No PriorsChasing Trillion-Dollar Companies, Founder Ambition, Token Budgets, & Regulatory Capture
CHAPTERS
- 0:00 – 0:31
Nuclear energy as a cautionary tale for AI safety tradeoffs
Elad uses France’s nuclear build-out versus U.S. stagnation to argue that safety-first politics can block abundant, beneficial technology. He frames this as a direct analogue for AI: where society sets the needle between risk and progress will create very different futures.
- •France’s high nuclear share vs. U.S. limited build-out as evidence of differing risk tolerance
- •Safety lobbying and regulation can have long-term negative externalities (clean energy delayed)
- •AI governance will require choosing a point on the risk–reward spectrum
- •Different policy choices create materially different societal outcomes
- 0:31 – 1:48
Show setup: what this episode will cover (and a brief program plug)
Sarah previews the main themes—risk management, RSI timelines, trillion-dollar company formation, and regulatory capture—then shares details about Embed, Conviction’s grant program. The conversation then transitions into the main discussion between Sarah and Elad.
- •Episode topics: risk management, RSI/ASI, trillion-dollar companies, regulatory capture
- •Embed grant program: funding + compute/services + founder network
- •Examples of prior cohorts to signal program fit and ambition
- •Transition from announcements into the dialogue
- 1:48 – 3:12
How many new trillion-dollar companies can AI really create—soon?
Elad argues the last five years were an unusually compressed period where multiple firms vaulted to massive valuations, which may distort expectations. He believes more large outcomes are likely, but far fewer true trillion-dollar companies will emerge in the next 3–5 years than investors assume.
- •Recent ‘zero-to-trillion’ leaps are historically unprecedented in speed
- •Typical trillion-dollar arcs take 15–20 years; AI created a rare valuation inflection
- •Exciting sectors (robotics, materials) may become huge—just not that fast
- •Distinguishes near-term probability (3–5 years) from long-term inevitability (20 years)
- 3:12 – 4:42
Tech waves as punctuated equilibrium: bursts, consolidation, then the next jump
Elad frames technology progress as cyclical bursts followed by consolidation, similar to punctuated equilibrium in evolution. AI is in a major burst now, but future step-changes will likely come from new breakthroughs rather than continuous proliferation of new giants.
- •Waves: internet → social → SaaS/cloud/security → crypto; AI as the current major wave
- •After a burst, winners consolidate; fewer new category-defining entrants appear
- •Another breakthrough could trigger the next startup explosion
- •Key question: what comes after current consolidators form?
- 4:42 – 7:14
TAM vs. revenue reality: what it takes to justify a trillion-dollar outcome
Sarah pushes that investors often underestimate AI-driven market expansion (e.g., outcome-based pricing). Elad agrees markets can be bigger than legacy proxies, but emphasizes that trillion-dollar valuations require extraordinary revenue scale with strong margins—an even rarer bar than “big TAM.”
- •Investors may misread AI apps by using old per-seat TAM models
- •Outcome-based pricing and usage-based economics can expand markets dramatically
- •Trillion-dollar companies typically need ~$50–$100B in high-margin revenue
- •Many opportunities can be $100B companies; far fewer can reach $1T soon
- 7:14 – 8:05
Market size vs. speed: why physical-world businesses scale more slowly
They separate the magnitude of a market from the velocity at which a company can capture it. Elad argues investors often conflate the two, especially for energy, robotics, or other physical-footprint businesses where ramp rates and deployment constraints limit near-term revenue scale.
- •A market can be enormous while remaining slow to penetrate
- •Physical deployment constraints reduce speed vs. purely digital markets
- •In the near term, inference is a clearer path to very large revenue
- •Investing ‘as if speed is there’ can misprice timelines and outcomes
- 8:05 – 10:22
Founder ambition in the shadow of ‘neo-labs’: flight to niches and hardware
Elad worries top founders are becoming less ambitious due to fear that frontier labs will encroach on large verticals. He observes a trend toward safer, narrower plays (or hardware ‘moats’) rather than head-on attempts to win big markets through product and distribution.
- •Perceived lab competition can deter startups from pursuing large categories
- •Trend toward niches, “American dynamism,” and hardware as avoidance strategies
- •Elad believes some markets are still winnable for startups vs. labs
- •Sarah agrees she’s seeing under-ambition as a recurring founder failure mode
- 10:22 – 14:04
When founders should sell: structured exit reviews and the time-cost of staying
Elad proposes a recurring, pre-scheduled board-level conversation to evaluate an exit rationally, given how quickly AI-era assumptions change. Both emphasize that avoiding the topic out of pride is irrational, and that the true cost is often founder time spent in stagnating, overcapitalized companies.
- •Some companies should ‘never sell’ (category-defining frontier leaders)
- •AI cycles compress time—facts change fast; revisit exit logic more frequently
- •Use a non-emotional, pre-planned board review cadence for exit decisions
- •Opportunity cost is founder time; secondary sales may not solve core issues
- 14:04 – 17:55
Financing, narrative risk, and perceived scale: capital may be abundant but not rational
Elad predicts capital availability will increase as big AI wins return funds to VCs, but Sarah cautions that private markets can remain wrong for long stretches. They discuss how financing structures must match thesis horizons and how “perceived scale” can distort both funding and strategic choices.
- •More venture capital may chase the next mega-outcomes as returns recycle
- •Valuation inflation can coexist with weak fundamentals; narratives can persist
- •Founders must avoid ‘margin-call-like’ financing fragility via structure and timing
- •Perceived scale influences who gets funded and at what terms
- 17:55 – 21:49
RSI/ASI timelines and the human psychology of ‘18 months away’
Elad describes a manic lab energy around code completion and early RSI, with people acting as if their productive window is short. Sarah finds the mindset psychologically destabilizing and notes RSI has been “18 months away” repeatedly; they explore second-order effects like burnout and life deferral.
- •Belief: code as ‘solved’ soon → light RSI thereafter; implications for work intensity
- •RSI predictions have recurred for years; timelines remain uncertain
- •Emotional/behavioral impacts: burnout, anxiety, postponing life decisions
- •Constraints may be more about data/compute than pure algorithmic possibility
- 21:49 – 24:14
Compute constraints, research power laws, and managing token budgets (ROIT)
They argue compute scarcity shapes the competitive landscape, reinforcing oligopolies and forcing prioritization. Elad highlights power-law contributions among researchers and introduces a “return on invested tokens” mindset for allocating compute to people and projects—mirroring how companies prioritize scarce engineering time.
- •Compute scarcity can enforce an oligopoly-like competitive structure
- •A small number of researchers drive outsized progress; hiring bars rise due to compute costs
- •‘ROIT’ framing: allocate token budgets where marginal impact is highest
- •Enterprises will move from “everyone try AI” to measured spend and prioritization
- 24:14 – 28:12
Where displaced talent goes: diffusion of AI capability into the broader economy
Elad argues even if some engineers or researchers are underutilized in top labs or big tech, many traditional enterprises will eagerly absorb them. Sarah suggests researchers without large compute allocations may find higher-leverage impact in bottlenecked domains like bio, supply chain, and real-world deployment.
- •Token allocation limits mean not everyone can be maximally productive inside top labs
- •Traditional enterprises can productively absorb strong engineering talent
- •High-impact opportunities outside core labs: bio, supply chain, diffusion into regulated fields
- •Power-law cutoffs don’t imply low quality—just resource-constrained prioritization
- 28:12 – 29:03
What could disrupt the current AI trajectory: CapEx pullbacks, architectures, and regulation
Sarah explores three disruption vectors: capital/CapEx reversals, technical shifts beyond transformers, and regulation that constrains model usage. Elad and Sarah view architecture changes as possible but likely to be quickly copied by major labs; they then pivot into the outsized role of policy and regulatory capture.
- •Potential disruption 1: markets reject CapEx/debt profiles, reducing build-out
- •Potential disruption 2: new architectures; hard to outperform transformers at scale
- •Likely outcome: breakthroughs get copied; labs leverage compute advantages
- •Potential disruption 3: regulation restricting use/open-source, slowing progress
- 29:03 – 34:42
Regulatory capture and ecosystem migration: California policy as a forcing function
They discuss how taxation and regulatory uncertainty could push founders and talent to relocate, reshaping innovation hubs. Elad predicts broader downstream effects (including potential future ‘exit tax’ talk), while Sarah points to Texas as an example of regulatory-driven ecosystem growth in energy and hardware.
- •California ‘billionaire tax’ concerns: forced asset sales, compliance complexity, chilling effects
- •Potential follow-on policies (e.g., exit tax) amplify migration incentives
- •Ecosystems self-assemble around talent density; regulation can redirect clusters
- •Texas emerges as a beneficiary hub for energy and hardware experimentation
- 34:42 – 37:55
Safety vs. progress: nuclear, pharma, and how AI rules could entrench incumbents
Elad argues one-sided safety regulation can slow beneficial progress and enable regulatory capture, citing pharma and nuclear as precedents. He warns that heavy compliance burdens could advantage large labs—especially if they can continue advancing internally while regulation slows everyone else—forcing society to make explicit risk–reward choices.
- •Regulatory capture can raise costs, slow iteration, and entrench incumbents
- •Risk-only frameworks (vs. risk–benefit) bias toward stagnation
- •Heavy AI safety burdens could create competitive moats for labs advancing internally
- •Society must choose tradeoffs: e.g., minor risks vs. major gains in health/productivity
- 37:55 – 39:29
Optimism for AI’s upside and closing call to resist overregulation
Elad closes with areas he’s optimistic about—productivity, education, healthcare, mobility, and quality-of-life improvements—while urging balanced safeguards. Sarah ends with a “call to arms” against regulatory capture and standard show sign-offs.
- •AI upside: productivity, education, healthcare, self-driving, elder care
- •Need safeguards, but historical precedent shows overregulation can stall progress
- •Tech’s impact has been accelerated by light regulation relative to other industries
- •Wrap-up: follow/subscribe and find transcripts online
