ARK • Disrupt

Last week, Tesla reported fourth-quarter earnings and updated investors on its robotaxi program. Since launching on June 22¹ through December 2025, the fleet has logged ~650,000 miles and now has deployed ~500 vehicles in Austin and the Bay Area.² After removing safety monitors and chase cars in some instances, Tesla expects the fleet to double every month and to serve ~25–50% of the US by year-end, pending regulatory approval.³

As outlined in ARK’s Big Ideas 2026, vertically integrated manufacturing should become Tesla’s key competitive advantage, both in early commercialization and in scaled deployment. At its current deployment pace, Tesla will surpass Waymo’s latest reported fleet of ~3,000 vehicles in roughly three months,⁴ supported by a cost structure that could make Tesla’s hardware, the Cybercab, ~50% less expensive per mile than Waymo’s sixth-generation robotaxi, according to our research shown below. Based on that cost advantage, Tesla could price its service as low as ~$0.25 per mile at scale, or less than one tenth the cost of US human-driven ride-hail today, also shown below.

Note: Values are rounded. Source: ARK Investment Management LLC, 2026, based on data from Yipit 2025, AAA 2025, and GetTransfer 2025. In addition to those sources, certain information presented may be the result of ARK’s internal analyses, which draw on various additional sources of information. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results. Forecasts are inherently limited and cannot be relied upon.

According to our research, robotaxis will transform Tesla’s business model from one-off vehicle sales into a recurring revenue, cash-generation machine with software-like margins, positioning it to capture a significant share of the ~$34 trillion enterprise value opportunity for robotaxi platforms by 2030.

We look forward to monitoring Tesla’s robotaxi expansion in the coming months.

For decades, scientists have been able to read the human genome, the “letters” representing the ~3.2 billion base pairs in DNA. Genes that encode proteins, molecules that carry out most cellular functions, account for less than 2% of the human genome. The remaining 98%—the genome’s “dark matter”—has been far more difficult to interpret.⁵ Scientists understand that non-coding DNA plays a critical role in regulating gene expression and are researching the impact of variations in non-coding regions affecting biology and disease.

Last week, Google DeepMind introduced AlphaGenome, an artificial intelligence (AI) model designed to help solve that problem.⁶ DeepMind has developed several important AI systems for biology, including AlphaFold, a prediction model for protein structures that earned its developers the Nobel Prize in Chemistry in 2024.⁷ AlphaGenome extends that work upstream to predict how long stretches of non-coding DNA regulate gene activity. The new model can analyze up to one million DNA “letters” as inputs to predict how sequence changes impact gene expression, RNA splicing, chromatin accessibility, and other regulatory features that shape cellular behavior.

AlphaGenome’s breakthrough is important because many genetic variants associated with disease lie outside of protein-coding regions. In other words, their impact can be indirect and difficult to interpret.⁸ While it is a research tool and does not replace clinical testing, AlphaGenome could help scientists prioritize causal variants, improve the interpretation of genetic studies, and inform the design of more targeted genetic medicines—ultimately narrowing the gap between DNA sequences and biological mechanisms of action. Much like AlphaFold accelerated our understanding of proteins, AlphaGenome could illuminate the impact of regulatory dark matter on disease and health.

Last week in the field of genetic medicine, some programs advanced, others paused, and new approaches surfaced. In other words, from early-stage genetic breakthroughs to late-stage clinical development, the field is evolving rapidly.

Intellia Therapeutics reported that the U.S. Food and Drug Administration (FDA) lifted its clinical hold on MAGNITUDE-2, Intellia’s Phase 3 trial evaluating the in-vivo CRISPR-based gene-editing therapy nexiguran ziclumeran (nex-z) for hereditary transthyretin amyloidosis with polyneuropathy (ATTR-PN).⁹ Approximately three months ago, the FDA imposed the clinical hold,¹⁰ based on a serious liver safety signal in a patient enrolled in Intellia’s Phase 3 trial of nex-z in transthyretin amyloidosis with cardiomyopathy (ATTR-CM). The patient subsequently died from septic shock after a perforated duodenal ulcer, not liver failure, although the ATTR-CM trial remains on hold. To date, serious liver safety events have been rare, occurring in less than 1% of patients across more than 450 individuals treated with nex-z. Intellia plans to resume enrollment and dosing in MAGNITUDE-2 following protocol amendments that include enhanced liver safety monitoring, and we expect similar amendments to accompany a potential lift of the clinical hold for the ATTR-CM trial.

In contrast, the FDA placed clinical holds on two of REGENXBIO’s investigational gene replacement therapies—RGX-111 for Hurler syndrome and RGX-121 for Hunter syndrome—both of which use the adeno-associated virus (AAV),¹¹ a commonly used viral vector, to deliver functional copies of genes into cells. The FDA placed the holds after a brain neoplasm, or abnormal mass of tissue that can be benign or cancerous, was detected in a five-year-old patient who received RGX-111 approximately four years earlier. Preliminary genetic analysis identified integration of the viral vector’s genetic material near a proto-oncogene, a normal gene that can promote cancer growth if disrupted, without determining causality. That investigation is ongoing.

In our view, this event highlights some of the ongoing challenges for viral vector–based gene therapies, particularly those for pediatric central nervous system disorders. These programs involve durable, often one-time treatments administered early in life, raising the importance of understanding long-term safety risks such as rare vector integration events. At the same time, treatment may take place after irreversible neurological damage, suggesting that gene replacement might slow disease progression rather than fully restore lost function. While gene replacement therapies can provide meaningful clinical benefit—especially in diseases with few or no disease-modifying options—the benefit-risk assessments become more complex as programs mature because of long-lasting exposure, delayed treatment, and partial functional recovery.

In answer to the future of genetic medicine, two other developments offered important information about gene therapy last week. The first came from Life Biosciences, a startup co-founded by longevity researcher David Sinclair, which received FDA clearance to initiate a first-in-human trial of a gene therapy designed to partially “reset” cells by restoring epigenetic information—the chemical instructions that help regulate gene expression.¹² Rooted in Sinclair’s information theory of aging, the approach proposes that aging is driven primarily by epigenetic dysregulation.¹³

Sinclair’s therapy builds on Shinya Yamanaka’s Nobel Prize-winning research, which demonstrated that four master regulatory proteins—transcription factors that play a central role in controlling gene activity—can rewind adult cells to their more youthful states.¹⁴ Intentionally, Life Biosciences delivers only three of those factors, avoiding one known to promote cancer. While often discussed in the context of longevity, the company’s first clinical application is targeting the loss of vision caused by optic nerve damage. Success in this setting could open the door to broader applications of cellular rejuvenation across multiple organs.

Future research on gene editing is likely to focus on longer gene edits targeting a larger swath of genetic diseases than those feasible with first-generation tools that simply disrupt disease-causing genes or make short edits to correct mutations. Last week, Eli Lilly announced a research collaboration with startup Seamless Therapeutics to develop programmable recombinase-based therapies for genetic hearing loss.¹⁵ Seamless is engineering recombinases to enable “gene-sized” site-specific DNA insertions, exchanges, or excisions. Along with recent advances like the Arc Institute’s bridge editing, programmable recombinases illustrate how the gene-editing toolkit is evolving to address mutations beyond the reach of first-generation gene-editing approaches.¹⁶ For diseases like Huntington’s and Friedreich’s ataxia, which involve repeat expansions, or diseases like Duchenne muscular dystrophy with mutations spread across multiple sites, gene editing capable of multiple kilobase edits could be applicable. Importantly, even partial excision of toxic base repeats within a gene can yield clinical benefit, as disease severity often correlates with repeat length.

Last week’s news highlights progress in both late-stage and early-stage genetic medicine. Late-stage gene-editing programs, like those from Intellia, are evolving toward the first in-vivo gene-editing therapies before the end of the decade, while early-stage genetic medicine innovation is expanding the range of solutions to genetic problems.

Ahead of its purported initial public offering (IPO), rumors are flying that SpaceX is exploring a merger with either Tesla or xAI.¹⁷ Historically, while separated financially, SpaceX and Tesla have shared talent and engineering resources.

A merger would introduce some interesting challenges. As separate entities, for example, transactions associated with an orbital data-center stack—fabs, chips, and rocket launches—would generate revenue for each company. If the companies were to merge, however, those transactions would become cost of goods sold to the combined entity.

A SpaceX–xAI combination is the higher probability, as suggested both by Elon Musk’s endorsement on X¹⁸ and, as of January 30, Polymarket’s odds: ~62% for a SpaceX–xAI merger versus ~13% for a SpaceX–Tesla merger.¹⁹

This year, its robotaxi platform is likely to start generating recurring cash flow that will support more leverage to scale Tesla’s platform globally. Meanwhile, SpaceX capital intensity will increase. A SpaceX–xAI pairing would give xAI direct access to SpaceX’s orbital data centers—and a leg up against rivals like OpenAI, which also is pursuing space-based compute resources.

Last week, Grail formally submitted to the FDA its final application for Premarket Approval (PMA) of the Galleri multi-cancer early detection (MCED) test, a key step in population-scale screening for more than 50 cancer types.²⁰ The submission initiates a review process that could span 12 to 18 months, with a potential approval window in mid-to-late 2027. FDA approval should unlock Medicare reimbursement through a potential National Coverage Determination (NCD) pathway, based on the recently advanced MCED legislation. Simultaneous to the federal payer route, Grail is also eyeing the U.S. Preventive Services Task Force (USPSTF) pathway. A Grade A or B recommendation from the USPSTF, which typically requires strong clinical utility data, would mandate $0 cost-sharing coverage across commercial insurance plans.²¹

A cornerstone of Grail’s FDA application is its Pathfinder 2 study, a prospective interventional trial involving over 25,000 participants with results at 99.6% specificity, or 0.4% false positives—a critical metric for testing asymptomatic populations.²² While overall sensitivity was 40%, it rose to 73% for a specified group of the most common and lethal cancers. The test also demonstrated a positive predictive value (PPV) of 62%: nearly two-thirds of patients that received a positive "cancer signal" ultimately were diagnosed with the disease. By adding Galleri to standard-of-care screenings, researchers effectively doubled the number of cancers detected from 67 to 133, 53% of which occurred within the more readily treatable Stage I or II window.

To validate the long-term impact of those findings, Grail is leveraging the National Health Service (NHS)-Galleri study, a large randomized controlled trial in the United Kingdom that includes 140,000 patients with three to four years of follow-up. The primary objective is to prove a "stage shift"—or statistically significant reduction—in the incidence of late-stage III and IV cancers compared to the control group. While Pathfinder 2 successfully identified early-stage cancers that standard screenings would have missed, the NHS data will elucidate whether those cancers were identified early enough to produce better outcomes. With a readout expected in mid-2026, the study could prove mortality benefits.

On the legislative front, the Nancy Gardner Sewell Medicare MCED Screening Coverage Act cleared a major hurdle in the House in January.²³ If signed into law, the bill would create a formal pathway for the Centers for Medicare and Medicaid Services (CMS) to cover FDA-approved blood-based cancer screening tests for seniors in 2029. The legislation targets the 50-65 age demographic and would reimburse at a ~$650 price point, the current Cologuard colorectal cancer test reimbursement. Importantly, the bill will preserve the independent USPSTF process as a path toward broad commercial coverage.²⁴

Ultimately, the FDA’s review will focus heavily on safety, specifically the risk of the "diagnostic odyssey" triggered by false positives. A low false positive rate is paramount to minimizing the number of healthy patients undergoing expensive and potentially invasive follow-up procedures, such as PET scans or biopsies. While relatively non-invasive, the cumulative cost to the healthcare system of unnecessary PET scans is a major concern for bodies like the USPSTF.

Even so, consumers are voting with their wallets, as Grail has sold ~500,000 non-reimbursed tests to date for ~$1000 each.²⁵ That utilization could increase with further regulatory and reimbursement milestones.