News Summary
Produced with editorial insights.
- Nvidia’s dominance is not just a business success but an ecosystem monopoly challenging even giants like Google and Meta, who struggle to break free from this complex web.
- There is a mysterious saying in the tech market: Nvidia is essentially not a chip manufacturer but a software company deploying hardware illusions to trap the world in its network.
- The key question before you now is: Will we remain mere spectators or become active players? The game has only just begun!
Recently, US President Donald Trump flew prominent Silicon Valley leaders Elon Musk of SpaceX, Tim Cook of Apple, and Jensen Huang of Nvidia on the special “Air Force One” aircraft to Beijing for “high-stakes” talks with Chinese President Xi Jinping, an event that stunned diplomats worldwide.
While it appeared to be a business visit on the surface, internally it was a competition among the world’s powerful leaders. This event made clear that the current global top concern is not tankers or ballistic missiles, but artificial intelligence (AI) and the tiny silicon pieces controlling it—the microchips. These minute chips govern everything from households to banking systems in every nation.
The world’s leading tech giants like Google, Microsoft, Meta, and Amazon lineup at Nvidia’s door for the world’s most advanced AI chips. Nvidia alone controls nearly 90% of the global market for cutting-edge AI chips.
Despite established competitors like AMD and Intel in hardware manufacturing, why are global billionaires eager to procure Nvidia’s chips? Why have other companies failed to develop alternatives to Nvidia’s offerings?
This article aims to clarify this complex global technology market game through examples from our own industry.
Difference Between CPU and GPU: An In-Depth Processor Comparison
To understand the politics in the chip market, it’s essential to grasp the distinction between two fundamental processors inside computers or mobiles — the CPU (Central Processing Unit) and the GPU (Graphics Processing Unit).
Let’s compare this to an orange orchard in our village. Suppose 10,000 oranges must be picked and delivered to the market. We have two options.
First option: A clever, strong porter (i.e., CPU) carries 100 oranges at a time and runs the shorter route. But he can’t carry more than 100 oranges at once, so he must run 100 times to deliver all 10,000 oranges. In technological terms, this is called ‘serial processing.’
Second option: Five hundred school children (i.e., GPU), each carrying a basket, are sent simultaneously via a wide road. Each carries 20 oranges, resulting in all reaching the market simultaneously and completing the task faster. This is ‘parallel processing.’
AI applications like facial recognition or language translation don’t require complex math but rather millions of simple operations done rapidly. Hence, GPUs are millions of times more useful than CPUs in AI. Nvidia realized this back in 2006 and focused fully on GPU development while other companies remained tied to CPUs.
The ‘CUDA’ Software Web
For two decades, Nvidia has concentrated on GPU chip development. Its success and power stem not only from hardware but from its software platform called CUDA (Compute Unified Device Architecture). CUDA teaches programmers a language for direct, optimized communication with GPUs.
CUDA can be thought of as a major highway that all customers and businesses have been using continuously for 20 years. All vehicles, petrol stations, hotels, and other infrastructure have been designed around this road.
Nvidia has strengthened its monopoly by advancing software development dubbed a ‘software moat.’
Although competitors like AMD introduce cheaper new chips, those using them must abandon the CUDA highway, leading to significant financial and technical challenges.
Thus, major tech firms must invest heavily and spend time developing new software alternatives. AI’s main digital platforms like PyTorch and TensorFlow run exclusively on CUDA, creating a ‘software ecosystem lock-in’ that prevents customers from switching to cheaper chip options.
Pre-made Software Libraries and Tools
AI software development doesn’t require rewriting code from scratch. Alongside CUDA, Nvidia has created various AI and tensor optimization software libraries making coding faster and easier.
This is comparable to having a pre-made mix of spices at a momos shop for quick and delicious momos. Competitors might have the utensils but without the spice mix, they must labor to create their own.
New and experienced programmers alike find themselves compelled to stay within Nvidia’s established ecosystem because their investment and time are tied to it.
Old Licenses and New Vehicles
How fast software code runs on hardware depends on the compatibility of components. The CUDA highway aligns perfectly with Nvidia’s wheels and gears, resulting in faster, more fuel-efficient rides.
Trying to use AMD’s vehicle on the CUDA highway results in mismatched wheels and road problems. Programmers face the challenge of constantly rewriting new software for new chips. Nvidia, however, ensures that codes over a decade old run in seconds on modern chips.
This can be likened to old driving licenses still being legally valid for operating today’s electric cars.
All these reasons have enabled Nvidia to eliminate competitors and maintain a monopoly.
Network Management Power: InfiniBand and DGX
To extend AI capabilities, millions of chips must be connected to form supercomputers. When thousands of chips join, the biggest issue is traffic jams—data bottlenecks.
This can be compared to trucks causing a jam on a narrow wooden bridge connecting two villages.
Nvidia acquired a networking company called Melanox and took over the InfiniBand technology, which enables thousands of trucks to move on a modern 24-lane highway without traffic jams at exceptionally high speed.
Nvidia does not just sell chips; it sells complete supercomputer systems called DGX, which package chips, InfiniBand cables, cooling systems, and software together. Competitors lack such extensive mega-networks.
The World’s Most Complex Factory: The Story of TSMC
Though a large company, Nvidia does not manufacture chips itself. It designs and delivers digital blueprints to Taiwan’s TSMC for production. More than 90% of the world’s most advanced AI chips are produced by TSMC.
Why can’t other wealthy nations or companies like Intel establish factories like TSMC? There are primary reasons:
First, enormous capital investment. Opening a modern chip factory costs $15-20 billion, exceeding Nepal’s total annual budget. Its key machine, the EUV lithography machine, costs over $300 million and is made only by a single company, ASML, mostly pre-booked by TSMC.
Second, atomic-level precision. TSMC currently manufactures commercially viable 2-nanometer chips, where even a single dust particle can ruin the chip. The factory’s cleanroom is 10,000 times cleaner than Mount Everest’s summit.
Third, production yield experience. TSMC is an expert factory achieving over 80% successful chip yield, while competitors continue to struggle.
One might think, “What difference does this make for us running Facebook in Kathmandu or Pokhara?” But the reality is much tougher.
Supply Chain Control and Speed
Building chips requires not only silicon but also high-speed memory (HBM), which can be likened to the village’s “grain banks.”
Three years ago, Nvidia paid billions of rupees upfront to secure deals with those who operate these mills and farms, locking down supplies.
Competitors might approach TSMC, but not only is this year’s production booked, future years are fully reserved as well.
Furthermore, Nvidia CEO Jensen Huang follows the “one-year product cycle” strategy, releasing new chips every year. Before the market fully absorbs old chips, a new generation is launched.
Opportunities for Nepal
Whether it’s the US political strategy of flying technology tycoons on Air Force One to Beijing or Nvidia’s software web, the message is clear: future conflicts will not be over territorial control but digital colonization where control over thought and data prevails.
Nvidia’s dominance is not just business success; this monopoly has even challenged giants like Google and Meta. As chip prices rise, our AI technology, educational tools, and mobile apps will become more expensive.
Unknowingly, we are becoming subjects of an invisible digital empire ruled by Jensen Huang.
But is it impossible to break this web? While we cannot build a hardware factory with billions of rupees, Nvidia has kept its AI software running openly on the internet on the ‘CUDA highway.’
This means dominating the tech world doesn’t require being a superpower. Sitting in an ordinary room in Kathmandu with a simple laptop, one can use Nvidia’s network to build AI models that could impact the global technology market.
In conclusion, the mystery of the tech market is this: “Nvidia is essentially not a chip manufacturer but a software company that lures the world with hardware illusions into its net.”
The question is clear: will we be spectators or players? The game has begun!
