Hook: A Portfolio That Spans the Entire Spectrum of AI's Physical Manifestation
The Q3 filing landed with the kind of clinical brevity that usually signals either confidence or concealment. MSX Pre-IPO III, the third vintage of the Barcelona-headquartered fund's pre-IPO strategy, had added two names to its roster: Neuralink and Anduril Industries. Two portfolio companies. Two lines on a cap table. But the implications ripple far beyond the fund's own return projections.
Neuralink, the brain-computer interface company founded by Elon Musk, and Anduril, the defense technology firm helmed by Palmer Luckey, represent the two most extreme poles of AI's migration from the digital realm into physical and biological reality. One embeds silicon into the human cortex; the other embeds autonomous decision-making into weapons systems. The juxtaposition is not incidental. It is a thesis.
The information available in the original announcement was, by any forensic standard, thin. Two data points, one of which was an opinion. No source citations. No financial terms. No valuation details. But the absence of detail is itself a data point. When a fund with MSX's pedigree takes positions in two companies with combined implied valuations approaching $400 billion, the silence around terms speaks volumes about competitive dynamics in the pre-IPO allocation market.
What follows is a systematic reconstruction of what this investment actually means. Not from the press release, but from the underlying technical, commercial, and structural realities of both companies.
Context: The Pre-IPO Market's New Physics
The pre-IPO market has undergone a structural transformation since the 2021 SPAC bubble burst and the subsequent 2022-2023 liquidity contraction. Traditional venture capital, with its 10-year fund cycles and tolerance for negative cash flow, has given way to a new species of investor: the pre-IPO crossover fund that demands near-term liquidity events and operates with the discipline of a public market arbitrageur.
MSX Pre-IPO III sits squarely in this category. The fund's stated strategy is to enter positions 1-2 years before anticipated IPOs, hold through the listing, and exit within 2-3 years at a target multiple of 3-5x. This is not venture capital in the traditional sense. It is a timing play predicated on the assumption that private market valuations still lag public market comparables for the same quality of assets.
The inclusion of Neuralink and Anduril in the same vintage tells me three things about the fund's internal thesis. First, that MSX believes both companies are within striking distance of public listings—Neuralink in the 2027-2028 window, Anduril in the 2026-2027 window. Second, that the fund views "AI-enabled hardware" as a distinct asset class with pricing power independent of the software-focused AI boom that has dominated public markets since late 2022. Third, and most tellingly, that MSX has concluded these two companies represent the most defensible positions in the AI value chain that extends beyond the data center.
The timing is not accidental. Both companies have hit critical regulatory and commercial milestones within the past 18 months. Neuralink received FDA breakthrough device designation in 2025 and approval for its CONVOY feasibility study in November of that year, which will connect the N1 implant to assistive robotic arms. Anduril crossed the $1 billion annual revenue threshold in 2025, a 150-180% year-over-year increase, and secured a $250 million contract from the Defense Innovation Unit for "large-scale expeditionary operations."
For a pre-IPO fund, these milestones transform the risk calculus. The companies are no longer pure story stocks. They have regulatory validation and revenue. The question is whether the valuations already price in the remaining execution risk.
Core: A Technical Teardown of the Portfolio's Underlying Assets
Neuralink's Engineering Reality: The Gap Between Headlines and Hardware
Neuralink's technical trajectory is frequently obscured by the celebrity of its founder. Setting aside the narrative, the engineering is genuinely impressive in specific, measurable ways. The N1 implant features 1,024 channels distributed across 64 flexible polymer threads, each approximately one-quarter the diameter of a human hair. The R1 surgical robot, which performs the insertion, uses real-time imaging to avoid vascular structures during implantation. These are not incremental improvements on existing brain-computer interface technology. They represent a fundamental re-architecture of the interface between neural tissue and silicon.
The 1,024-channel count matters because it addresses the primary constraint that has limited brain-computer interfaces to laboratory curiosities: bandwidth. Synchron's Stentrode, the leading non-invasive competitor, operates with approximately 16 channels. Precision Neuroscience's Layer 7 array offers higher resolution but has yet to demonstrate the long-term stability that Neuralink's approach theoretically provides. The difference between 16 and 1,024 channels is not 64 times more capability. It is several orders of magnitude more in terms of the complexity of signals that can be decoded.
As of the most recent data, Neuralink has performed three human implantations. The first, in January 2024, was followed by a second in August 2024, and a third in 2025. All three patients have demonstrated the ability to control digital interfaces with thought. The CONVOY study, approved in November 2025, extends this to physical control of assistive robotic arms. These are meaningful clinical data points, but they constitute a sample size that would be considered inadequate for any traditional medical device submission.
The critical technical risk is not the initial implantation but the long-term behavior of the electrodes. Neural tissue is a hostile environment for foreign objects. Glial scarring, electrode degradation, and signal attenuation over time are the failure modes that have historically limited the lifespan of invasive brain-computer interfaces. Neuralink's flexible thread design is intended to mitigate these issues, but the longest any human patient has carried the device is approximately two years. The FDA's requirement for longitudinal safety data, which will be the gating factor for commercial approval, remains an open question.
From a cryptographic perspective, the security architecture of the N1 implant deserves more scrutiny than it has received. The device transmits neural data wirelessly, which creates an attack surface for interception or injection. The most sensitive biometric data a human being possesses—their raw neural activity—will be transmitted over the air. Neuralink has not published a detailed security whitepaper, and the absence of such documentation, given the sensitivity of the data involved, is a concern that the market has not adequately priced.
Anduril's Software-Defined Warfare: The Lattice Architecture
Anduril presents a different technical profile. The company is not building a single product but a platform—the Lattice operating system—that integrates sensor data, autonomous decision-making, and human oversight into a unified command and control framework. This is the "software-defined defense" thesis in its purest form: hardware becomes interchangeable, software becomes the moat.
The Lattice architecture processes data from a heterogeneous network of sensors and effectors. Its product matrix includes the Anvil counter-drone system, the Dive-LD autonomous underwater vehicle, the Ghost 4 reconnaissance drone, and the Roadrunner missile system, which is notable for its vertical takeoff and landing capability and its reusability. The Roadrunner is a particularly significant technical achievement, as it combines the speed of a missile with the recoverability of a drone, fundamentally changing the economics of air defense.
The partnership with OpenAI, announced in June 2025, is more significant than the press coverage suggested. The collaboration is focused on counter-unmanned aircraft systems, which are currently the highest-priority air defense problem for the US and its allies. The technical challenge is not detection but classification and response—distinguishing hostile drones from commercial traffic, and making engagement decisions at machine speed. This is a problem domain where large language models, OpenAI's core competency, are not obviously suited. The partnership may prove to be more about political signaling than technical integration.
The December 2025 acquisition of Callum, the British 3D-printing drone manufacturer, for $1.3-1.4 billion, tells me something important about Anduril's manufacturing strategy. The company has recognized that software superiority cannot compensate for hardware supply chain vulnerability. By bringing additive manufacturing capabilities in-house, Anduril is positioning itself to produce drones at a scale and speed that traditional defense contractors, with their reliance on precision machining and specialized suppliers, cannot match. This is the same logic that drove the original "thousands of drones" concept—warfare is becoming a contest of production rates, not just technological sophistication.
The revenue trajectory supports the technical assessment. Anduril is projected to generate $1 billion in 2025 revenue, a 150-180% increase from the prior year. The gross margins on software-embedded defense products are substantially higher than traditional hardware contracts. The company claims to be on track for break-even in 2026 and profitability in 2027. These numbers, if accurate, would make Anduril one of the fastest-growing defense companies in history.
The Technical Complementarity That the Market Has Missed
The standard analysis of this portfolio treats Neuralink and Anduril as independent bets on unrelated technologies. This is a misreading. The two companies represent opposite sides of the same coin: the integration of artificial intelligence into human biological systems and into autonomous physical systems.
Neuralink's trajectory points toward human augmentation—the enhancement of human cognitive and motor capabilities through direct neural interfaces. Anduril's trajectory points toward human substitution—the replacement of human decision-making in combat environments with autonomous systems. These are not contradictory. They are complementary expressions of the same underlying trend: the erosion of the boundary between organic intelligence and artificial intelligence.
For a pre-IPO fund constructing a portfolio thesis, this complementarity provides narrative coherence that can be marketed to limited partners. "From brain to battlefield" is a compelling story. But the story has a technical foundation that is rarely articulated. Both companies are building the infrastructure for a world in which AI is not a tool that humans use but an environment that humans inhabit. Neuralink is building the interface. Anduril is building the autonomous agents that will operate within that environment.
Contrarian: What the Bulls Get Right
A rigorous analysis requires acknowledging the validity of the opposing case. The bearish narrative on this portfolio is well-rehearsed: Neuralink is a clinical-stage company with three patients and no revenue, valued at $80-100 billion; Anduril is a defense contractor with a single dominant customer, facing competition from incumbents with decades of relationship capital. Both valuations are said to embed "AI narrative premium" that could evaporate if the hype cycle turns.
The bulls would respond, correctly, that this analysis misses the structural shift that both companies represent.
For Neuralink, the counterargument is that the relevant comparison is not other brain-computer interface companies but the entire class of medical device platforms that have achieved platform status. Consider the trajectory of Intuitive Surgical's da Vinci system, which faced similar skepticism about its clinical utility and cost-effectiveness in the early 2000s, and which now generates over $7 billion in annual revenue from a installed base of robotic surgical systems. The key insight is that platform companies in healthcare do not win on the first indication. They win on the breadth of applications that the platform can address.
Neuralink's 1,024-channel platform, if it achieves commercial approval for its initial indication (assistive communication for paralyzed patients), becomes the foundation for a series of additional indications: motor restoration, sensory feedback, treatment of neurological disorders, and ultimately cognitive enhancement. The market for the initial indication may be small—hundreds of thousands of patients globally—but the addressable market for the platform across all indications is in the tens of millions.
The technical question that will determine whether this platform thesis holds is whether the electrode array can maintain signal quality over multi-year time horizons. If the electrodes degrade significantly after 2-3 years, requiring explantation and re-implantation, the platform economics collapse. If, however, the flexible thread design achieves the stability that the preclinical data suggests, the platform becomes a permanent infrastructure layer for human-machine interaction.
For Anduril, the bull case is about the structure of the defense procurement market. Traditional defense contractors like Lockheed Martin, RTX, and Northrop Grumman are optimized for a procurement environment characterized by large, multi-year development programs with cost-plus contracts. They are not optimized for the current threat environment, which demands rapid iteration, software updates, and production at scale for attritable systems.
The conflict in Ukraine has demonstrated that modern warfare consumes drones at a rate that traditional defense manufacturing cannot sustain. Anduril's software-defined approach, combined with its additive manufacturing capabilities, positions it to produce systems at the speed of software development rather than the speed of hardware manufacturing. The $250 million DIU contract for "large-scale expeditionary operations" is a validation of this thesis by the most demanding customer in the world.
The bull case also rests on the geopolitical tailwind. Global defense spending is projected to reach $2.5 trillion in 2025, and the share allocated to autonomous systems is growing from a low single-digit base. Even a conservative estimate of Anduril capturing 10-20% of the defense AI market over the next decade implies a revenue trajectory that justifies a $250-300 billion pre-IPO valuation.
The contrarian case that the market is missing is more subtle than the standard bull-bear debate. Both companies may be under-valued relative to the strategic optionality they provide to the fund. Neuralink's technology, if it achieves platform status, has applications that extend far beyond medicine. The same neural interface that allows a paralyzed patient to control a robotic arm could, in principle, allow a soldier to control a drone swarm or a worker to interface with industrial machinery. Anduril's Lattice platform, if it becomes the standard for autonomous defense systems, has obvious applications in border security, critical infrastructure protection, and eventually civilian domains like autonomous logistics.
The portfolio construction, in other words, may be more intelligent than the sum of its parts. MSX has positioned itself at the intersection of two technologies that are individually compelling and jointly transformative.
Takeaway: The Accountability Imperative
The MSX Pre-IPO III portfolio is a bet on a specific vision of the future: one in which artificial intelligence becomes physically embodied, both within the human nervous system and within autonomous systems operating in contested environments. The fund's limited partners are purchasing exposure to that vision, with a 2-3 year holding period and a 3-5x return target.
The risks are not symmetrically distributed. Neuralink's clinical timeline could slip, as FDA approval processes are notoriously unpredictable. Anduril's revenue concentration in the US Department of Defense creates single-customer risk that no amount of international expansion can fully mitigate. The valuations in both cases embed assumptions about execution that the companies have not yet demonstrated.
But the more significant risk is the one that neither the fund nor the companies are likely to articulate. The convergence of human augmentation and autonomous warfare represents the most consequential technological development since the splitting of the atom. The same neural interface technology that could restore function to paralyzed patients could, in a different regulatory and ethical framework, become a tool for surveillance and control. The same autonomous systems that can protect military bases from drone attacks could, in the wrong hands, become instruments of indiscriminate violence.
The question that the pre-IPO market is not asking—but should be—is whether the governance structures of these companies are adequate to the magnitude of the technologies they are developing. Neuralink's handling of animal welfare allegations and its lack of published security research for the N1 implant are not disqualifying, but they are warning signs. Anduril's aggressive lobbying and its positioning at the center of the autonomous weapons debate raise questions about the company's approach to the ethical dimensions of its products.
The market is pricing these companies as if the only risks are execution and market timing. The deeper risks are governance and accountability. A fund that has the sophistication to construct a portfolio thesis spanning the full spectrum of AI's physical manifestation should also have the sophistication to demand transparency from its portfolio companies on the dimensions that will ultimately determine whether these technologies serve human flourishing or undermine it.
The 2-3 year holding period is short relative to the significance of what is being built. The 3-5x return target is modest relative to the stakes involved. The real question is not whether MSX's limited partners will see their projected returns. It is whether the companies they are backing will prove worthy of the trust that the market is placing in them.
Based on my experience auditing the governance structures of high-consequence technology companies, the evidence so far is mixed. Neuralink's technical achievements are real, but the company's approach to security and ethics has been reactive rather than proactive. Anduril's growth is impressive, but the company has not demonstrated that it can navigate the ethical complexities of autonomous warfare with the same competence it has shown in software engineering.
The market will make its judgment in time. The rest of us would be well advised to watch closely, not just the valuations, but the decisions that will determine whether this portfolio represents the future of human capability or the future of human vulnerability.