For nearly a decade, Nvidia has been untouchable. The company controls over 90% of the AI accelerator market. Every startup building AI dreams bought Nvidia GPUs. Every cloud provider bowed to their pricing. Every major tech company paid the “Nvidia tax.”
Jensen Huang commanded the industry like a tech emperor, and the market rewarded him with valuations that seemed infinite. But empires built on monopoly power never last. In November 2025, the cracks became impossible to ignore.
Meta is in talks to spend billions on Google’s Tensor Processing Units instead of Nvidia’s chips. Amazon built Trainium chips and shipped 500,000 units to Anthropic. Microsoft designed Maia.
One by one, the giants started asking the same question. Why are we locked into Nvidia when we could own our own silicon? The answer came in the form of a reality check that wiped $115 billion off Nvidia’s market value in a single week.
When Market Dominance Becomes A Liability
Nvidia did not get here by accident. The company spent years building CUDA, its software ecosystem, creating what looked like an unbreakable moat. Every AI engineer has learned CUDA.
Every framework is optimised for CUDA. Every library supported CUDA. Developers faced a choice: switch to an alternative chip and rewrite everything, or stick with Nvidia. The math was simple. Nvidia won.
But that same lock-in that made Nvidia invincible is now making it invisible to the giants who matter most. Meta is not some random startup—Meta plans to spend up to $72 billion on AI infrastructure this year.
That’s not a rounding error. That’s a fundamental business decision by one of Nvidia’s largest customers that their current supplier is no longer acceptable. When the giants start looking elsewhere, the spell breaks.
The problem is not that Google’s TPUs are better. They are not. Nvidia is more flexible, more versatile, and can run any AI model anywhere computing happens.
But that flexibility comes with a cost premium. The average data center chip costs $20,000 to $35,000 per unit. Google can produce TPUs at a fraction of that price.
Credible analysis suggests Google pays roughly 20% of what big companies pay for Nvidia hardware. That is not a niche advantage. That is a structural cost difference that becomes impossible to ignore as AI spending reaches astronomical levels.
The Bubble Whispers Are Getting Louder
There is something else happening beneath the surface. The entire AI industry is showing signs of strain. Experts from every corner of the financial world are whispering about an AI bubble.
The S&P 500 is trading at 23 times forward earnings while the broader FTSE trades at 14 times. Thirty percent of the S&P 500 is held up by just five companies. That is the greatest concentration in half a century.
This mirrors the dot-com crash in structure and tension. OpenAI tripled its valuation from $157 billion to $500 billion in a single year. Startups with negative revenue and uncertain paths to profit are raising billions.
The Bank of England issued an official warning about overvaluation in AI firms. When central banks start sounding alarms, the party is ending. Nvidia’s stock is a bet on infinite AI spending growth.
If that bet falters, if companies start realizing they spent too much on too little, Nvidia falls the hardest. The Meta deal is not the real threat to Nvidia. The real threat is that companies are starting to think clearly.
They are asking themselves if they really need Nvidia, or if they just assumed they did. That question, once asked, cannot be unasked.
Breaking Free From The Monopoly Prison
What Google understood, and what other giants finally recognized, is that the AI era was supposed to unlock innovation and reduce dependency. Instead, it created a new bottleneck. Nvidia became the gatekeeper.
Every company building AI had to go through Nvidia. Every data centre had to be built around Nvidia. Every budget had to include the Nvidia premium. The most profitable move is to build your own silicon.
This is not new in technology. AWS built Trainium and Graviton because they recognised relying on Intel was a strategic vulnerability. Microsoft pursued Maia for the same reason. Google pursued TPUs for the same reason.
The pattern is always the same. As long as a supplier works with you, you benefit from the partnership. The moment the partnership becomes a dependency, you start building alternatives.
Nvidia’s response has been to retreat to the classic argument: their platform is a generation ahead. They still supply Google. They still run every AI model. Their Blackwell architecture is the best.
All of these things might be true, but they sound increasingly desperate. If Nvidia was so far ahead, why would Meta consider switching away? If the technology gap were truly generational, why would price matter at all?
The answer is that neither of these arguments holds water anymore. Good enough and cheaper wins in infrastructure. It always does.
The Commoditization Has Begun
What is really happening is commoditization. In infrastructure markets, once multiple viable alternatives exist, price becomes the primary variable. Nvidia had no real alternatives for years.
Now it has several. Google has TPUs at a quarter of the cost. AMD is pushing harder. Every hyperscaler has in-house options or is building them. The competitive advantage Nvidia built for a decade is evaporating in months.
The irony is that Nvidia built a platform so good that it became the standard everyone wanted to escape. Any company large enough to control their own destiny will now evaluate the ROI for building custom silicon.
For Meta, for Amazon, for Microsoft, the math probably works. They will own their chips. They will reduce cost. They will reduce dependency. They will win.
Smaller companies and startups will stick with Nvidia. There will always be a market for general-purpose accelerators. But the margin is coming out of the high-value deals.
The contracts that drove Nvidia’s stock to the moon are getting shopped around. Some will go to Google. Some will go to in-house silicon. Some will be split between multiple suppliers.
Nvidia is not going away. But the emperor is no longer wearing clothes. The company will remain powerful and profitable. But the era of unchallenged dominance is over.
The house that Jensen built cracked when the giants realized they could build their own.
When Overpriced Meets Over-Hyped
There is one final layer to this story that nobody is discussing directly but everyone is thinking about. Nvidia’s stock is not just a bet on AI hardware dominance. It is a bet on AI itself, justifying the most aggressive valuations in market history.
The company’s stock price assumes AI capex will keep climbing forever. It assumes every company on Earth will spend hundreds of billions on these chips. It assumes Nvidia will capture most of the value.
But what if the AI boom is not quite as infinite as people think? What if companies start asking whether 5% of revenue spent on AI infrastructure is really optimal? What if the return on AI projects starts mattering again?
The first sign that the market is asking these questions would be a major customer reducing orders or diversifying suppliers. That is exactly what is happening now.
Meta’s flirtation with Google chips is not a vote of no confidence in Nvidia specifically. It is a vote of no confidence in the narrative that drove Nvidia to a $3 trillion valuation.
That narrative was that AI is so important, so transformative, that cost does not matter. But cost always matters at scale. Always.
The question now is how many other giants will follow Meta down this path. If it is just Meta, Nvidia survives and thrives. If it is Amazon, Microsoft, Google, Meta, and Apple all building their own silicon… Nvidia has a scaling problem.
Once you have a scaling problem in infrastructure, you have a real problem. The cracks are showing. The emperor just realised the crowd had got smaller. The competitors are no longer intimidated.