The Diss Track That Broke the Seal
In April 2024, Drake released a diss track aimed at Kendrick Lamar that featured something unprecedented in hip-hop beef: the voice of Tupac Shakur, dead for nearly three decades, rapping new bars written for a fight he could never have witnessed . The track, 'Taylor Made Freestyle', also summoned a synthetic Snoop Dogg, and within hours the internet was ablaze—not merely with the usual hip-hop drama, but with a more fundamental unease. Tupac's estate threatened legal action . Fans debated authenticity, consent, the boundary between homage and desecration. But the genie, as they say, was out of the bottle. A tool that had been the province of research labs and tech demos had crossed into the realm of popular culture, wielded by one of the world's most famous artists in one of music's most public feuds.
The incident was a fracture line, a moment when the abstract promises and warnings about artificial intelligence became suddenly, jarringly concrete. If Drake could summon the dead to settle scores, what else had become possible? And how had we arrived here, at a juncture where the technology to fabricate reality—voices, faces, entire videos—had become accessible enough for a rapper's PR team to deploy on a Friday night?
The answer lies in an extraordinary compression of technological development that has unfolded over the past five years, a period in which AI has evolved from a specialist concern into the organising principle of the global economy, a driver of geopolitical competition, and a force reshaping everything from drug discovery to disinformation. This is the story of the boom: how it happened, what it has wrought, and why the people building it are themselves increasingly uncertain about what comes next.
The Unsolved Problem
To understand the acceleration, you must first grasp the breakthrough. For half a century, one of biology's grand challenges had resisted solution: the protein-folding problem. Proteins are the molecular machines of life, chains of amino acids that twist and fold into precise three-dimensional shapes, and those shapes determine their function. Knowing a protein's structure is essential to understanding disease, designing drugs, grasping the mechanics of life itself. But predicting how a given sequence of amino acids will fold had defied researchers since the problem was first articulated in 1972.
Then, in 2020, DeepMind's AlphaFold solved it . The AI system could predict protein structures with an accuracy that stunned the scientific establishment, essentially cracking a puzzle that had consumed entire careers . It was a watershed: not merely an incremental improvement but a category-shift in what machines could do. AlphaFold demonstrated that AI could tackle problems requiring not brute-force calculation but something closer to intuition, pattern recognition at a scale and subtlety beyond human capacity.
The significance extended beyond biology. If AI could solve protein folding, what other intractable problems might yield? The breakthrough arrived at a moment when the underlying technology—deep learning, neural networks trained on vast datasets—was maturing rapidly. The same architectural principles that powered AlphaFold were being applied to language, to images, to sound. The laboratory, for decades a place of careful incremental progress, was suddenly producing tools that worked astonishingly well in the wild.
The Generative Explosion
What followed was a cambrian explosion of generative models, systems that could create rather than merely classify or predict. OpenAI introduced Jukebox, an AI that could generate genre-specific music, complete with rudimentary vocals . Google introduced MusicLM, which could generate high-fidelity music from text descriptions . These were not mere curiosities; they were glimpses of a future in which the production of culture—the writing of songs, the making of images—might be partially or wholly automated.
The release of Stable Diffusion by Stability AI marked another inflection point . Here was an image-generation model released not as a controlled experiment but as open-source software, downloadable by anyone. Within weeks, millions were using it to conjure photorealistic images, surreal art, marketing copy, and inevitably, pornography and propaganda. The genie was not merely out of the bottle; it had been duplicated a million times over and handed out for free.
OpenAI's DALL-E 2 followed, producing images far superior to its predecessor . One demonstration showed an astronaut riding a horse—an absurd juxtaposition that nonetheless revealed the system's capacity to understand and combine concepts in ways that suggested genuine semantic comprehension. Then came Sora, a video-generation model that could create startlingly realistic clips from text prompts . Marketers immediately recognised its potential for storytelling, for producing bespoke content at a fraction of traditional costs . Suno, an AI music generator, prepared to release its fourth version, promising dramatically improved vocal quality and production values .
The pace was dizzying. Technologies that would once have taken years to mature were being iterated monthly. And underpinning it all was a single company whose chips had become the sine qua non of the AI age.
The Rise of the Pick-and-Shovel Seller
In June 2024, Nvidia surpassed Microsoft to become the most valuable company in the world . It was a fitting symbol. Nvidia's graphics processing units, originally designed to render video-game graphics, had proven uniquely suited to training the massive neural networks that powered generative AI. Every major AI lab, every tech giant, every startup with ambitions in machine learning needed Nvidia's chips. The company had become the arms dealer of the AI gold rush, and its valuation reflected the scale of the boom.
The financial stakes were staggering. Microsoft had invested billions in OpenAI over several years and was now integrating generative AI across its product suite—Bing, Dynamics 365, productivity tools . The bet was existential: that AI would not merely augment existing software but fundamentally remake how people worked, searched, created. Google, facing an unprecedented competitive threat to its search monopoly, rushed out its own models. DeepMind's Gemini represented the company's attempt to build a system that could match or exceed OpenAI's offerings . Shopify deployed AI agents to analyse complex merchant data and forecast growth at global scale, running multiple subagents in parallel over long time horizons .
The 2026 Stanford AI Index Report documented a sharp increase in AI adoption in medicine , a signal that the technology was moving from experimental to infrastructural, from curiosity to dependency. And the geopolitical dimension was becoming impossible to ignore.
The New Great Game
China led the world in sheer volume of AI patents, with nearly 13,000 granted in 2024 . But the United States dominated in impact: American patents were cited nearly seven times more often, a metric suggesting deeper innovation, more foundational work . The divergence illuminated two distinct strategies. China was pursuing quantity, flooding the zone with applications, deploying AI across state surveillance, social credit, industrial optimisation. America was betting on quality, on breakthrough research from labs like OpenAI, DeepMind, and a constellation of university partnerships.
The competition was not merely economic but ideological. Control over AI meant control over the infrastructure of the 21st century: the algorithms that would mediate information, allocate resources, identify threats, design weapons. Vietnam's parliament passed the country's first AI law, requiring transparency from providers and mandating that AI products be clearly labelled . It was a gesture towards governance, an acknowledgement that the technology had outrun the regulatory frameworks designed for an earlier era.
Yet the historical arc was longer than the headlines suggested. The MIT AI Lab, founded by John McCarthy and Marvin Minsky in 1959 , had been grinding away at these problems for more than six decades. The boom was not a spontaneous eruption but the culmination of sustained investment, theoretical breakthroughs, and—crucially—the availability of two things that earlier researchers lacked: vast datasets scraped from the internet, and the computational power to process them.
The Dark Horizon
But as the technology proliferated, so did the threats. Generative AI was now widely adopted for social engineering, phishing schemes, deepfake scams, and automated disinformation campaigns . The same tools that could compose a symphony or design a drug could fabricate a politician's confession or generate a synthetic child for exploitation. The barrier to entry for malicious actors had collapsed. A sophisticated phishing campaign that once required linguistic skill and cultural knowledge could now be generated in seconds, in any language, tailored to any target.
A 2025 report from CrowdStrike documented the escalation: state-sponsored hackers, criminal syndicates, and lone actors were all using AI to automate and scale attacks . Google's security team warned that generative AI would fundamentally level up cyber threats, enabling adversaries to operate with a speed and sophistication previously unimaginable . The asymmetry was stark: building a robust defence required vast resources and expertise, but launching an attack required little more than an internet connection and a subscription to a generative model.
And the more existential risks loomed larger. Researchers studying AI governance identified a core problem: many actors with the ability to deploy AI had both the incentive and the capacity to do so even if such deployment risked catastrophic harm—and they might do so unintentionally, for a substantial period of time, before the consequences became apparent . The scenario was not far-fetched. Imagine an AI system optimised for engagement inadvertently amplifying conspiracy theories into mass violence, or a financial algorithm triggering a cascade that crashed markets, or a biological research tool misused to engineer a pathogen.
The risk was not merely that someone would build a dangerous AI on purpose, though that was certainly possible. The more insidious danger was that the technology's very power and accessibility created conditions in which catastrophic outcomes could emerge from banal decisions, from the ordinary operation of systems designed to maximise profit or efficiency or user satisfaction . Unsafe development or misuse of AI could cause harms of a scale and kind that humanity had not previously confronted. And the window to establish safeguards was closing as the technology diffused.
The Reckoning
We return, then, to Drake and the ghost of Tupac. The diss track was trivial in one sense—a publicity stunt in a genre long defined by provocation and spectacle. But it was also a perfect distillation of the moment: a demonstration that the tools once locked in research labs were now in the hands of anyone with resources and motive, and that the questions they raised—about authenticity, consent, truth, identity—had no easy answers.
The AI boom has delivered genuine marvels: proteins folded, diseases diagnosed, efficiencies extracted from systems too complex for human oversight. It has also delivered a torrent of synthetic media, a new terrain for geopolitical competition, and an array of threats that evolve faster than our capacity to understand them. The optimists see a future of abundance, in which human creativity is augmented and drudgery automated. The pessimists see a tightening spiral of disinformation, surveillance, and accidental catastrophe.
What is certain is that the boom has ended one era and opened another. The technology is no longer speculative. It is embedded in the economy, the culture, the sinews of daily life. We are all, now, living in the aftermath of decisions made in laboratories and boardrooms over the past five years—decisions about what to build, what to release, what risks to accept in pursuit of competitive advantage.
The question is not whether AI will reshape the world. It already has. The question is whether we can shape the trajectory of that transformation, or whether we will simply be swept along by it, adapting as best we can to a landscape remade by forces we only partially understand and imperfectly control. The boom, in other words, is not the end of the story. It is the beginning of a much longer, more uncertain one, and the opening chapters have been written in extraordinary haste.
"Many actors are able and incentivised to deploy catastrophically harmful AI for a substantial period of time—and they might do so unintentionally."
The future will be determined not by the technology alone but by the choices we make about it: what we permit, what we prohibit, what we decide is worth the risk. Those choices are being made now, in piecemeal fashion, by legislators in Hanoi, executives in Silicon Valley, researchers at DeepMind, and yes, by artists in Los Angeles who summon the voices of the dead to settle scores. The boom has given us power. What we do with it will define the age.