks: "Which sticker are you standing on?" If you are on sticker #5, it only looks for cars on sticker #5 or the six stickers right next to it. It’s like having a secret code for every spot-on Earth, so Uber never has to search it just "knows" exactly where to look instantly. This is how they are so efficient at hailing rides for customers.
Uber’s foundational technical moat is H3, a hexagonal hierarchical spatial index. Traditional maps use latitude/longitude, which requires complex spherical trigonometry computationally difficult math to calculate distances.
Each H3 hexagon is represented by a 64-bit integer. To "zoom out" from a street-level hexagon to a city-level view, the system simply truncates the bits (a bit-shift).
H3 uses a hierarchical structure where each parent hexagon contains seven child hexagons at the next resolution level. This allows Uber to aggregate data (like demand or weather) at 16 different scales (Resolution 0 to 15) without re-calculating the underlying data points.
This is why Uber can calculate surge pricing for every hexagon in a city every few seconds. Competitors using traditional latitude/longitude math often face "latency lag," where their prices are based on data that is already 2 minutes old.
If you are on "Hexagon A," Uber doesn't have to look through the whole book to find you a car. It only looks at "Hexagon A" and the stickers touching it.
If Uber wants to see the whole city, it doesn't need new stickers; it just looks at the "Big Stickers" that the little ones are sitting on. It’s like a puzzle that is already put together.
Uber processed 3.5 billion trips with near-zero matching latency. A competitor using standard GPS-radius math would see their servers melt under that much data. Uber has turned global geography into a simple, high-speed lookup table.
Most legacy taxi apps and smaller regional players (like Bolt in Europe or DiDi in its earlier stages) relied on traditional Latitude and Longitude point-matching or Geohashing.
The Problem: If a rider is at 40.7128, -74.0060, the server has to search for every driver's coordinate and calculate the distance using the Haversine formula.
Doing this math for 100,000 people simultaneously causes "Search Latency." This is why, on smaller apps, you often see a "Searching for drivers" spinner for 10–30 seconds. Uber’s H3 system eliminates this wait because it doesn't do "math" it does a "lookup."
The proof of Ubers moat is that competitors switched from their systems to their H3 system
DoorDash explicitly moved away from its old system to H3. They realized that in dense cities like NYC, their old "Coordinate-to-Store" matching was causing their servers to crash during dinner rushes.
Even though DoorDash uses H3, they are 5+ years behind Uber in data "training." Uber has 15 years of "Hexagon-level data" (knowing how traffic flows specifically in Hexagon #882 on a rainy Tuesday). DoorDash has the tool, but Uber has the History.
Layer 2. The Cross-Platform Flywheel
One of Ubers most valuable economic weapons is their ability to acquire customers for free.
Zero-Cost Acquisition: Most apps pay Google or Meta $15–$30 to find one new customer. Likely 20-30% of new Uber Eats users in 2025 were acquired directly from the Mobility (ride) app.
The Uber One Lock-in: Uber’s membership program has hit 36 million members as of late 2025. These members are the "whales" of the ecosystem: they spend 3x more than non-members and have a 35% higher retention rate.
The Economic Moat: Once a user pays for Uber One, they are "psychologically locked." If they want a pizza, they won't even check DoorDash because they've already paid for "free" delivery on Uber. This creates a "sunk cost" moat that makes it nearly impossible for competitors who just do rides or only food to compete.
Cross Seling:
Approximately 20–25% of Uber’s consumers are now "multi-platform," meaning they use both Mobility (Rides) and Delivery (Eats).
Data from Q3 2025 shows that these cross-platform customers spend 3x more than customers who only use one service.
Uber's AI now uses "Intent-Based Cross-Selling." For example, if you book a ride to work at 8 AM, the app automatically offers you a Starbucks order to be waiting for you at your destination. This isn't a random ad; it's a contextual utility that drives conversion rates 10% higher than standard promotions.
Cross Selling to drivers:
Because Uber already owns the "Rider" relationship, they can acquire a "Delivery" customer for $0. Competitors like DoorDash have to spend an estimated $15–$30 in marketing to acquire that same customer. This "CAC Advantage" is a moat that starves competitors of high-intent traffic.
Layer 3. AV Orchestrator
Uber wants to become the Marketplace for AV's
The Partnership Moat: In October 2025, Uber partnered with Stellantis and NVIDIA to deploy 5,000 Level 4 vehicles using the NVIDIA DRIVE AGX Hyperion 10 architecture. Uber isn't building the cars; it's building the software marketplace they run on. This is part of Uber's broader goal to scale a global fleet to 100,000 autonomous vehicles starting in 2027.
The "Data Factory": Uber and NVIDIA are currently running a "Robotaxi Data Factory" powered by the NVIDIA Cosmos platform. Uber is feeding 3 million+ hours of human driving "edge cases" how a human driver handles a sudden parade or a broken fire hydrant into the AI.
The Hybrid Advantage: Pure AV companies like Waymo hit a wall during demand spikes or bad weather. Uber’s moat is its Hybrid Network: if a robotaxi gets stuck in a snowstorm, the app instantly dispatches a human driver. No standalone robotaxi company can match Uber’s 100% reliability.
Layer 4. Advertising
Although ride-hailing operates on slim margins (around a 30% take rate), advertising delivers hefty profits (exceeding 70% gross margins). During 2025, Uber's ad business reached a $1.5 billion annual run rate.
Uber channels these lucrative ad revenues to subsidize discounts for users. Essentially, Restaurants shell out for promotions on Uber, and the company redirects those funds to knock a few dollars off your trip back home.
Competitors like Lyft, missing a robust delivery and advertising operation, struggle to offer Uber's cut-rate fares without incurring losses. In effect, Uber imposes a "tax" on restaurants to accumulate resources that fortify its control over the transportation landscape.
Layer 5. The Last 100 Meters
Uber’s proprietary machine learning model, DeepETA, processes up to 500,000 requests per second. It uses an encoder-decoder architecture with self-attention to predict arrival times with millisecond precision, accounting for specific "drop-off" vs. "pick-up" behaviors.
After 15 years and 40+ billion trips, Uber has the world’s most precise map of Points of Interest (POI). They know exactly which "hidden" airport pillar or stadium side-entrance is the most efficient.
By eliminating just 45 seconds off every pickup through better data, Uber effectively increases its "fleet capacity" by 10% without adding a single new car. Competitors without this 15-year data history can't duplicate this.
Layer 6. Regulatory
Uber has spent over a decade fighting "Worker Classification" (Independent Contractor vs. Employee) legal battles. In late 2024, the California Supreme Court upheld Prop 22, solidifying Uber’s model.
Any new startup trying to start a company today would face "misclassification" lawsuits and large insurance requirements on Day 1. Uber has the legal experience and lobbying infrastructure to work with these laws in 70+ countries.
Uber has 9.5 Billion in cash and over 8 billion in free cash flow. Uber can simply "out-wait" any legal or competitive threat. They have turned legal compliance into a fixed cost that smaller rivals simply cannot afford.
Network Effects
One thing I didn't talk about much was network effects meaning riders attract more drivers and riders attract more drivers. This is a Powerful flywheel, but I figure it is obvious and talked about a lot.
3⃣ Management
Uber manages their business for "entire lifetime experience." This means they will intentionally take a short-term hit on a ride's profit if it keeps a user in the Uber One ecosystem. This is a quality that I see in many great companies like Amazon and MELI.
It is a sign that management values long term dominances of their market over short term numbers to appease wall street.
The CEO (Dara) is required to hold 10x his base salary in Uber stock. This ensures he is personally "punished" or "rewarded" exactly like an outside investor.
Dara is one of the largest individual shareholders.
While I don't think Uber has the best management philosophy and policy like a Pro Medicus. I think they have a long-term outlook and are building UBER for decades to come.
Uber manages their business for "entire lifetime experience." This means they will intentionally take a short-term hit on a ride's profit if it keeps a user in the Uber One ecosystem. This is a quality that I see in many great companies like Amazon and MELI.
It is a sign that management values long term dominances of their market over short term numbers to appease wall street.
The CEO (Dara) is required to hold 10x his base salary in Uber stock. This ensures he is personally "punished" or "rewarded" exactly like an outside investor.
Dara is one of the largest individual shareholders.
While I don't think Uber has the best management philosophy and policy like a Pro Medicus. I think they have a long-term outlook and are building UBER for decades to come.
4⃣ Growth Runway
As of late 2025, Uber's non-restaurant delivery (Grocery, Alcohol, and Retail) has reached an annualized run rate of $12.5 billion. This is a 25% increase from early 2024.
CEO Dara Khosrowshahi noted in late 2025 that 75% of Uber’s rideshare customers still haven’t ordered grocery or retail through the app. Converting this existing user base is a "zero-cost" growth opportunity.
The Walmart/Amazon Rivalry: By adding 1,000+ partners like Aldi, Best Buy, and The Home Depot in 2025 alone, Uber is positioning itself as the "instant" alternative to traditional e-commerce.
Advertising is also a highly profitable market they are just taping into. Unlike Facebook, Uber knows where you are going (e.g., a stadium or a mall) and what you want to eat. This high-intent data allows them to charge premium rates to advertisers.
AV Orchestration is probably the most important thing and needed for long sustainable growth Uber needs to accomplish. Uber plans to offer robotaxi services in 10+ global markets by the end of 2026. Uber has partnered with Waymo, NVIDIA, WeRide, and Baidu (Apollo Go). In late 2025, Baidu and Uber announced a plan to launch fully autonomous trials in London by early 2026.
As of late 2025, Uber's non-restaurant delivery (Grocery, Alcohol, and Retail) has reached an annualized run rate of $12.5 billion. This is a 25% increase from early 2024.
CEO Dara Khosrowshahi noted in late 2025 that 75% of Uber’s rideshare customers still haven’t ordered grocery or retail through the app. Converting this existing user base is a "zero-cost" growth opportunity.
The Walmart/Amazon Rivalry: By adding 1,000+ partners like Aldi, Best Buy, and The Home Depot in 2025 alone, Uber is positioning itself as the "instant" alternative to traditional e-commerce.
Advertising is also a highly profitable market they are just taping into. Unlike Facebook, Uber knows where you are going (e.g., a stadium or a mall) and what you want to eat. This high-intent data allows them to charge premium rates to advertisers.
AV Orchestration is probably the most important thing and needed for long sustainable growth Uber needs to accomplish. Uber plans to offer robotaxi services in 10+ global markets by the end of 2026. Uber has partnered with Waymo, NVIDIA, WeRide, and Baidu (Apollo Go). In late 2025, Baidu and Uber announced a plan to launch fully autonomous trials in London by early 2026.
5⃣ Summary
In summary Uber has excellent Financials with an underappreciated moat it is not impenetrable but with good execution from management I think they can make the moat much wider than it already is if they become the marketplace for AVs, and so far, their management is doing all the right things to accomplish this.
This is the biggest risk for Uber long term if there are only 1 or 2 players in the AV space like Tesla and Waymo. Uber may slowly become less and less valuable to these companies as they get better with AVs. They may be able to build their own marketplaces for AVs and overtime slowly phase uber out.
If there are more players in this market, then just a few this will benefit Uber greatly and they will become the aggregator of all the AV companies. No one is going to want to have 4-15 apps of all the AV companies, all if those companies will be aggregated in one app the UBER app, but if there are only one or two players customers may just use their native apps and ditch Uber overtime, but Uber's hybrid model and demand control could mitigate it.
I am betting that the former of these options happens and that uber becomes the multi trillion-dollar toll booth of AVs like they are now the toll booth of ride hailing and food delivery.
In summary Uber has excellent Financials with an underappreciated moat it is not impenetrable but with good execution from management I think they can make the moat much wider than it already is if they become the marketplace for AVs, and so far, their management is doing all the right things to accomplish this.
This is the biggest risk for Uber long term if there are only 1 or 2 players in the AV space like Tesla and Waymo. Uber may slowly become less and less valuable to these companies as they get better with AVs. They may be able to build their own marketplaces for AVs and overtime slowly phase uber out.
If there are more players in this market, then just a few this will benefit Uber greatly and they will become the aggregator of all the AV companies. No one is going to want to have 4-15 apps of all the AV companies, all if those companies will be aggregated in one app the UBER app, but if there are only one or two players customers may just use their native apps and ditch Uber overtime, but Uber's hybrid model and demand control could mitigate it.
I am betting that the former of these options happens and that uber becomes the multi trillion-dollar toll booth of AVs like they are now the toll booth of ride hailing and food delivery.
Typo in network effects should be riders attract more drivers and drivers attract more riders.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
