Three Trillion-Dollar Companies in Five Years, But the Next One Won't Come That Fast
Three companies going from zero to a trillion-dollar valuation in five years is a historical anomaly, not a new normal. What's genuinely scarce isn't a big market, it's the speed to reach $50 billion in revenue within five years.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Three trillion-dollar companies in five years is an anomaly, not a new normal
Elad points out that over the past five years Anthropic, OpenAI and SpaceX went from near zero to trillion-dollar valuations, whereas historically SpaceX starting in the early 2000s and Google from the '90s typically followed a 15-to-20-year arc. He thinks many people now assume another batch of trillion-dollar companies will emerge within three to five years, and that this doesn't hold. He can think of "maybe one more" himself, but refuses to say which. Sarah pushes back from the other side: what she encounters more often is investors lacking imagination, valuing AI companies on old per-seat, per-lawyer TAMs, when coding has already proven that consumption and value can grow 100x.
— Elad GilA trillion-dollar market needs $50 billion in revenue, not a big TAM
Elad reframes the question from "is the TAM big" to "where does a $50 to $100 billion revenue stream for a single company exist." He mentions a blog post he wrote in 2010 about how hard it is to reach a $10 billion valuation, a number that is now just some neolab's seed round. His judgment: plenty of companies can reach $5 billion or $10 billion in revenue, very few can reach a trillion — that's an order-of-magnitude difference. Sarah adds the key distinction — people conflate market size with speed of arrival; what she doubts isn't the size of markets like energy or robotics, but whether physical-goods companies have enough footprint to get there that fast.
— Elad GilThe best founders are building small businesses out of fear of the labs
Sarah observes two things happening at once: in the mid-to-late stage market, investors keep betting on "can reach a trillion" speed, and she doesn't think that speed exists; meanwhile a cohort of very strong founders is turning to niche markets because there's far less fear of neolabs and far less head-on competition. She names Harvey, Open Evidence, Decagon, Sierra and Cognition as having entered large markets that could plausibly appear on a lab's roadmap. Elad pushes this further: what worries him isn't the median founder, it's that the best founders are turning timid — a shift in the trend line, and in his view a negative signal.
— Sarah GuoEvery company has a 12-to-18-month optimal selling window
Elad's framework: a tiny handful of companies (he names Anthropic and OpenAI) should absolutely not sell anytime soon; but most companies in any era should at least consider selling, and there's usually a 12-to-18-month window in which the company is worth the most it will ever be worth. He cites Ben Horowitz's practice — a pre-scheduled annual board meeting whose agenda is to ask, unemotionally, "should we consider an exit in the next six months," driven by process rather than by the founder or the investor. Sarah adds a new question: are you capturing value on the curve of falling costs and rising capability, or are you on the wrong side of it? If the latter, and you have no idea how to get across, you should sell.
— Elad GilFundraising will get easier; your real cost is time
Elad's judgment: because several three-trillion-dollar-scale companies rose in a short period, a lot of VC money is flowing back, funds are getting bigger, and the money has to be deployed, so valuations will keep rising over the next year or two and fundraising will get easier rather than harder. He therefore shifts the decision's center of gravity from "can I raise" to "what is your company's true expected outcome" — don't listen to investors, the media or Twitter, sit down and do the math yourself, including future dilution and the number of years it will take. He stresses that the biggest opportunity cost isn't money but time: the cohort of founders from 2020 and 2021 who are still running companies that don't work five years later missed the entire AI shift. Sarah cautions that private markets can stay irrational for a long time, and fundraising ability is equivalent to avoiding a margin call.
— Elad GilRSI arrives every 18 months, and it's been arriving for five years
Elad describes the restlessness inside the labs: code is basically solved by the end of the year, some kind of light RSI may arrive by the end of next year, models start training large parts of themselves, and early on it's mostly post-training. Sarah offers a counterpoint — her data is that a group of very smart and self-reflective scientists have felt, every 18 months for the past five years, that recursive self-improvement or ASI is 18 months away. She thinks extending from code to training code and data pipelines is easier to believe, but the hard part is how to collect data in unverifiable, more complex domains, and the accessibility of physical compute, which may be more of a limiting factor than algorithmic feasibility. She also mentions people from labs coming to ask her whether they should get married, because they don't know what the world will look like in 18 months.
— Sarah GuoCompute scarcity lets a few dozen researchers capture 80% of the results
Elad proposes a "human power law": in any field, at most a few dozen people drive most of the progress — breast cancer research, subfields of math, subfields of physics, the startup ecosystem — and AI research is no exception. When compute becomes the truly scarce resource, compute starts being allocated differentially to those people, and some labs have slowed researcher hiring unless the candidate clears an extremely high bar — because the bottleneck isn't the researcher's salary, it's the compute allocated to them. From this he draws out the concept of ROIT (return on invested tokens): given a token budget, who should get it and why. He also analogizes to internal tools teams having historically been starved of resources, arguing this explains why "the death of SaaS" is overstated.
— Elad GilCalifornia is pushing the whole ecosystem out
On California's billionaire tax, Elad argues the bill is written quite broadly, and once passed it can be re-enacted and thresholds lowered in future years, and around 2028 there's also discussion of an added exit tax, effectively penalizing those who leave a second time. His judgment is that lawmakers want this outflow to happen, and the two genuinely negative signals from California are this bill and ballot-harvesting-related measures. Sarah says she already knows quite a few people leaving now or planning to leave before the fall. She thinks the key to migration isn't weather but whether you can gather enough smart people doing the same thing — Boston was Silicon Valley's counterpart in the '80s, lost the competition in the early '90s, but biotech is still there today. She names Texas as an ecosystem forming around energy and hardware.
— Elad GilArchitecture isn't the variable; regulatory capture is
Sarah presses on whether non-transformer architectures could change the landscape, and Elad's answer is: whatever the underlying architecture, the industry will eat all available compute and power, so the pressure for more memory-efficient, more power-efficient architectures is greater than ever, but catching up to transformer at scale while matching hardware is still very hard. He offers two low-probability scenarios: some neolab secretly builds something better and raises more compute on the strength of it, or one person on a team jumps to Anthropic or OpenAI and the knowledge spreads — the latter being what has actually happened so far. He then cites an interview with Janssen of Janssen Pharmaceuticals on regulatory capture: drugs get more expensive and slower for two reasons, regulatory capture and the FDA looking only at risk and not at benefit. He worries the same structure will form inside the labs.
— Elad GilIn their own words · checked verbatim
And so suddenly we had this massive inflection in terms of valuations of these companies. And I think a lot of people now are assuming that there's a bunch of other trillion dollar companies that will be formed in three to five years. And, you know, that's unprecedented in human history.
Elad Gil2:04
You need a hundred billion of revenue or 50 to a hundred billion pretty easily. And so then the question with good margin, right? So then the question is, where are the 50 to a hundred billion dollar revenue streams for single companies?
Elad Gil6:06
People are conflating the two things right now, in my opinion.
Sarah Guo8:08
It's like a 12 to 18 month period, usually where the company's worth the most it'll ever be worth.
Elad Gil10:32
But the biggest opportunity cost is your time. Your most productive years of your life are on the line right now.
Elad Gil15:15
The data I have is that a number of very smart and even very self-aware research scientists have felt that, you know, there was some knee in the curve on recursive self-improvement or ASI 18 months away, every 18 months for the last five years.
Sarah Guo19:17
Which is a really interesting human power law, right? If you actually look at it in any field, there's at most a few dozen people who drive the field.
Elad Gil22:27
70% of France is still nuclear in terms of its power generation. Right? 70%! Where are all the accidents? And where are all the kerfuffles? And nothing. Nothing's happened.
Elad Gil36:10
Figures
| Companies going from near zero to a trillion-dollar valuation in five years | 3 (Anthropic, OpenAI, SpaceX) | 2:04 |
| Window when a company is worth the most | 12 to 18 months | 10:32 |
| Years of a normal cycle compressed into one year of AI | 3 to 4 years | 11:12 |
| France's nuclear share of electricity generation | 70% | 36:10 |
| US nuclear share of electricity generation | 18% | 36:10 |
| Japan's nuclear share of electricity generation | 25% | 37:11 |
| Years since the US built a new reactor | 40 years | 36:10 |
Glossary
- punctuated equilibrium
- An evolutionary theory: long periods of stability interrupted by short bursts, which Elad uses as an analogy for technology waves.
- RSI
- Recursive self-improvement: models participating in improving their own training, considered a key step toward ASI.
- ROIT
- Return on invested tokens: a term Elad coined for how a given token budget should be allocated and what return it yields.
- regulatory capture
- When a regulator is swayed by the industry it is supposed to oversee, so that the bar ends up serving incumbents.
- exit tax
- A tax on asset appreciation levied when leaving a jurisdiction; California is discussing adding one.
How to listen
AI founders deciding whether to keep raising or sell their company, and early-stage investors who need to recalibrate how many trillion-dollar companies there will be and the pace of exits.
The show intro and embed project recruiting from 0:00 to 1:44 can be skipped.