Retail investors have poured money into quantum computing stocks like IonQ, Rigetti, and D-Wave for one reason: they want to own the next Nvidia. The problem is that quantum computing, as an investment thesis, rests almost entirely on hope rather than engineering. Once you actually map out where algorithmic speedups exist — and where they don't — the case collapses quickly. Meanwhile, the companies most likely to define the next era of compute are all private, and the public markets won't see them until VCs decide the time is right.
The Quantum Algorithm Problem
To understand why quantum is mostly a dead end, start with workloads. AI inference and training are dominated — roughly 99% — by matrix multiplications. For classical silicon, scaling that problem is exponential: you're fighting a 2-to-the-n wall. Quantum offers no meaningful escape from that wall.
The two quantum algorithms that get the most attention outside of cryptography are Grover's algorithm (unstructured search) and HHL (sparse linear systems). Grover's gives you a quadratic speedup — going from n to the square root of n. That sounds useful until you realize a GPU cluster can approximate the same gain, and you still have to account for quantum state preparation time, superposition overhead, and error correction. The juice is not worth the squeeze.
HHL is similarly limited. It's essentially a modified Grover's applied to sparse matrix problems, and the setup costs eat most of the theoretical gain. Neither algorithm addresses the core bottleneck in modern AI workloads.
03:20
Comparison table: clock speed, core count, and wattage across GPU, neuromorphic, and quantum hardware
Watch at 03:20 →
The one genuine quantum advantage is Shor's algorithm for decryption. There, the speedup is dramatic — polynomial versus exponential for classical machines. But decryption workloads represent a vanishingly small fraction of total compute time. It wouldn't rank in the top 100 use cases by compute hours. So the only place quantum wins decisively is also the place nobody is spending meaningful compute budget.
The White House has drawn the same conclusion. Rather than fund quantum computing hardware companies, the current policy is to invest in quantum information science — essentially paying mathematicians and physicists to think about whether any new algorithms might emerge. That's a smart allocation: it costs almost nothing compared to building machines that currently solve no important problems. No good quantum algorithm exists for the workloads that matter. Building the hardware first is backwards.
Why Retail Investors Keep Buying Quantum Anyway
The honest answer is that quantum stocks are a proxy bet on not missing the next compute wave. Investors who remember what Nvidia did to portfolios are looking for the next version of that trade. Since the actual next-generation compute companies — like Unconventional AI, which recently raised $500 million at a $5 billion valuation — are private and accessible only to firms like Andreessen Horowitz, Lux Capital, and DCVC, public market investors are left with whatever the VC ecosystem decides to list, usually at prices designed to benefit the sellers.
Quantum stocks fill that gap psychologically, even if they don't fill it logically. The better framework is to accept that in this cycle, the most asymmetric compute bets are not available to retail investors at formation. The lotto ticket many people are buying in quantum is a ticket to a drawing that probably won't pay out.
What Unconventional Compute Actually Looks Like
The more interesting question is what might genuinely displace or supplement silicon transistors over the next decade. Several approaches are worth watching:
- Biological compute: Companies like Final Spark and Cortical Labs grow neurons on a dish and use biological cells to perform computation. The human brain runs on roughly 20 watts. An Nvidia GPU rack needs closer to 1,000. If biological compute can be made programmable and reliable, the efficiency advantage is enormous. We don't yet know how to program these systems at scale, but the direction is real.
- Optical computing and co-packaged optics: Optical interconnects have been a research focus for decades and have suffered persistent noise and precision problems. Near-term applications are mostly in interconnects rather than core compute, but the space is active. Key public companies in this area — including Lumentum, Coherent, and Fabrinet — trade as interconnect businesses today.
- Analog computing: Analog died commercially in the 1980s, largely due to component imprecision and noise. Some companies are revisiting it with ternary transistors and better error-correction approaches. It remains a long shot, but the term covers a wide range of architectures.
- Ternary and novel transistor designs: At least one company (Ternary) is working on three-state transistors as an alternative to binary silicon logic.
07:45
Overview of unconventional compute categories and representative companies
Watch at 07:45 →
The National Security Dimension
There is a structural argument for accelerating beyond silicon that has nothing to do with raw performance: geography. The overwhelming majority of advanced chip fabrication happens in Taiwan and South Korea — both directly adjacent to China. The concentration of that supply chain is a significant national security vulnerability. Whether the next compute architecture is biological, optical, analog, or something not yet named, reducing dependence on TSMC-adjacent fabs is an independent reason to fund alternatives aggressively.
Silicon transistors have been the dominant compute substrate for over 50 years, and physical scaling limits are now a real constraint. The next 10 years will see serious capital flow into alternatives — not because any single replacement is obvious, but because the incumbent is visibly running out of runway and the geopolitical stakes are high enough that governments will fund the search regardless of near-term commercial viability.
The Investment Takeaway
If you are evaluating compute investments, the framework is straightforward. Ask whether a technology has a genuine algorithmic advantage on the workloads that actually consume compute — primarily large matrix operations for AI. Quantum does not. Biological compute might, eventually, on efficiency grounds. Optical and analog face engineering challenges that are hard but not theoretically impossible.
The companies best positioned to win the next compute cycle are not yet public. By the time they are, much of the return will already have been captured. That is the actual shape of the opportunity — not a quantum lottery ticket, but a patient search for the engineering breakthrough that makes the next substrate manufacturable at scale, preferably on American soil.
