This interview is about spotting real tech turning points and investing in them. Guest Michael Dempsey says GANs (AI that generates images) are making animation explode, and digital celebrities like Hatsune Miku can make more money than human stars because they can perform in multiple places at once. Key picks: SpaceX (drastically cut launch costs), Planet Labs (cheap satellites from phone parts), and Ono Food Co. (robot food truck, he invested). He warns many robot companies promise 90% reliability but customers prefer cheap human labor instead.
Michael Dempsey, General Partner at Compound, discusses the core theme of investing in "bleeding edge technology" in a podcast. He argues that identifying key "inflection points" in technological evolution is the source of investment opportunities, and distinguishes between genuine and false inflect
Michael Dempsey, General Partner at Compound, specializes in applying brand building and direct customer relationships to technology-intensive fields. The main theme of this issue is identifying genuine "inflection points" in technological evolution and capturing investment opportunities from them. Dempsey argues that Generative Adversarial Networks (GANs) are driving the "animation eating the world"—from virtual influencers to digital celebrities, the creation and distribution of IP are undergoing fundamental transformation, with an economic model superior to that of human entertainers.
Michael Dempsey argues that the core of distinguishing real from false inflection points lies in assessing whether a technology possesses "production-level scalability," rather than merely being a scientific project or a novelty experience.
Dempsey defines an inflection point as an "unlock"—making many previously impossible things possible. He reviews two decades of history: internet penetration, the launch of AWS in 2006, the App Store's debut, and the commercialization of 4G/5G—these are real inflection points. In contrast, the 2017 cryptocurrency craze is a classic "false signal": at the time, everyone was saying "it's now," but actual transactions still took over 30 minutes to complete a single Bitcoin transfer, "that can't be the future."
VR is another case of misjudgment. After Facebook acquired Oculus for over $1 billion, the market believed VR was ready for mass adoption. However, Dempsey points out that people were "too enamored with the novelty" and failed to seriously examine processing power, computational demands, and cost barriers. "If you used VR early on, the hallmark experience was getting motion sickness every time you tried a new demo—because this is a hardcore engineering problem." From the outset, Zuckerberg positioned it as a "decades-long bet," not a near-term scalable business.
Key judgment: Acquisitions by large tech companies (e.g., GM acquiring Cruise for over $1 billion) are not inflection point signals in themselves, but rather "FOMO driven by existential threats." Dempsey emphasizes the need to understand the logic behind these acquisitions—"If Tesla solves autonomous driving first, that would be an existential threat to the entire automotive industry"—but the technology itself remains far from production-level readiness.
Dempsey argues that "Animation is Eating the World," driven primarily by GANs (Generative Adversarial Networks), which drastically reduce the cost and scale up the production of animated content, while the economic model of IP outperforms that of human entertainers.
Technological Breakthrough of GANs: Proposed by Ian Goodfellow, GANs involve two models competing against each other (one generating, one discriminating) until equilibrium is reached. Initially used to generate realistic images, researchers at the University of Tokyo applied them to anime face generation, which inspired Dempsey's team to invest in a "machine learning-first animation studio."
Evolution of Animation Technology: From hand-drawing every frame → cel animation (fixed backgrounds with superimposed moving images) → Disney's multiplane camera → digital workflows → real-time rendering (Unreal Engine, Blender). Dempsey summarizes: "The arc of animation is from 'creating beauty' to 'scaling beauty' to 'doing more beautiful things with fewer people.'"
Demand-Side Explosion: Platforms such as Netflix, Amazon, and Hulu are "investing heavily in animation" because animation has extremely high retention rates within households, and IP can extend into adjacent areas like gaming. During the pandemic in 2020, animation required no on-site filming, marking "a massive inflection point."
Economic Model Advantages of Digital Celebrities: Take Hatsune Miku as an example—she is a UGC-style digital celebrity based on Vocaloid software, where fans can design her outfits, create songs and dances, and she performs as a holographic projection. Key data: Over $1 billion in merchandise sales. Dempsey highlights the economic advantage: "Nothing prevents Hatsune Miku from performing in Tokyo at 7 PM, Los Angeles at 8 PM, and Boston at 9 PM—or appearing everywhere simultaneously." In contrast, "humans are beautiful, volatile beings"—difficult to scale.
Another Case: The digital celebrity Michaela (created by Brud). She can appear simultaneously in New York and Los Angeles, sing, walk the runway, and adjust her image as she ages. However, Dempsey also points out the "uncanny valley effect"—she looks human but not quite, which erodes trust. Early companies concealed that she was not a real person, sparking controversy.
Core Competency of Creators: Dempsey believes that the most successful creators of digital celebrities need to understand "which markets are severely underserved." His team found that the 13–17-year-old female demographic lacks content supply—"this is an underserved market."
Dempsey believes the robotics industry has experienced a "false dawn"—many companies promised 90% reliability, but customers were unwilling to pay for 90%. His strategy is to either invest in full-stack robotics companies (absorbing R&D costs internally) or invest in the infrastructure layer.
Historical lesson: Around 2016, the market loudly proclaimed "robots are coming." The R&D spillover from the smartphone wars (Apple, Samsung, etc.) drove sensor costs down sharply, reducing robot prices from $1 million to $100,000. However, the problem was that customers were unwilling to retrofit factories with 90% automation (huge CapEx), and "a robot with five fingers is often the cheapest option—humans." After many companies completed pilot projects, customers refused to renew or scale up.
Dempsey's response: He invested in Ono Food Co. (a robotic food truck)—"both a brand and a robotics company"—using a full-stack approach to "short-circuit" the hassle of dealing with large enterprises. He also invested in Fort Robotics (safety-grade robotics infrastructure) and a stealth "human-in-the-loop" company, helping automation firms achieve 100% reliability at lower cost.
Key judgment: "We like doing hard things. A full-stack robotics company can continuously capture value during the R&D process—from 60% to 80% to 99.99% automation, each step creating more economic value."
Dempsey believes the true inflection point for the space economy has yet to arrive—it will only come when the "Las Vegas principle" (what happens in space stays in space) is realized, i.e., privatized space stations. Computational biology, on the other hand, is "one of the biggest opportunities of the next decade," but investors must be wary of the risk that "tech investors pretend to understand biology."
Space: Dempsey cites the "macroscope" concept from Orbital Insight founder Jimmy Crawford—society understands the value of a microscope (continuous magnification) but has yet to grasp the value of "continuous reduction" (observing the Earth as a whole in a repeatable manner). Planet Labs' nanosatellites and synthetic aperture radar (capable of penetrating clouds and imaging at night) are currently visible applications. However, the true turning point is the "privatized space station"—a veteran of the space industry told Dempsey that, at that point, four of the six seats could be sold to millionaires and two to scientists, using the former to subsidize the latter.
Computational Biology: Dempsey notes that human understanding of biological organisms was "astonishingly limited" a decade ago, but progress is now accelerating at an extremely rapid pace. "If I had to make a single technology bet, this could be the biggest opportunity over the next one or two decades." Key breakthroughs include high-throughput screening (automated experiments) and DeepMind's protein structure prediction (AlphaFold). However, he warns: "Some VCs now say, 'These companies are essentially tech companies—thank you, biotech investors, for the past 30 years of innovation; we'll take it from here'—and buy in at double the valuation. This is a dangerous game."
Lessons from 2020: The rapid development of vaccines proved that "if capital and incentives are sufficiently aligned, scientific problems can be solved quickly." Dempsey believes this should serve as a template for solving other scientific problems in the future.
Dempsey introduces the macro theme "Cyberpunk is now"—dystopian elements from science fiction are manifesting in reality, creating new investment opportunities.
Specific manifestations include: the four big tech CEOs testifying before Congress with the attitude of "we're here, we're dealing with it, but we're not really listening"; state-sponsored cyberattacks (e.g., the SolarWinds incident); information warfare during the 2016 election; and U.S.-China competition. These trends point to several investment directions: privacy-preserving machine learning (how to protect data while enjoying the benefits of ML), adversarial attack defense (the more automated systems, the greater the risk of attack).
Dempsey's investment methodology follows two paths: either enter a category, think about "what kind of business should exist," then find founders to collaborate with; or think about "what kind of future we believe in," then derive investable startups.
| Position | Guest Stance | Key Data |
|---|---|---|
| SpaceX | Bullish (a model of creating its own inflection points) | Significantly reduced launch costs; all R&D has commercial use cases |
| Planet Labs | Bullish (a case of technological innovation) | Uses smartphone sensors to manufacture low-cost nanosatellites |
| Ono Food Co. | Invested (full-stack robotics) | Robotic food truck, highly fault-tolerant and low-cost |
| Fort Robotics | Invested (infrastructure) | Safety-grade robotics infrastructure |
| Brud (Michaela) | Neutral (case study) | Digital celebrity that can appear in multiple locations simultaneously |
| Epic Games (Unreal) | Bullish (core infrastructure) | Fortnite cash cow subsidizes engine development |
| Unity | Neutral (one of the duopoly) | Competes with Unreal in the game engine market |
| DeepMind | Bullish (technological breakthrough) | Solves protein structure prediction using non-traditional ML methods |
| Hatsune Miku | Bullish (UGC-style digital celebrity) | Has sold over $1 billion in merchandise |
| Riot Games (League of Legends) | Recommended for research | A model for IP expansion outward |
1. "A real inflection point makes the previously impossible possible; a false inflection point seduces you with novelty while ignoring production-grade scalability." — Dempsey cites VR and 2017-era cryptocurrencies as false signals, stressing the need to examine processing power, cost thresholds, and engineering difficulty.
2. "Animation is eating the world—the economic model of IP outperforms human artists because digital celebrities can perform in multiple venues simultaneously, never age, and never have mood swings." — Hatsune Miku has sold over $1 billion in merchandise, proving the monetization power of digital IP.
3. "The robot with five fingers is often the cheapest option—humans." — Robotics companies promise 90% reliability, but customers are unwilling to retrofit factories for 90%; full-stack robotics firms "short-circuit" this dilemma by absorbing R&D costs themselves.
4. "The Las Vegas principle: once you have a privatized space station, 'what happens in space stays in space'—that is the key to unlocking the space economy." — Dempsey believes current space investment opportunities remain concentrated at the "Earth-to-space" interface, and the true inflection point has yet to arrive.
5. "Human understanding of biological organisms was astonishingly limited a decade ago, and is now advancing at breakneck speed—this may be the biggest opportunity over the next one to two decades." — But Dempsey warns tech investors not to "pretend to understand biology" and "catch a falling knife" from biotech investors at double the valuation.
6. "Cyberpunk is now—the dystopian elements of science fiction are manifesting in reality, from Big Tech congressional hearings to state-level cyberattacks." — This gives rise to investment directions such as privacy-preserving machine learning and adversarial attack defense.
7. "2020 proved that if capital and incentives are sufficiently aligned, scientific problems can be solved quickly—vaccines compressed from 5 years to under 1 year." — Dempsey believes this should serve as a template for solving other scientific problems in the future.
8. "You can't bring facts to a feelings fight." — Dempsey found, while investing in ML-driven animation tools, that artists would rather spend 3–5 hours manually completing a task than use a 100% accurate automation tool, because creation is a "romantic craft."