Equity Mates Investing Podcast Sep 21, 2026 43m 24m saved
With Cathie Wood, co-founder and CEO of ARK Invest
Anthropic's revenue run rate went from $9 billion in December to $65 billion six months later, and Cathie Wood says that is why the comparison to the tech and telecom bubble does not hold.
The bubble case is that too much capital chases one opportunity and destroys the return on it. Wood's answer is that the capital is being repaid inside a year, and she gave the numbers on a single gigawatt of data center capacity to show it.
"Its revenue run rate has gone from $9 billion in December to $65 billion in six months."
Cathie Wood, co-founder and CEO of ARK Invest, on Equity Mates, has run the firm since 2014, when her team bought Nvidia below a $10 billion market capitalization on the argument that autonomous vehicles would be run by graphics chips.
The full interview is covered here so you can skip it. 43 minutes of audio, 19 minutes of reading.
Here are the 12 calls that matter.
Key Takeaways
Anthropic's revenue run rate went from $9B in December to $65B six months later, which Wood says is more revenue added in seven months than Salesforce added in 25 years
She also says the firm's gross margins are reported at 80% and that it has been operationally profitable for two quarters
A gigawatt of data center built for $25B–$30B was resold for $34B a year, and Google later paid $50B per gigawatt
Healthcare is the most inefficiently priced part of the AI trade, on her argument that technology analysts will not cover regulated industries and healthcare analysts will not cover fast-moving ones
Intellia and CRISPR Therapeutics were both $200B companies in 2020; Intellia is now under $5B
Discovering one new drug costs $2.4B and takes 13 years — Wood expects AI and sequencing to cut that to $600M–$700M and eight years within five years
Insurers already pay $2.2M for a one-time cure, because the alternative is decades of emergency treatment
Bitcoin fell about 50% in this crypto winter against 75%–95% in previous ones, and she blames a Binance software fault, quantum-computing fear and AI taking the attention
Chinese and American models will split the way Android and iOS did — volume share to one side, profits to the other, with the cloud providers collecting the revenue either way
Humanoid robots in the home are about five years away, not the two Elon Musk suggests, and the bottleneck is that the training data does not exist yet
1. The Great Acceleration
The interview opened on the theme of ARK's Big Ideas 2026 report, and the host asked how fast technology is actually moving. Wood said the pace has surprised the people whose job is not to be surprised.
Even a first-principles research shop did not expect this rate
We have even been surprised and it takes a lot to surprise us because we're doing deep first principles research and to see AI taking off as quickly as it is.
Cathie Wood
She credited competition with China for part of it, and said the feeling of speed comes from the fact that every industry is affected at once. Her account of adoption has two directions. Chief executives are pushing it down through their companies because they cannot afford to be the one that did not. And people in or near entry-level jobs are pulling it up, because the same tool that could replace them is the one that makes them more productive.
Adoption is arriving from both ends of the org chart
So top down, bottom up by company.
Cathie Wood
The second thing happening, she said, is embodied AI: the digital and physical worlds converging.
2. Robotaxis, Then Robots
Her example of embodied AI is autonomous driving, and the point she wanted to make about it is that the speed is an illusion created by a long, invisible build-up.
The first commercial robotaxi is eight years old
But the first commercialization of a robo taxi was in 2018. It was Waymo in the US.
Cathie Wood
And it helps us explain or illustrate slowly then all at once.
Cathie Wood
ARK's research had expected Tesla to get there before June of last year; it did not, and is now moving faster than Waymo. The reason, on her telling, is data. Tesla has six to seven million vehicles collecting it. Wood owns three of them, two Model Ys and a Model 3, and has been sending road data back since 2018, which for Tesla started in 2015.
She flagged one reason Australian listeners will feel this later than Americans: robotaxis are electric, so the charging network has to exist before they can run continuously.
On humanoid robots she was willing to disagree with the person she agrees with most.
Musk's timeline is real, in Musk units
That's Elon time. It's probably going to be here in 3 to five years.
Cathie Wood
And that's probably closer to 5 years, I think, that we'll have humanoid robots in our homes.
Cathie Wood
The gap between three and five years is unit growth, which is what brings the cost down to something a household can afford. The gap before that is data, and there is none, because nobody has recorded housework at scale. Figure AI, held in ARK's venture fund, has started giving people cameras and sensors to wear while they do chores.
There is no corpus of a person doing the dishes
Because there is no data on it. And for AI to really take off, you have to have a big body of data.
Cathie Wood
She told the same story about Physical Intelligence in San Francisco, whose chief executive will not let a fund into a round unless someone from that fund's investment committee visits in person. Wood went. What she found was a room full of people doing physical tasks so a robot could learn from watching them.
3. Cancer Caught At Stage One
The application she rates highest is not transport.
Her ranking puts medicine above everything else AI touches
I think healthcare is the most profound application of AI we are now diagnosing cancer in stage one
Cathie Wood
The mechanism is two technologies arriving together: sequencing and AI. She expects diagnosis to move earlier than stage one. Freenome, another venture-fund holding, was co-founded by an oncological surgeon who went and learned AI because he thought the state of cancer data was unacceptable; the company is now detecting colorectal cancer at stage one from a blood test rather than a colonoscopy, and is in trials on precancerous polyps, which shed into the blood as they turn.
Behind that sits single-cell sequencing, and her framing of what it makes a human body into.
A body is now a readable dataset
So what are we? We are walking data factories because there's something called single cell sequencing.
Cathie Wood
She put the number of cells in a human body at 35 to 40 trillion. With sequencing, AI and gene editing together, she said, two rare blood diseases are already being cured, with CRISPR Therapeutics as the company doing it. The pipeline item she thinks matters more is a one-time gene edit for high cholesterol, which she called the category killer. One of the hosts looked it up mid-interview and reported that in trials it has cut blood lipid levels by roughly 50%.
Her complaint is that none of this is in the price.
Nobody believes it, which is what the early stage of any innovation looks like
this is what happens in innovation a lot of people doubt in the beginning doubt and then boom
Cathie Wood
4. How ARK Found Nvidia
Asked how she separates real innovation from a fad, Wood answered with the firm's own origin story rather than a framework.
In 2014 she sent Tasha Keeney to find out what an autonomous vehicle actually was, because Elon Musk had put one in his master plan. ARK publishes its research as it goes rather than when it is finished, on the reasoning that this is how you meet the people building the thing, and Keeney worked social media, academia and technology analysts.
The answer came back as a component, not a concept
It appears that the brains of an autonomous vehicle are going to be GPUs.
Cathie Wood
What made that useful was where the stock was.
In 2014 the company was priced as a fading PC parts business
Nvidia was selling off. It was below $10 billion in market cap. It was selling off because it was nothing more than a PC gaming chip company and PCs were being displaced by laptops.
Cathie Wood
ARK made it a top holding and spent the years from 2014 to late 2022 saying GPUs, autonomous driving, AI and robotics to an audience she says was not listening. She was candid that the firm took profits along the way, and that some would call it premature. Her defense is the question of what those profits then bought — Palantir, she said, at $9 or $10, which she saw as the same theme approached from a different angle.
5. The Analyst Divide
Asked what the next Nvidia is, Wood named a sector rather than a stock, and explained the mispricing as a staffing problem.
Technology analysts will not go near healthcare because it is slow, regulated, bureaucratic, political and dependent on insurers and governments. Healthcare analysts will not work the way technology analysts do.
The mispricing is structural, not a matter of opinion
So, I think that's the most inefficiently priced part of the market as it relates to AI.
Cathie Wood
Her illustration is Intellia, which she said has more robust patents than most and sits at an earlier stage than CRISPR Therapeutics. ARK first bought in the late 2010s. By 2020 Intellia and CRISPR were each $200 billion companies by market capitalization. Intellia today, she said, is under $5 billion.
She also reported an objection she has heard from healthcare analysts and does not accept: that curing a disease is a bad business model, because it ends the annuity of a daily pill or a monthly injection.
She treats the annuity objection as disqualifying
And I've heard some healthcare analysts saying curing disease that's a terrible model.
Cathie Wood
First of all, anything that's good for humanity is going to be highly valued.
Cathie Wood
The commercial answer is what payers already do.
Insurers are not waiting to be convinced about the price of a cure
Insurance companies today have no problem paying $2.2 million for a cure.
Cathie Wood
Her example is a rare inherited blood disease whose sufferers, for fifty or sixty years with no cure available, went into hospital ten to twenty times a year on an emergency basis for transfusions. That is the alternative the $2.2 million is measured against.
The money is currently spent on the wrong side of the line
Turning sick care into health care I think is one of the biggest opportunities we have.
Cathie Wood
I would say 75 to 80% of all healthcare dollars are dedicated to sick care.
Cathie Wood
6. $2.4B And 13 Years
The third healthcare argument is about the cost of making a drug at all.
One new drug, counting the failures on the way
So people are very surprised to learn that a large pharma or biotech company in order to discover and develop one new drug and this includes the trial failures along the way cost $2.4 billion and takes 13 years.
Cathie Wood
Her forecast is that AI and sequencing take that to $600 million to $700 million and to eight years or fewer within five years. The value of the time saving is patent life: on a 21- or 22-year patent, five extra years of revenue is a large fraction of the whole return.
She was explicit that this is an argument for active management rather than for the sector. Some companies will move quickly enough and some will not, and the ones that do not are simultaneously losing what she put at $300 billion of revenue to patent expiries and generic pricing.
7. Bitcoin's Third Winter
Asked whether she is still bullish on Bitcoin, Wood said yes, and then walked through why the past year was bad.
This drawdown was mild by the standards of the previous ones
The previous crypto winters went down, Bitcoin went down 75 to 95%.
Cathie Wood
This one, she said, took Bitcoin down about 50%. She gave three causes.
The first was the flash crash of 10 October last year, when a software fault at Binance triggered auto-deleveraging. A fund or decentralized-finance firm running a long on one exchange and a matching short on another was, in her account, left completely unhedged; estimates of the damage she quoted range from $15 billion to $30 billion, and a lot of people were wiped out.
The second was quantum computing, on the theory that Bitcoin is the most attractive target available because Satoshi's coins have never moved to more resistant wallets and a great deal of Bitcoin is simply lost. Her research says the timeline is not close: machines are working with about 100 qubits at a time, and cracking the cryptography needs hundreds of thousands.
Quantum is a real risk on a timeline nobody is trading
It's not going to happen for a long time.
Cathie Wood
The third was attention.
And then the third thing that happened was AI. AI was taking all of the oxygen out of the room and Bitcoin miners were leading the way.
Cathie Wood
Miners, she said, shifted from mining to selling their power to data centers. She put the price move as going from a little below $60,000 to roughly $80,000, and thinks it is bottoming.
Her case for the asset has not changed
We still think that Bitcoin is three revolutions in one.
Cathie Wood
The three are a technology and a currency native to the internet, which she says did not exist because nobody coding the internet in the early 1990s expected commerce to happen on it; a global monetary system that is private, rules-based and capped at 21 million units; and a new asset class, which she expects to become the largest. ARK's published targets are a base case in the mid-$700,000s and a bull case of $1.5 million, previously dated to 2030, and she said that date could still hold if institutions move in size.
The one fundamental headwind she named is stablecoins, and her reading of the Bitcoin community's response is that they are an interim step for people in emerging markets who cannot afford Bitcoin's volatility but will eventually prefer it to the dollar that backs the stablecoin.
8. The Big Four, And Throughput
Beyond Bitcoin, Wood named four assets ARK follows: Bitcoin, Ethereum, Solana and Hyperliquid. Her observation about bear markets is that they are when builders build, and what she watches is how the ecosystems specialize against each other — Hyperliquid on perpetual futures, Solana on high-throughput financial transactions, Ethereum on smart contracts.
The constraint she thinks decides the next phase is transaction throughput, and the number comes from Cloudflare. Its chief executive, Matthew Prince, has said that agentic commerce, meaning software agents transacting with each other in fractions of a cent, will require 20 million to 100 million transactions per second. Wood's comparison is that Visa, an entire traditional payments network, does about 20,000 payments per second. Ethereum and Solana are far below what would be needed. She disclosed that she advises a blockchain called LayerZero, which is aiming at 2 million.
9. What 2022 Taught Her
The hosts raised ARK's 2021 run and the 2022–23 fall, and asked what it taught her about risk. Her first answer was that the drawdown made no sense to her, because the tech and telecom bust happened when the technologies were not ready and cost too much, and this time they were ready.
Her second answer was a correction to her own playbook. ARK's normal behavior is to concentrate into its highest-conviction names in a bear market and diversify during a bull market, on the reasoning that a correction can arrive at any time. In that particular bear market, she said, the firm should instead have moved into the largest cash-generative technology companies.
She also drew a line under what volatility does and does not measure.
Volatility is a reading on uncertainty, not on risk
And the volatility is not a measure of risk which is a really important thing for your audience to understand. It's a function of uncertainty and our research gets us to a higher point in terms of certainty than I would say most investors.
Cathie Wood
The thing that actually damaged the portfolios was slower unit growth, and the cause was a supply-chain disruption she had not expected to run for three years, compounded by the fastest rate-rise cycle on record. Electric vehicles were the clearest casualty. Her reason for thinking the combination will not repeat is that it took a pandemic plus simultaneous fiscal and monetary expansion by every government at once to produce it.
The lesson was about unit growth, not about volatility
But we had no idea that the supply chain issues would last three years. And what kills us is when unit growth slows down.
Cathie Wood
10. Anthropic's Numbers
Asked directly whether the AI data center trade is a bubble, Wood said she is pleased that the question is being asked.
She accepts the mechanism and denies that it has happened yet
So, we believe we're in the early stages of the AI revolution.
Cathie Wood
Too much capital chasing an opportunity does eventually hit returns on invested capital, she said.
We are not anywhere near that right now.
Cathie Wood
Her evidence was Anthropic, which ARK owns in its venture fund.
The revenue is arriving faster than any comparable company booked it
Its revenue run rate has gone from $9 billion in December to $65 billion in six months.
Cathie Wood
That is as much revenue added in seven months as Salesforce added in its entire 25-year history, on her account. She also passed on, as a rumor she believes is well founded and attributed to the Financial Times without being certain of the source, that Anthropic's gross margins are 80% and that it has been profitable on an operating basis for two consecutive quarters. The company is preparing to go public, Wood noted, which gives it a reason to present well.
The harder number is the data center economics. She described xAI, now owned by SpaceX, building capacity at what she put at $25 billion to $30 billion per gigawatt, then selling capacity it was no longer using to Anthropic at $34 billion per gigawatt per year — more than a 100% return in the first year. Google, she said, subsequently paid $50 billion per gigawatt.
Her answer to the elder statesmen of technology who lived through 2000 and refuse to do it again is that the earlier cycle planted the seeds without the conditions to grow them: cloud in 2006, deep learning in 2012, transformer architecture in 2017.
The build-up is the point, not the recent move
What I've just described is 25 years of the rubber band stretching.
Cathie Wood
11. China And Commoditization
The hosts put the commoditization risk to her: Chinese models priced at a fraction of American ones. Wood noted that Palantir's Alex Karp, who spent a long time calling large language models commodities, has begun to soften that.
Her analogy for the split is the phone market
We actually think that these if you look at what's going on in China versus the US, what you'll find is it's a little bit like iOS and Android.
Cathie Wood
Android has 80% of the unit share out there. Apple has 80% of the profits, right?
Cathie Wood
The cost structure of running a large model, she argued, is much the same wherever it sits, and she would not guess at Chinese state subsidy. What she was certain about is where the money lands.
Whoever's model it is, the cloud bill is American
But the real winners from all of this are the cloud providers
Cathie Wood
Researchers in the United States use Chinese models, she said, and that usage accrues to whoever hosts them. ARK itself cannot use them, being a regulated financial firm. China's constraint is chip access, which it compensates for with power it has and the United States does not.
Her evidence that the market is putting a high value on the output is what her own colleagues pay.
A single knowledge worker's tooling bill has moved by two orders of magnitude
I'll tell you, our chief futurist, Brett Winton, I think he's using $4,000 a month, not $20 a month, $4,000 a month.
Cathie Wood
Winton, an MIT-trained engineer, would rather run ten projects himself than train a junior analyst, which is the substitution she thinks the price is actually measuring.
12. The Next Trillion
Asked to name the next company to reach a trillion dollars, Wood took both the public and the private list and went to Elon Musk.
She named Neuralink, which ARK owns, as a company at a very low valuation today that she can see reaching a trillion dollars eventually — not the next one there, but one that gets there. That would be Musk's third trillion-dollar company.
Her explanation for why is a thesis about data rather than about any one product. From ARK's 2019 podcast with him onward, she said, Musk understood that the companies he was assembling would each produce a different kind of data, and that the winner in AI is whoever holds the deepest domain expertise and the largest body of high-quality, differentiated data. X, SpaceX, Tesla, The Boring Company, the robotics data he is starting to collect, and Neuralink's data about the brain — the system that neural networks were modeled on in the first place.
She said Musk told her about nine months ago that his companies were converging more, and sooner, than even he had expected, and pointed to xAI being folded in.
Her largest number in the interview is not for a listed company
Just keep an eye on SpaceX. I think it is going to be the one of the most valuable companies talking about tens of trillions of dollars.
Cathie Wood
Bonus Insights
Her position on Twitter was a research decision, not a media one
Wood said she never expected that Twitter, which she thought of as a network for teenagers and celebrities, would become ARK's most important research channel. Publishing unfinished work publicly is how the firm meets the people building the technology it is trying to price.
She learned the term "OG" from Elon Musk on her own podcast
Recording with Musk alongside Tasha Keeney, Wood heard the phrase for the first time and had to ask Keeney what it meant.
The investment-committee visit is a condition of entry, not a courtesy
Physical Intelligence's chief executive will not accept money from a fund unless a member of that fund's investment committee comes to see him in person. Wood made the trip to San Francisco herself.
The charging network is why Australia gets robotaxis late
Her answer to a local audience was infrastructure rather than regulation: robotaxis are electric and cannot afford downtime, so the charging build-out has to come first, the same sequence the United States went through with range anxiety.
Wood's bottom line is that the AI cycle is 25 years of accumulated research arriving at once, that the return on the capital going into it is still being repaid inside a year, and that the part of it nobody is pricing is medicine.
Products, Companies & Tools Mentioned
Anthropic (Held in ARK's venture fund; run-rate revenue of $9B in December to $65B six months later, reported 80% gross margins and two profitable quarters, and preparing to go public)
Tesla (Six to seven million vehicles collecting road data since 2015; now moving faster on robotaxis than Waymo, on Wood's account)
Waymo (Ran the first commercial robotaxi service in 2018, the starting point for her "slowly then all at once" argument)
Nvidia (ARK's original thesis trade: bought below a $10B market capitalization in 2014 on Tasha Keeney's finding that autonomous vehicles would run on GPUs)
Palantir (Bought at $9 or $10 with proceeds from trimming Nvidia; its CEO Alex Karp is the source of the argument that large language models are commodities, which she disputes)
CRISPR Therapeutics and Intellia Therapeutics (Both around $200B of market capitalization in 2020; Intellia now under $5B. CRISPR's pipeline includes a one-time gene edit for cholesterol Wood calls the category killer)
Freenome (In ARK's venture fund; detecting colorectal cancer at stage one from a blood test, co-founded by an oncological surgeon who taught himself AI)
Figure AI and Physical Intelligence (Both collecting the physical-task data humanoid robots do not yet have; Figure is giving people cameras and sensors to wear while doing housework)
Binance (The software fault on 10 October last year that triggered auto-deleveraging and, on her estimate range, $15B to $30B of damage)
Ethereum, Solana and Hyperliquid (The other three of the four crypto assets ARK follows, specializing into smart contracts, high-throughput transactions and perpetual futures)
Cloudflare (Matthew Prince's estimate that agentic commerce needs 20M to 100M transactions per second, against about 20,000 payments per second at Visa)
LayerZero (A blockchain Wood advises, targeting 2 million transactions per second)
Neuralink, SpaceX and The Boring Company (The Musk companies she says are converging into one differentiated data set; she can see Neuralink reaching a trillion dollars and SpaceX reaching tens of trillions)
Salesforce (Her benchmark for the Anthropic revenue comparison, and the building the interview was recorded in)
Books & Resources Mentioned
ARK's Big Ideas 2026 report (The source of the "great acceleration" theme the interview opens on)
Listen to the full episode
🔴 YouTube | 🔗 Episode page
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: