"The World's Best AI is Tesla"—Antonio Gracias and Gavin Baker on the Intersection of AI, Defense, and Energy
Valor Equity board member and early Tesla investor Antonio Gracias, and Atreides Management CIO Gavin Baker. From AI to the SaaS collapse, Tesla, humanoid robots, defense, and energy, their conversation was full of insights for investors.
In this article, we summarize the core points from an investor's perspective.
1. The SaaS Collapse—"Evolve or Die"
1-1. HubSpot from $850 to the $200 range
Baker stated the current situation bluntly: "The risk of SaaS is definitely priced into the stock. HubSpot peaked at $850 and is now in the $200 range. This means the market has already made its judgment."
In the era of vibe coding (intuitive coding using AI), even non-engineers can build their own CRM in a few hours. This fundamentally questions the raison d'être of $100,000-a-year SaaS tools like HubSpot.
1-2. "Software CEOs have only 2-3 months to make good decisions"
Baker warned, quoting Maximus Decimus Meridius from Gladiator: "What software company CEOs decide now will echo in eternity. They only have 2-3 months to make good decisions."
Gracias said, "I haven't invested in a pure software company in 6-7 years." Instead, he is focusing on the strategy of acquiring existing private software companies and making them AI-native.He says it's the same idea Larry Ellison used when building Oracle—acquiring PeopleSoft and Siebel, cutting 80% of the staff, and turning the software into AI.
2. Anthropic and AI Model Efficiency
2-1. Token efficiency determines cost and quality
Baker analyzed Anthropic's competitive advantage from the perspective of token efficiency.
"Anthropic is about four times more capital-efficient than OpenAI and has low cash burn despite growing rapidly. And currently, it is the best in terms of the maximum time an AI can work on a single task."
In reasoning models, the higher the token generation, the better the quality of the answer. Anthropic is said to be able to generate answers of the same quality as Google with about half the tokens. Since 1 token equals cost, the difference in token efficiency is directly linked to economic competitiveness.
2-2. A new evaluation axis: The "values" of a model
Gracias added a more philosophical perspective: "Unlike knives or guns, AI models have values. A model embodies the values of its creator. Therefore, we should ask who designed the model and with what optimization function."
If the optimization function is the "pursuit of truth," safety increases. Citing Grok 4.2 as an example, he evaluated a design where multiple agents verify each other's output as "good for humans."
3. Tesla—A new perspective as "the world's most advanced AI"
3-1. Handling the most complex tasks with A4 chips
Gracias' most notable remark was about Tesla.
"In terms of intelligence efficiency per token, the world's best is actually Tesla Autopilot. This is because it operates on highly limited hardware—the A4 chip—without cloud access, and handles the most complex task humans perform at the edge: driving a car."
Large models like GPT are trained with hundreds of megawatts of power. In contrast, the human brain consumes about 20–30W. Tesla's Autopilot performs superhuman judgments with a power consumption of about 50W, close to that of the human brain. They stated that this means, "For the first time in history, we have achieved superhuman AI with power consumption equivalent to the human brain."
3-2. Optimus and the End of the "Specialized vs. General-Purpose" Debate
A major point of contention in the humanoid robot field was the debate over whether they should be general-purpose or specialized. However, Baker stated, "That debate is over."
"Humanoid robots can now learn from videos of human movements. And we have a massive amount of video footage of human movement. With this, it is almost certain that general-purpose humanoids will win."
He also predicted that the competition would become a head-to-head battle between Tesla and Chinese firms.
4. Defense and Energy—The Invisible Risks for the U.S.
4-1. China's Unmanned Systems and the "West's Only Contender"
China's drone swarms are a real military threat. Those beautiful drone shows are actually military exercises and a venue for China to demonstrate its technological prowess to the world.
"If we didn't have Optimus (Tesla), there would be no true competitor in the U.S. or the West against Chinese platforms. That is extremely dangerous."
Valor has welcomed former NATO Supreme Allied Commander and Commander of U.S. European Command, Chris Paul, as a new partner to accelerate its investment strategy in defense technology.
4-2. The Reality of Purchasing 100% of Nuclear Fuel from Russia
The fact shared by Gracias regarding energy issues was shocking.
"It was as surprising as when I learned that the U.S. space program once used Russian engines. And when I looked into nuclear fuel, I found that the U.S. does not enrich any nuclear fuel domestically and is purchasing it from Russia."
To bridge this gap, Valor is investing in companies with domestic nuclear fuel manufacturing capabilities in the U.S.
4-3. The Solution of Space Data Centers
As a unique approach to energy shortages, Baker discussed space data centers.
"A common misconception is the image of a Pentagon-sized building floating in space. In reality, it's a rack. A Starlink V3 satellite consumes 20kW, and an Nvidia Blackwell rack consumes 130kW. If you increase the power consumption of Starlink by five times, you get a space-based computer equivalent to one rack."
If multiple satellite racks are connected by lasers, a data center is created. The speed of lasers traveling through a vacuum is faster than fiber optics, and since it reaches user devices directly via Starlink, it is possible that latency could be lower than that of terrestrial data centers.
5. The Paradox That Power Shortages Prevent a Bubble
Finally, Baker presented a paradoxical perspective: "Power shortages might actually be a good thing."
Citing Carlotta Perez's book "Technological Revolutions and Financial Capital," he pointed out that every technological innovation in history has been accompanied by a bubble. The South Sea Bubble, the railway bubble, and the dot-com bubble—all were cases where over-investment in new technology led to a collapse.
"Power shortages and wafer shortages are making over-investment and over-construction physically impossible. In other words, the AI cycle might be more gradual and long-lasting. That is a good thing for everyone here."
However, he did not say there is "no bubble at all." His view is that the "scale of the bubble and the depth of the collapse" could be limited.
Summary
What their conversation points to in common is the reality that AI has moved beyond the traditional framework of "software" and has deeply penetrated the physical world, energy, security, and hardware efficiency.
Whether SaaS companies will survive, whether Tesla will be re-evaluated as an AI company, and whether the U.S. can become self-reliant in defense, energy, and semiconductors—these are all one big question connected by the "lens of AI."