In 2019, Nvidia was worth around $60 billion.
By mid-2024, it had crossed $3 trillion — briefly overtaking Microsoft and Apple to become the most valuable company on Earth.
That is not a typo. A company that most people outside gaming and tech had barely heard of five years ago became the most valuable business in human history in roughly the time it takes to complete a university degree.
How does something like that happen? What does Nvidia actually make? Why is it suddenly at the centre of every conversation about artificial intelligence, global supply chains, and the future of the economy?
This is the story of how Nvidia went from selling graphics cards to gamers to becoming the company that the entire world is fighting to buy chips from — and what it means for the global economy.
What Nvidia Actually Makes
Nvidia makes semiconductors. Specifically, it designs a type of chip called a GPU — a Graphics Processing Unit.
GPUs were originally invented for one purpose: rendering the complex, rapidly-moving visuals in video games. A modern video game requires billions of mathematical calculations per second to display realistic lighting, shadows, textures, and movement. Traditional CPUs — the general-purpose chips that run most computing — are not designed for this kind of parallel, high-volume number-crunching. GPUs are.
Nvidia did not invent the GPU. But it perfected it. Through the 1990s and 2000s, Nvidia built an increasingly dominant position in the gaming graphics market, competing primarily with AMD and (at various points) Intel.
Then two things happened that changed everything.
The Accidental Discovery That Changed Nvidia’s Destiny
Around 2007 and 2008, researchers at universities began noticing something unexpected: Nvidia’s GPUs were extraordinarily useful for scientific computing.
The same parallel processing architecture that made GPUs great at rendering game graphics also made them ideal for any task that required doing millions of similar calculations simultaneously — which, it turned out, described an enormous range of scientific and research workloads.
Nvidia saw this and made a pivotal decision. In 2006, it released CUDA — a programming platform that allowed developers and researchers to harness GPU power for general computing tasks beyond graphics.
CUDA was not an immediate commercial success. It was a long-term bet. Nvidia was essentially saying: we think GPUs will be useful for computing broadly, not just gaming, and we are going to build an ecosystem around that belief before anyone else does.
For years, it was a niche bet. Then artificial intelligence changed everything.
Why AI Runs on Nvidia
Training an AI model — whether it is a language model like the ones behind ChatGPT or an image recognition system — requires processing incomprehensibly large datasets through complex mathematical operations billions of times.
This is, at its core, exactly what GPUs are designed to do.
When deep learning research began accelerating in the early 2010s, AI researchers discovered that training their models on Nvidia GPUs was orders of magnitude faster than training on traditional CPUs. A task that might take weeks on a CPU cluster could be done in hours on a GPU cluster.
And because Nvidia had spent years building the CUDA ecosystem — the software, the tools, the developer community, the libraries — the entire AI research world had built its workflows on top of Nvidia’s infrastructure.
This created something economists call a switching cost moat. Not just a product advantage but an ecosystem advantage. By the time companies like Google, Meta, OpenAI, and Microsoft needed enormous amounts of GPU computing to train their AI models, the entire infrastructure of AI research was already running on Nvidia hardware and software.
Switching to a competitor would not just mean buying different chips. It would mean rewriting years of software, retraining teams, and rebuilding an entire development stack. The cost was prohibitive.
Nvidia did not just build a better product. It built the road that everyone else’s cars were already driving on.
The Product That Powered the AI Boom: The H100
When ChatGPT launched in November 2022 and triggered the current AI investment frenzy, one product sat at the centre of the spending that followed: the Nvidia H100 GPU.
The H100 is not a consumer product. You cannot buy one at a technology store. A single H100 chip costs between $25,000 and $40,000. Training a large AI model requires thousands of them, connected in specialised clusters that can cost hundreds of millions of dollars to build and operate.
Every major technology company — OpenAI, Google, Meta, Amazon, Microsoft, and dozens of others — scrambled to acquire as many H100s as they could. At the peak of the frenzy in 2023, the H100 had a waiting list measured in months. Companies were reportedly paying significant premiums on secondary markets just to get chips sooner.
The economic logic was clear: whoever had more computing power could train better AI models faster. Computing power was the limiting factor in the AI race. And Nvidia was the only meaningful supplier of that computing power at the required scale.
How Nvidia Makes Money
Understanding Nvidia’s business model requires understanding that it does not actually manufacture its own chips.
Nvidia is what is called a fabless semiconductor company. It designs chips but outsources their physical manufacture to foundries — most importantly, TSMC (Taiwan Semiconductor Manufacturing Company), the Taiwanese company that manufactures the most advanced chips in the world.
This model has significant advantages. Nvidia does not need to invest tens of billions of dollars in manufacturing facilities. It can focus entirely on chip design — where its expertise lies — and let TSMC handle the extraordinarily complex and capital-intensive process of actually building the chips.
Nvidia’s revenue comes from selling its designed chips to customers. Its business is divided into several segments:
Data Centre (the dominant segment): Sales of GPUs to cloud providers, AI companies, research institutions, and enterprises for AI training and inference. This segment grew from around $15 billion in 2022 to over $47 billion in 2023 to an annualised rate exceeding $100 billion in 2024. This is the segment that drove Nvidia’s valuation explosion.
Gaming: Nvidia’s original business. GeForce GPUs for consumer gaming PCs. Still significant — around $10-15 billion annually — but now a secondary business compared to data centre.
Professional Visualisation: High-end GPUs for architects, designers, film studios, and engineers. Smaller but high-margin.
Automotive: Nvidia’s Drive platform for autonomous vehicles and advanced driver assistance systems. Growing but not yet a major revenue contributor.
OEM and Other: Chips for laptops and other devices using Nvidia integrated graphics.
The data centre segment now dominates. Nvidia’s gross margins on data centre chips exceed 70% — meaning for every £100 of data centre revenue, Nvidia keeps over £70 after the cost of having the chips manufactured. These are extraordinary economics for a hardware company.
The Geopolitical Dimension
Nvidia’s rise has introduced it into one of the most consequential geopolitical conflicts of our time: the competition between the United States and China over AI leadership.
In 2022 and 2023, the US government imposed increasingly strict export controls on advanced semiconductor technology, specifically restricting the sale of Nvidia’s most powerful chips — the H100 and its successors — to China.
The rationale was strategic: advanced AI chips are dual-use technology. They can train AI systems for commercial applications but also for military applications, surveillance, and weapons development. The US government decided that allowing China unrestricted access to the world’s most powerful AI chips was a national security risk.
For Nvidia, this created a complex situation. China had been a significant market — contributing around 20-25% of data centre revenues before the restrictions. Losing access to that market was a real financial hit.
Nvidia’s response was to design export-compliant chips for China — stripped-down versions of its flagship products that met the regulatory requirements while still being commercially useful. Whether this workaround will survive further tightening of restrictions is an ongoing uncertainty that every Nvidia investor watches closely.
The Competition Trying to Catch Up
Nvidia’s dominance in AI chips is not uncontested. Several serious challengers are spending billions trying to break its grip.
AMD has produced competitive AI chips — the MI300X series — and has signed significant deals with Microsoft, Meta, and others. AMD is the most credible near-term competitor, but still commands only a small fraction of the AI chip market.
Google has developed its own AI chips — Tensor Processing Units (TPUs) — which it uses for internal AI workloads and offers through Google Cloud. They are highly optimised for Google’s specific AI work but not available as standalone products.
Amazon has developed the Trainium and Inferentia chips for AI training and inference on AWS. Like Google’s TPUs, these are internal-first but sold through cloud services.
Intel has been trying to re-enter the high-performance computing and AI chip market but has struggled to execute at the level required to challenge Nvidia’s leadership.
Startups — Cerebras, Groq, SambaNova, and others — are building specialised AI chips with specific performance advantages. None have yet achieved the scale or ecosystem depth to challenge Nvidia broadly.
The common challenge for all of them: CUDA. Nvidia’s software ecosystem, built over nearly two decades, is so deeply embedded in AI research and deployment that competing on hardware alone is insufficient. Any serious Nvidia challenger needs to build a software ecosystem compelling enough to pull developers away from a platform they have spent years mastering.
That is a formidably difficult task. Not impossible — but formidably difficult.
What Nvidia’s Rise Tells Us About the Economy
Nvidia’s story is about more than one company’s success. It is a case study in several of the most important forces shaping the global economy right now.
The winner-takes-most dynamics of platform businesses. Nvidia did not just win the AI chip market. It built a platform — CUDA, the developer ecosystem, the software stack — that makes its position self-reinforcing. The more developers build on CUDA, the more valuable CUDA becomes, the harder it is for competitors to attract those developers away.
The chokehold of semiconductor manufacturing. Everything in the AI economy runs on chips. Chips are extraordinarily difficult to manufacture. Only a handful of companies — TSMC chief among them — can manufacture the most advanced chips. This concentration of manufacturing capability is one of the most significant strategic vulnerabilities in the global technology supply chain.
The weaponisation of technology supply chains. The US export controls on Nvidia chips represent a new kind of economic statecraft — using technology access as a geopolitical lever. This is the opening move in what will likely be a long and complex competition over who controls the foundational infrastructure of the AI economy.
The speed of value creation in technology. A $60 billion company became a $3 trillion company in five years because it happened to own the infrastructure that the most important technological transition in decades required. This kind of value creation — and the inequality it creates — will be one of the defining economic stories of the coming decade.
What This Means for Ordinary Investors
Nvidia’s rise has made it a dominant holding in most global stock indices. If you hold an S&P 500 index fund or a global equity ETF, you almost certainly own Nvidia — its weighting in the S&P 500 has, at peak, exceeded 6%.
For investors, the key questions now are:
Can Nvidia maintain its dominance? The competitive dynamics described above suggest yes, in the medium term. The CUDA moat is real. But competition is intensifying and the pace of change in AI hardware is rapid.
Is the valuation justified? At its peak valuation, Nvidia was priced at extraordinary multiples of earnings — a bet that AI infrastructure spending would continue at the current pace for years. If AI investment slows, or if a competitor achieves meaningful ecosystem adoption, the valuation could compress significantly.
What happens if US-China tensions escalate further? Additional export restrictions could cut Nvidia off from a significant global market. This is a real risk that is difficult to quantify.
None of this is investment advice. It is context — the kind of context needed to understand what you own in your index fund, why financial headlines keep mentioning Nvidia, and why a chip company became the most valuable business on Earth.
The Bottom Line
Nvidia’s ascent from gaming graphics company to AI infrastructure monopolist is one of the most remarkable business stories of the modern era.
It was built on a 2006 software bet called CUDA that nobody paid much attention to for a decade. It was accelerated by the AI boom that nobody predicted at the scale it arrived. It was protected by an ecosystem moat that competitors have spent billions trying to breach and largely failed.
Whether Nvidia maintains its position or eventually faces a serious challenger, its rise has already reshaped the global technology economy, ignited a geopolitical contest over semiconductor supply chains, and made a handful of people extraordinarily wealthy.
Understanding how it happened is not just business history. It is a window into how the global economy is being reorganised around artificial intelligence — and who the winners and losers of that reorganisation are likely to be.
Eueezo covers global business and finance in plain English — no jargon, no paywalls. Subscribe to our weekly briefing below.
