ARK • Disrupt

Two weeks ago, the U.S. Food and Drug Administration’s (FDA) Chief Medical and Scientific Officer, Dr. Vinay Prasad, previewed a “plausible mechanism” pathway to accelerate the delivery of bespoke therapies.¹ Published in the New England Journal of Medicine last week, the FDA’s paper provides an early blueprint for this future.²

The FDA’s bold regulatory shift recognizes what families of rare disease patients have understood for years: the biology associated with many genetic conditions is well established. With this new approach, to support a marketing authorization, the FDA is likely to accept streamlined evidence—including data generated across several consecutive patients—when both the genetic cause of a rare disease is known and a therapy targets it directly.

The FDA is responding to lessons learned earlier this year from Baby KJ, the infant who received a first-in-kind, custom gene-editing treatment for a rare metabolic disorder.³ KJ’s story captured global attention and stirred both hope and heartbreak in families whose children face similarly devastating diagnoses. Prasad’s move suggests that the FDA understands that Baby KJ’s case was not a fluke, and that soon more patients might gain access to previously unimaginable therapeutic options.

The implications for the healthcare industry are profound. Instead of waiting years to populate and collect data in rare-disease trials, companies should be able to advance modular therapeutic platforms that scale across many genetic variants by relying on shared mechanisms of action, chemistry, and manufacturing.⁴ In such cases, evidence of biological plausibility, target engagement, and clinical improvement could enable therapies to move from treatment of just a few patients to many patients—potentially compressing drug development timelines dramatically.

Importantly, the new framework may not be limited to rare diseases. While prioritizing pediatric, fatal, and severely disabling genetic conditions, the FDA also acknowledges the potential to expand it to common diseases—particularly those lacking effective treatments—and to model modalities beyond gene editing, including small molecules and antibodies.

While the FDA has yet to provide formal guidance about this pathway in practice, the direction is unmistakable. Regulators are trying to catch up to the pace of scientific progress and address the urgent, unmet needs of patients who cannot wait a decade for trials to read out.

This year, the price of bitcoin has been buffeted by a battle between early adopters and institutions. So-called “whales” have been selling, while public companies and ETFs have been buying. Indeed, this dynamic has characterized this bull cycle.

“Liveliness” is a metric that captures this phenomenon. Liveliness measures the ratio of total coindays destroyed relative to total coindays ever created: it rises when long-dormant coins make a move on-chain and falls when holders are inactive. At 0.89, its highest level since 2018, liveliness suggests that longer-term holders have been taking profits at the highest rate since 2021, as shown below.⁵

Source: ARK Investment Management LLC, 2025, based on date from Glassnode and BitcoinTreasuries.net as of November 15, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency. Past performance is not indicative of future results.

While early adopters have distributed liquidity, institutional demand seems to have absorbed it. “Vaulted” supply, meaning “stagnant” coins, has declined throughout the bull market, as shown below:

• from 7,969,098 BTC at the start of 2024,
• to 7,553,973 BTC at the start of 2025,
• to 7,318,450 BTC as of November 15, 2025.

Source: ARK Investment Management LLC, 2025, based on date from Glassnode and BitcoinTreasuries.net as of November 15, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.

Conversely, institutional balances have surged. The number of bitcoins in US spot ETFs has increased, as shown below:

• from zero at the launch of US spot bitcoin ETFs on January 11, 2024,
• to 1,125,507 BTC on January 1, 2025,
• to 1,332,379 BTC as of November 15, 2025.

Source: ARK Investment Management LLC, 2025, based on date from Glassnode and BitcoinTreasuries.net as of November 15, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.

Public company bitcoin holdings, including digital asset treasuries (DATs), have increased, as shown below:

• from 271,996 BTC at the start of 2024,
• to 598,995 BTC at the start of 2025,
• to 1,056,367 BTC as of November 15, 2025.

Source: ARK Investment Management LLC, 2025, based on date from Glassnode and BitcoinTreasuries.net as of November 15, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.

Combined ETF and public company holdings reached 2,388,746 BTC as of November 15. From January 1, 2024, to November 15, 2025, the combined flows from ETFs and public companies, plus the reduction in vaulted supply, totaled 1,466,102 BTC, suggesting that institutional demand has offset early-adopter selling pressure. Year-to-date data suggests the same, with a net delta of 428,721 BTC.

Together, these data suggest a structural transfer of bitcoin supply from long-time holders to institutions initiating positions in a new asset class. Through year-end and in 2026, given the sensitivity of institutions to changes in dollar liquidity, the end of quantitative tightening⁶ and lower interest rates—both of which we believe are likely—should bode well for BTC.

ARK’s research suggests that humanoid robots will reach human level performance in 2029. As ARK’s Chief Futurist Brett Winton noted recently,⁷ the difference in the difficulty of scaling autonomous robotaxis and general-purpose humanoid robots will not be incremental. Informed by Tesla‘s AI compute scaling, our research suggests that humanoid robots will be ~200,000x more complex to scale than robotaxis have been.

Fundamentally, autonomous mobility is a function of compute and performance: in robotaxis, the higher the investment in compute, the higher the number of miles between interventions. By regressing that relationship, we have modeled the non-linear amount of compute required for each order of magnitude improvement in autonomy, as shown below.

Source: ARK Investment Management LLC, 2025. This ARK analysis draws on a range of external data sources as of November 14, 2025, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

Musk contends that Full Self-Driving (FSD) must exceed human performance,⁸ represented by the dashed circle in the chart above, before robotaxis will scale. Humanoid autonomy should follow a similar path.

Our estimated 200,000x complexity ratio represents the degree of perfection a humanoid robot will have to achieve before operating reliably in the real world. Based on our established AI improvement curve, the graph above highlights the enormous compute that will be necessary to solve an autonomy challenge 200,000x more complex than robotaxis.

According to our research, humanoids will be capable of human-level performance across varied tasks in 2029, as shown below. While commercial sales of early-stage, limited-function robots could begin in the next few years, generalized performance will demand much more investment in compute, $150 billion at the very least.

Source: ARK Investment Management LLC, 2025. This ARK analysis draws on a range of external data sources as of November 14, 2025, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.