#473: Nuclear Energy Continues To Gain Momentum In The United States, & More

The US is turbocharging its nuclear ambitions, thanks to hyperscaler demands and multilateral support following decades of neglect.¹ In his recent Executive Order, President Trump set forces in motion to increase nuclear capacity in the US ~4x by 2050, starting with the construction of ten large reactors by 2030.² During its second-quarter earnings call last week, Cameco—which owns 49% of Westinghouse—reported a “wall of business” for the AP1000,³ a certified Generation III+ reactor well aligned with Trump’s goals.⁴

Thanks to this gale-force tailwind, the AP1000 is poised to become the leading large-scale reactor in the US, a strategic imperative. Today, China is building ~30 large-scale nuclear reactors, compared to 0 in the US, as shown below.

XXX 

Source: ARK Investment Management LLC, 2025, based on data from PRIS 2024.⁵ For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. 

Small Modular Reactors (SMRs) are likely to play a key role in the nuclear race. With more than 120 technologies under development,⁶ companies focused on SMRs are in start-up mode. Their go-to-market strategies and execution will separate the winners from losers. 

Increasingly, both the US and China are turning to nuclear as a strategic energy source for data centers in the AI race. Large-scale reactors and emerging SMR technologies are central to their strategies. Design and deployment will be critical to meeting US energy security goals, as well as scaling data centers and reducing emissions.

Ethereum Layer 2 (L2) solutions are execution environments designed not only to enhance Ethereum’s scalability by processing transactions offchain but also to take advantage of its robust security by netting them onchain. Built on top of Ethereum, these blockchains host their own ecosystems—complete with applications, liquidity, and users—and settle compressed transaction data on Ethereum Layer 1 (L1) for verification. L2s are critical to Ethereum’s vision of scaling to enable global demand and use cases from DeFi to microtransactions while preserving the network’s security.

Last year, one of the notable L2s, Coinbase’s Base generated ~$89 million in revenue⁷— only 14% of went to Ethereum.⁸ Base generates revenue primarily from sequencer fees, also called base fees, which users pay to transact.

That said, Base is facing competition, notably from Robinhood’s announced L2, Kraken's L2 Ink, the Securitize/Ethena-backed chain Converge, and projects like Worldcoin. Both Coinbase and Robinhood have opted for technical stacks that provide out-of-the-box infrastructure, ease of deployment and Ethereum as the “mother” blockchain technology.

Base has the first-mover advantage, supported by a growing user base and strong network effects. Robinhood could have an edge in tokenization, leveraging its deep equities offerings as a bridge between traditional finance and blockchain infrastructure.

Both Coinbase and Robinhood believe that financial markets will rebuild onchain, their swift development of L2 platforms is a strategic push to capture the lead. With tech stacks tailored for the onchain opportunity, each company is aiming to create scalable, seamless platforms while navigating complex regulatory environments. Coinbase’s ecosystem is more native to the opportunity, while Robinhood is using its superior user interface as one of its competitive advantages.

Ultimately, success will hinge on which platform can abstract away crypto complexity and deliver the better user experience. As the L2 landscape evolves, competition between Base and Robinhood will power the innovation in onchain finance.

CRISPR’s in-vivo gene editing potential is significant, despite hurdles associated with specificity, immunogenicity, and efficiency. Researchers in the United States and Switzerland recently demonstrated a sophisticated approach that combines microbial genome mining with generative AI to engineer a new, improved CRISPR enzyme: OpenCRISPR-1.⁹

First, the researchers conducted a comprehensive data-mining operation across microbial genomes, scanning 26.2 trillion bases, an unprecedented effort that identified more than 1.2 million CRISPR operons¹⁰ and significantly expanded known CRISPR-Cas9 variants. Leveraging that massive dataset, they fine-tuned ProGen2, a generative protein language model initially trained on hundreds of millions of protein sequences.

Following this hierarchical training process, the model generated more than 350,000 candidate enzymes. Rigorous computational filtering then identified 209 high-potential candidates that subsequently were validated in human cells. OpenCRISPR-1 emerged as the leading enzyme, distinct from the widely used SpCas9 by over 400 amino acid mutations and differing from its nearest natural relative by more than 180 mutations.

The benefits were provocative. OpenCRISPR-1 achieved comparable on-target efficiency to SpCas9 with a significant enhancement in specificity, reducing off-target edits by 95%. Its off-target activity profile matched known Cas9 enzymes, with no unexpected cleavage sites. Moreover, OpenCRISPR-1 demonstrated lower immunogenicity than Cas9, making OpenCRISPR-1 potentially less detectable by the human immune system.¹¹

While the enzyme requires further validation across diverse genomic targets and delivery systems before confirming its clinical viability, OpenCRISPR-1's approach—combining extensive natural database mining with targeted generative AI—seems to mark a significant leap forward. Now, researchers can engineer highly specialized, efficient, and minimally immunogenic CRISPR enzymes, potentially reshaping the therapeutic market by enabling truly personalized "n-of-one" treatments tailored to individual genetic profiles...potentially creating new markets.

Tesla has designed its own custom silicon to power Autopilot and Full Self-Driving. Samsung fabricated Tesla’s current generation chip, the AI4, but TSMC won the bid for AI4’s successor, AI5. Elon Musk stated last year that AI5 will have ~10x the capability of AI4.¹²

Now, Tesla has signed a $16.5 billion deal with Samsung’s new Texas-based fab to manufacture its AI6 chip.¹³ Musk will be deeply involved¹⁴ in “maximizing manufacturing efficiency,” signaling a deep level of collaboration. 

A champion of vertical integration, Tesla’s use of custom silicon allows it to tailor its hardware stack to meet the software demands of its cars and data centers. Unlike the chip for its cars, Tesla’s data center chip, Dojo, has yet to unseat Nvidia as its primary source of training compute. Tesla is one of the largest enterprise purchasers of Nvidia GPUs,¹⁵ which could change with AI6. Musk recently highlighted the potential to converge inference and training chips into a single design for both its cars and data centers. 

Three months into spearheading the recalibration of U.S. Food and Drug Administration (FDA) standards, Dr. Vinay Prasad is out as Chief Medical and Scientific Officer and Director for the Center of Biologics Evaluation and Research (CBER).¹⁶

Prasad's departure comes on the heels of a high-profile standoff between the FDA and Sarepta Therapeutics over its Duchenne muscular dystrophy gene therapy, Elevidys.¹⁷ After a third patient linked to Sarepta’s AAVrh74-based platform died—notably, first publicly disclosed by outside media rather than the company itself¹⁸—the FDA requested Sarepta halt all shipments.¹⁹ Sarepta ultimately complied,²⁰ but the episode created a backlash from the Duchenne patient community and advocacy groups, especially parents, concerned that the FDA’s intervention would deprive families of a therapy that could halt or reverse progression of a severe and ultimately fatal condition.

Sarepta seemed to provide the opening for more controversy. Some right-wing commentators,²¹ high-profile media outlets,²² and political communities accused Prasad of engaging in past partisan politics that could undermine the mandates established by HHS Secretary Robert F. Kennedy, Jr. 

Prasad’s short-lived tenure signals the FDA’s willingness to challenge legacy practices. Protracted drug development timelines,²³ escalating costs,²⁴ and the 90% failure rate in clinical trials²⁵ collectively have prevented the treatment of many diseases. Something needs to change.²⁶ As breakthroughs in genetic medicine and AI-enabled drug discovery create transformative opportunities to accelerate clinical timelines and outcomes, the FDA needs to modernize.

What comes next for the FDA? Dr. George Tidmarsh, the newly appointed Director of the Center for Drug Evaluation and Research (CDER), now is serving also as Acting Director of CBER—Prasad’s former post.²⁷ An oncologist and biotech veteran, Tidmarsh brings both clinical and industry chops to the role, as Dr. Makary guides the FDA to accelerate cures without sacrificing transparency or scientific integrity.²⁸ Under his leadership, the FDA is likely to facilitate innovation—streamlining regulatory paths and leveraging AI and big data to increase efficiency. Dr. Makary’s next appointee to the CBER is likely to share those priorities and catapult the FDA into the new age. 

Last week, ARK analysts Brett Winton, Sam Korus, and Frank Downing visited Austin to test Tesla’s Robotaxi. The experience was surreally normal. When autonomy works, it fades into the background. You stop thinking about the missing driver and enjoy the ride. We spent the downtime recording a video…coming soon!

Our two big takeaways?

1. Driverless is underrated: A taxi without a driver is better.
2. Normalization is fast: Many people see robotaxis as sci-fi but, in cities where they operate, no one bats an eye.

Now that Tesla's service is live, we will be able to track its momentum. Removing the safety driver will be key to scaling. In 2020, Tasha Keeney highlighted how a human-driven ride-hail service would accelerate Tesla’s autonomous effort.²⁹ Here we are!

In our view, framing robotaxis as a threat to Uber and Lyft is too narrow a perspective. Human-driven ride-hail currently constitutes only 1–3% of US vehicle miles traveled (VMT).³⁰ Thanks to their dramatically lower costs, robotaxis will address and compete not only with Uber and Lyft but with all drivers, both personal and commercial. In that context, the disruption of human-driven ride-hail like Uber and Lyft is likely tobe a blip on the path to the transformation of transportation.