One of the biggest expectations surrounding quantum computers is that they may one day train AI models significantly faster than classical systems. This naturally raises an important question: what exactly makes quantum computers potentially faster, and can they truly disrupt the way large AI models are trained? To understand this, we first need to look at what goes into training a large language model (LLM) and then examine how quantum computing compares to classical computing methods. LLM Training Consider a modern open model like Llama 3.1, released in 2024, which contains hundreds of billions of parameters and was pre-trained on trillions of tokens. Training such a model using just a single GPU would theoretically take thousands of years. To make this feasible, companies rely on massive clusters of GPUs working together. For example, dedicating around 16,000 GPUs to the same training task can reduce the time from thousands of years to just a few months. This approach is known as parallel computing, where a massive computational workload is divided into smaller distributed tasks across thousands of processors. Parallel computing is currently the backbone of modern AI training and is the primary reason we can build such large models within realistic timeframes. The reason training these models takes so long in the first place lies in the nature of the computations involved. Most AI systems today, especially neural network-based architectures, rely heavily on matrix multiplication. This process involves multiplying large grids of numbers repeatedly during training. These calculations are measured in floating-point operations, or FLOPs, and training a frontier-scale model can require an astronomical number of operations—on the order of 10^25 FLOPs. Any new computing paradigm, including quantum computing, must somehow compete with or outperform this massive computational demand. A common assumption is that quantum computers could simply perform all these mathematical operations simultaneously. This idea stems from the perceived evolution of computing hardware from CPUs to GPUs, and potentially from GPUs to quantum processing units (QPUs). When CPUs reached their limits for heavy computation, GPUs emerged as specialized hardware capable of handling parallel tasks far more efficiently. Since both CPUs and GPUs are transistor-based and operate on binary data (zeros and ones), this transition was relatively seamless. However, the shift from GPUs to quantum computers is fundamentally different. Classical computers store and process information deterministically in binary form. In contrast, quantum computers use qubits, which can exist in probabilistic states—representing both zero and one simultaneously through superposition. This means that moving computation from GPUs to QPUs is not a simple hardware upgrade but a complete shift in how information is represented and processed. The tokens and numerical data used to train AI models, which are currently encoded in binary form, would need to be translated into quantum states. Encoding deterministic classical data into probabilistic quantum systems is not straightforward and presents a major technical challenge. Nvidia might be to store classical binary data directly in qubits as definite zeros or ones. While this is technically possible, it defeats the purpose of quantum computing. The real advantage of quantum systems lies in leveraging superposition and entanglement, not in mimicking classical deterministic behavior. Forcing quantum computers to operate like classical machines would eliminate their potential benefits. Meanwhile, classical parallel computing has already proven to be extremely effective, as demonstrated by the dramatic reduction in training time when scaling GPU clusters. This leads to a deeper and more practical question: what are quantum computers actually good for in the context of AI? Researchers and scientists are still actively exploring where quantum computing fits within AI innovation. Trying to directly compete with well-established parallel computing methods for training AI models may not be the most logical target. An analogy can be drawn from the CPU-to-GPU transition. Even though GPUs are vastly superior for heavy parallel tasks, CPUs are still used for everyday activities like browsing the internet, handling spreadsheets, and managing basic system operations. Not every task requires the power of a GPU, and the same logic may apply to quantum computers. Do We Even Need Quantum Computers for LLM Training? In a similar way, it is worth asking whether quantum computers truly need to take on the burden of training large language models when existing GPU-based parallel systems already perform the task effectively. This skepticism has fueled ongoing debates about whether the industry is aiming at the right problem by trying to apply quantum computing directly to AI training. Instead of replacing GPUs, quantum computers may eventually work alongside them in a complementary manner. The economic reality of the quantum computing industry also plays a significant role in the hype surrounding it. For any technology to succeed, a strong market must exist, as market demand drives investment, innovation, and competition. Currently, AI represents one of the largest and fastest-growing technology markets, attracting massive capital. Some of this capital could potentially flow into quantum computing research, especially if quantum technologies demonstrate synergy with existing AI hardware rather than attempting to make GPUs obsolete. Industry leaders have even suggested that future computing paradigms could surpass today’s dominant technologies. Just as GPUs eventually overtook CPUs in many high-performance computing tasks, quantum processors may one day reshape the hardware landscape. However, investing in quantum computing remains risky due to its highly experimental nature. Unlike GPUs, which found early adoption through industries like gaming, quantum computing is still searching for its “killer application” that can justify large-scale commercial deployment. The Challenge of Scaling Quantum Computers Another challenge lies in how scaling works in quantum computing. In classical computing, progress has historically followed Moore’s Law, where transistor density doubles roughly every two years. However, this model does not directly translate to quantum systems. Even the most advanced quantum laboratories currently operate machines with only hundreds of qubits. Following a Moore’s Law-like trajectory would still leave quantum computers far below the scale needed for large AI training by the next decade. Moreover, current quantum computers are highly error-prone. While classical computers require near-perfect determinism, quantum systems are inherently probabilistic and extremely sensitive to noise and environmental disturbances. Issues such as short coherence time, limited scalability, and imperfect gate fidelity make reliable large-scale computation extremely difficult. As a result, even though a machine with a hundred qubits may sound powerful, it is not yet practical for training large AI models. Different Approaches to Building Quantum Hardware Different companies and research labs are experimenting with various hardware approaches, including trapped ions, superconducting qubits, and quantum annealing, each attempting to overcome key limitations in stability and scalability. Alongside hardware innovation, entirely new algorithms and quantum gates are being developed to better harness quantum state space. This highlights that quantum computing is not just about faster hardware, but about rethinking computation itself. Will GPUs Continue to Dominate AI Training? The relationship between quantum computing and AI remains an open question. Rather than immediately replacing GPUs in training large language models, quantum computers may initially find value in niche optimization problems, new algorithmic frameworks, or entirely different industries. It is also possible that future AI systems will continue to run primarily on GPUs, just as everyday applications still rely on CPUs today. Large language models may eventually be seen as specialized tools for language processing, much like how spreadsheets specialize in numerical computation. What remains clear is that quantum computing holds enormous potential, but its true role in AI has yet to be fully defined. Instead of viewing it as a direct competitor to classical computing, it may be more realistic to see it as a complementary technology that unlocks new forms of computation that classical systems cannot achieve. The future may not be about replacing existing hardware, but about discovering entirely new ways of processing knowledge that were previously unimaginable with classical computers. You can also read; Google New Quantum Chip Willow and The Parallel Universe Post navigation Apple’s 4 Step AI Strategy To Win AI Race vs Google vs Microsoft