- TheTechOasis
- Posts
- Is the AI Bubble Bursting?
Is the AI Bubble Bursting?
![](https://media.beehiiv.com/cdn-cgi/image/fit=scale-down,format=auto,onerror=redirect,quality=80/uploads/asset/file/cd2cd8f4-45ca-4e20-99ee-c85fff15f7b2/image.png?t=1721466041)
MARKETS & GEOPOLITICS
Is the AI Bubble Finally Crashing?
As you know, this newsletter is forward-looking. We analyze the present to strategize about the future and make better decisions.
Recently, many of the world’s most important stocks crashed, making investors ask if the recent crash in chip companies and AI stocks is a wake-up call to the fragility of their valuations, which, as we covered, are partly based on hype rather than reality in the latter case.
The explanations provided by news and ‘experts’ have been, mildly put, only scratching the surface of what is really going on, as latent in this crash is the possibility of this being the beginning of a great market debacle.
Therefore, today, we are uncovering the true reasons the markets are so volatile and, importantly, how you should approach the valuations of many of the most important corporations, alongside some hidden opportunities and risks.
A Problem Going Back Decades
Words spoken by the right—or wrong—people can carry much weight, and this time, the impact was very negative.
Huge Sell-Offs
This is the summary of recent events in the markets concerning stocks from semiconductors and AI stocks:
NVIDIA fell 8%, losing its $3 trillion value
Microsoft 4%, although we need to factor in the Crowdstrike mess, too
Apple fell 5%
ASML, a chip manufacturing equipment EUV company, fell 18%
Applied Materials, another equipment and software provider for chip manufacturing, fell 14%
TSMC, the global foundry of advanced chips, 12%
And so forth, combined for a total loss of $1.1 trillion, with $500 billion wiped on Wednesday alone. But why?
Although there’s much more than meets the eye, as you’ll see in a minute, the shortest answer publicly provided is that the symbiosis between Joe Biden and Donald Trump’s comments caused the massacre.
On the one hand, a Bloomberg report indicated that Biden was seriously considering a tougher stance on companies exporting chip manufacturing equipment or GPU cloud services to China.
As for the former, the US is obsessed with preventing China from becoming a powerhouse chip manufacturer.
Regarding the latter, the US Government felt that the penalties they were enforcing on companies like NVIDIA to prevent China from acquiring chips and GPUs were being sidestepped by the Land of the Red Dragon thanks to US hyperscalers providing GPU services to Chinese companies.
Biden’s stance is extremely geopolitical and much less economical; China manufacturing chips would cause a massive deflationary effect due to increased competition and supply increase, but could be devastating to the geopolitical interests of the US in the region.
On the other hand, Trump’s comments have been directed more toward Taiwan, scaring the living shit out of investors… and the aforementioned country itself.
Specifically, Trump said the following:
"I know the people very well and respect them greatly. They did take about 100% of our chip business. I think Taiwan should pay us for defense." He’s implicitly referring to how Taiwanese manufacturers, high on Barbie and TV manufacturing in the 1980s, went to the US to learn as much as possible about the chip manufacturing process and took the entire industry back to Taiwan.
He also said, "You know, we're no different than an insurance company. Taiwan doesn't give us anything." This is a lie, as Taiwan, an ally of the US, plays a crucial role in ensuring US relevance in East Asia as a blocker of China's expansion.
Although undeniably clumsy, some have taken its comments too far to the extreme.
But what is this all about?
Taiwan is a strategic ally of the US, with the latter acting as a deterrant toward China, which has a historical revindication on Taiwan, which they consider part of the ‘One China.’
And they have been clear about the idea of using force if necessary.
Contrary to Biden, Trump’s comments focus more on economics and have little regard for geopolitics. However, I feel this is just a move to rack up more money, and Taiwan will remain a crucial piece in the puzzle even in a Trump administration no matter what.
But blaming the crash on the comments of the presidential candidates would be only scratching the surface of what is really going on:
The threat of global-scale war.
Formosa, The Key Island
Taiwan is crucial to US interests and maintaining world dominance. Nonetheless, Trump’s candidate VP, JD Vance, has a clear focus on Asia to Europe’s detriment (and NATO’s).
But why is Taiwan, an island 100 miles from mainland China, so important?
Besides being a geographical blocker to China’s growing presence, being a key piece of the ‘First Island Chain,’ Taiwan is also the global foundry of chips; it produces 65% of global chip production and +90% of advanced chips (sub-10 nanometer).
As we covered in our NVIDIA deep dive, NVIDIA, Apple, or Qualcomm are all fabless companies. They design chips and outsource manufacturing.
Well, Taiwanese companies like TSMC are the opposite. They are foundries, or fabs, that do not design but instead only manufacture third-party chips.
Advanced chips are crucial to AI, as most accelerated computing (GPUs, NPUs, TPUs) use sub-10 nm chips (NVIDIA’s Hopper and Blackwell GPUs use 4nm nodes, for instance).
They are also fundamental to cars, laptops, and smartphones, with examples like Apple’s M/A series chips or Qualcomm’s Snapdragon series (powering Microsoft’s laptops), all in the 2-4 nm range.
One notable exception is Groq’s LPU hardware, which uses 14 nm chips (4nm expected for 2025); older chips but the fastest inference system in the world.
For that reason, Groq’s entire production is US-only, as processors are built by Global Foundries (casually one of the few companies alongside Intel not affected by the crash), and their architecture doesn’t require off-chip memory (largely sourced from South Korea and thus impacted by geopolitical tensions in the region).
So, what does all this mean?
Simply put, a blockade of the Taiwan Strait would, in the words of a Taiwanese MP, “destroy the global economy.”
This may sound like an exaggeration, and the true effects are impossible to measure, but the consequences of the world's inability to build advanced chips would be catastrophic.
Therefore, Taiwan’s foundry supremacy is considered a ‘silicon shield’ of sorts (in clear reference to the main material used to build these chips), as a way to prevent China from invading.
But is this all we have to protect Taiwan?
A Matter of Power, Energy, and Feasibility
China must be very careful in its approach to conquering Taiwan for various reasons, from balance of power, energy, to technology.
Balance of power in South China Sea
In a nutshell, invading Taiwan is one of the most complex operations one could ever achieve, logistically speaking.
For starters, invading an island, especially a mountainous one, is extremely difficult.
Secondly, China's balance of power in the South China Sea is complicated, with close US allies nearby in South Korea, Japan, the Philippines, and obviously Taiwan, meaning the outcomes in case of conflict are all but certain.
Thirdly, the Bashi Channel, between Taiwan and the Philippines, is one of the most important chokepoints in the world for telecommunications (most underwater data pipelines between East Asia and the open Pacific go through here) and is a key supply corridor, especially in wartime.
Energy
Unlike the US, China is not energetically self-sufficient (besides coal), which represents 53% of its power mix. Other energy sources are mostly imported by sea, especially oil and LNG (Liquefied Natural Gas) through the Strait of Malacca.
As 90% of Chinese trade and 80% of oil imports travel by sea, a potential blockade from the US Navy on the Malacca Strait, known as the Malacca problem, would seriously constrain China.
Specifically, if India (which controls the Strait) sided with the US, which seems the most likely outcome, blocking the Strait would prevent China from establishing maritime commerce with all of Europe through the Gate of Tears (the Bab al-Mandab Strait) and oil from the Middle East through the Ormuz Strait.
How much money are we talking about?
Every year, around $5.5 trillion in trade goes through these routes between Europe/Middle-East with China/Japan/South Korea.
However, the biggest problem for China (and the rest of the world) is technology.
Technology
Regarding the article's center point, China isn’t interested in killing the semiconductor supply chain.
Nevertheless, the Chinese Communist Party (CCP) relies heavily on technology produced in Taiwan to control its citizens through so-called smartphone ‘super-apps’ like WeChat, which are essential for conducting any type of consumption in great cities, and the extensive network of cameras, which rely on AI to identify bad actors and punish bad activity through the social scoring system (although it must be said that some argue this is a myth).
Nevertheless, with its booming EV market, it is still in dire need of chips.
Thus, it aims to become a chip design and manufacturing force and reduce its dependency on Taiwan. However, the US Government has tried to prevent this from happening.
But this threat of potential war alone goes nowhere near fully explaining why the markets are so volatile.
Trillions of Dollars in Storytelling
As I covered in my report on the AI bubble, AI stocks have seen their valuations increase by multiple trillion dollars since the release of ChatGPT in November 2022.
In full Silicon Valley style, tech tycoons built a beautiful narrative around AI's promises, from eliminating the need for jobs and creating a Universal Basic Income to achieving superintelligence, AI that becomes the supreme intelligence in our world.
Investors loved the idea, applauding multiple-billion investments into the main AI hardware, GPUs, even catapulting NVIDIA into being the most valuable company in the world.
Naturally, in the absence of revenues, the horse can only be beaten for so long, and the world is finally realizing that the industry is largely underdelivering on its promises.
Goldman Sachs shared a pessimistic report on the clear unbalance between investment and returns, with a total CAPEX estimate of $1 trillion for GenAI (money poured into GenAI-related assets, research, etc.).
In parallel, Sequoia puts that value at $600 billion…
with Barclays sharing an investor report along the same lines and quantities as GS.
And what revenues should we expect from those investments in 2024?
The Economist projects $20 billion GenAI revenues between Amazon, Google, and Microsoft for 2024… COMBINED, 160 times less than these three companies have added since ChatGPT’s arrival.
In the private markets, while OpenAI triples the run rate to $3.4 billion for 2024, the next competitor, Anthropic, is still way below the $1 billion annual revenue mark, and the rest are revenue-less only but a few exceptions.
All things considered, as we covered in a previous report, I estimated, using Sequoia’s CAPEX estimate, that the industry was overinvested in a multiple ranging from 12 to 20 times revenues, as the run rate for 2024 will not surpass $50 billion (realistically falling closer to $30 billion), three times less than what NVIDIA alone will turnover in 2024.
In a nutshell, AI is currently a lot of storytelling and bluffs with little to account for.
Painfully, I predict that the technology's undeniable short-term value will mostly come from open-source implementations.
LLM augmentation through LoRA adapters is the only way to reduce hallucinations to a threshold enterprises are comfortable with, and private models can’t reasonably perform these fine-tunings at scale due to their absurd sizes or the complexity of handling such architectures at scale.
In other words, AI isn’t only an overinvested technology, but all signs point to the need for a new ‘reasoning’ breakthrough, not yet materialized, for these companies to dare hope to achieve some return on their investments.
If that wasn’t enough, if Trump wins, they should not expect much benevolence from the administration regarding regulation, as JD Vance, the candidate Vice-President, has been very outspoken about the importance of open-source AI and is blatantly against AI regulation.
So, with no clear path to enterprise revenues, the flashbacks to the dot-com bubble era, with clear similarities (no clear way to monetize their business), will have more than one investor panicking.
Valuation-wise, the comparison with the dot-com bubble doesn’t hold. For the moment.
All things considered, can we build a clear narrative to guide our market decisions moving forward?
Absolutely.
A Guide Against the Hype
Before we move onto the latent opportunity sectors, we start with the most predictable and center point of the whole drama: AI stocks.
Leaders, but at what cost
When considering AI software stocks, the list is quite consolidated: the ‘Fabolous Five’ (Apple, Microsoft, NVIDIA, Alphabet (Google), and Amazon) plus Meta, Tesla, and Oracle in the US and ByteDance and Alibaba in China.
While they all have strong non-GenAI cash cows, their cloud businesses have grown tremendously (Fab Five, ByteDance, Alibaba, and Oracle) or are investing heavily in that direction (Tesla for autonomous cars and humanoid model training and NVIDIA for GenAI model serving through NIMs and robotics simulation).
At first glance, things look great: from a P&L standpoint, they mostly enjoy healthy business models and large margins. But the equally exciting and worrying thing about this group becomes prominent in their future plays.
For starters, they are all slowly converging into the same company, with huge repercussions for the US economy and the world's future.
As their capacity to create trillion-dollar products or businesses is very limited, something required to justify their valuations, and with notable flops like Google Glass, Meta’s Metaverse, or Apple’s Vision Pro, they are all betting their entire futures on the same approach:
Becoming pillars of culture.
![](https://media.beehiiv.com/cdn-cgi/image/format=auto,onerror=redirect/uploads/publication/logo/9b355c4a-4dbc-4f53-b676-a166fee6812c/thumb_Isotipo_negro.png)
Subscribe to Full Premium package to read the rest.
Become a paying subscriber of Full Premium package to get access to this post and other subscriber-only content.
Already a paying subscriber? Sign In