For as long as humans have been asking questions, we’ve also been trying to see what comes next. From reading tea leaves to studying the stars, the idea of knowing the future has held certain power and allure for millennia.
Simulations have long fascinated popular culture. In Greg Egan’s Permutation City, entire universes are computed from cellular automata and consciousness persists as software across infinite iterations, demonstrating the raw computational power to rewrite reality itself. And in Westworld, Bernard Lowe wields predictive engines to run millions of branching timelines, simulating every possible decision to anticipate and shape better outcomes.
But what if that power of foresight wasn’t pure fiction. Because you might not realise it, but advances in AI – specifically AI-accelerated simulation – are creating something that looks uncannily like it.
Simulating fiction – until now
Traditionally, if we were to take what Bernard Lowe does in Westworld – running millions of branching simulations to explore possible outcomes – and attempt to replicate it with real-world computational resources, it would require massive infrastructure and could take days, weeks, or even months.
The only way to know a simulation’s outcome was to run it. This high-fidelity modeling has long been essential in fields like climate science, manufacturing, drug development, and particle physics. Yet these simulations often demand vast computational resources, consuming time and energy on massive high-performance computing (HPC) clusters before generating a single usable result.
High-fidelity modelling at ENEA’s new CRESCO8 supercomputer illustrates just how demanding modern science has become. Before any physical test is run inside a fusion research environment, ENEA researchers use CRESCO8 to generate detailed simulations of how superheated plasma will behave under different magnetic confinement conditions. These models predict everything from turbulence patterns to plasma–wall interactions, allowing scientists to compare predicted behaviour with real-world test results and refine reactor designs far more safely and efficiently. Until recently, running a single high-resolution plasma simulation could take many hours of computation across thousands of parallel cores, consuming significant energy simply to model a few milliseconds of physical reality.
A fundamental shift in scientific capability
With CRESCO8, ENEA can now integrate AI-accelerated modelling techniques directly into its fusion research workflow. Lightweight neural models trained on traditional physics-based simulations can predict plasma behaviour or turbulence evolution almost instantly, acting as “AI solvers” that replicate the end state of large-scale simulations without the heavy computational load. Tasks that once demanded sprawling HPC runs can now be executed on a fraction of the hardware, often on a single GPU, radically reducing both energy consumption and turnaround time while preserving scientific accuracy.
Similarly, at The European Organization for Nuclear Research, known as CERN, the use of generative adversarial models (GANs) in their particle-physics research has already shown that AI can replicate collision-event images that once required vast HPC clusters and hours of processing in a matter of seconds.
For both ENEA and CERN, the impact is transformative: simulation workflows that were once bottlenecks are becoming real-time tools, unlocking faster experimentation, cleaner energy use, and a new era of scientific agility.
Modeling entire galaxies in weeks
This transformation is not limited to particle physics and the field of nuclear fusion, with astrophysics researchers now experiencing similar leaps in simulation speed and fidelity thanks to advancements in HPC infrastructure.
Modeling cosmic evolution requires tracking processes that unfold across staggering timescales. A star takes several hundreds of thousands of years to form, and galaxies can take several tens of millions of years to revolve around themselves. But, as Dr Ana Duarte Cabral, a Royal Society University Research Fellow working out of the Cardiff Hub for Astrophysics Research and Technology says, we are now able “to create a model of an entire galaxy, tracking the formation and death of generations of stars, in only a matter of weeks. Before, the simulations took more than three times longer to run.”
At Cardiff University, improved systems driven by cutting-edge server architectures have doubled the available compute capacity and exceeded initial performance benchmarks by 46%. This allows researchers to process gravitational wave detection events and share data with the global astronomy community significantly faster.
The result is another form of ‘future-seeing’: an ability to explore cosmic evolution at a pace that is no longer measured in months or years, but in weeks. Researchers can test hypotheses about stellar formation, black hole interactions, and galactic dynamics with a level of speed and scale that was once inconceivable.
The implications of these developments extend far beyond efficiency. When a simulation takes 24 hours, researchers might explore a handful of scenarios. When it takes seconds – or when an entire galaxy can be modeled in weeks instead of months – they can explore thousands. That ability to examine millions of possible outcomes before conducting a real-world experiment gives researchers something uncanny: a way to compute the future rather than wait for it.
From the cosmos, back down to Earth
While cosmic modeling may feel distant from everyday experience, the same AI-accelerated approaches are entering domains that shape our daily lives too.
- In pharmaceuticals, AI solvers dramatically speed the modeling of molecular interactions, supporting rapid new drug development and making research into rare diseases economically viable.
- In transportation, autonomous systems gain the capacity to train on millions of synthetic driving scenarios – every hazard or weather pattern – creating a reservoir of ‘experience’ no human driver could ever accumulate.
- In engineering and manufacturing, design teams can explore performance trade-offs in minutes, accelerating innovation and reducing the need for physical prototyping.
Not magic – but is this the closest we’ve come?
We still can’t literally predict the future. But we are entering an era where scientists and industries can compute the most probable future, with unprecedented speed and accuracy.
AI solvers don’t replace HPC, they elevate it. They compress the wisdom of countless simulations into models capable of delivering answers in seconds. This blend of high-performance computing and high-performance intelligence gives researchers something humanity has chased for millennia: a clearer view of what lies ahead.
It isn’t magic, but it might be the closest we’ve come yet.
Dr. Valerio Rizzo
Dr. Valerio Rizzo is EMEA Head of AI and SME at Lenovo. A Tech Evangelist with over 10 years’ experience leading R&D initiatives and delivering advanced IT solutions for global organisations, Dr. Rizzo is a trusted advisor in AI, ML and immersive technologies, with a strong background in neuroscience research and a proven ability to bridge customer needs, innovation, and business strategy.
