The $2 Billion Validator: Why Ackman Is Right About Meta’s “Deep Discount”
In the fast-moving world of Silicon Valley, company values can change drastically based on a single earnings call. In this environment, Bill Ackman has made a clear choice. Pershing Square has built a stake in Meta Platforms worth roughly $2 billion. As of late 2025, this amounted to 10% of the fund’s capital. The market is currently nervous about Mark Zuckerberg’s large forecast for capital expenditures. Ackman, however, sees a “deeply discounted” asset trading at roughly 22 times forward earnings.
Ackman is right. The doubt surrounding Meta’s move to “Personal Superintelligence” misses the bigger picture. We must look at the company’s massive infrastructure plans, its control of software through PyTorch, and the internal debates led by Yann LeCun. We should also consider the history of human cognitive development. When you analyze these factors, it is clear that Meta is not just burning cash. It is building the only defense that matters in the 21st century.
The “Shock and Awe” of Meta Compute
The argument against Meta is based almost entirely on the price tag. During the Q4 2025 earnings call, CFO Susan Li shared a number that worried Wall Street. She announced a 2026 capital expenditure guidance of $115 billion to $135 billion. This is nearly double the $72.2 billion spent in 2025.
However, a closer look at the “Meta Compute” plan shows that this spending is both insurance and a weapon. Zuckerberg stated clearly that being the most efficient at building infrastructure “will become a strategic advantage.” The company is starting work on gigawatt-scale sites. These include the $10 billion “Hyperion” facility in Louisiana and a center in Lebanon, Indiana. These sites are designed to handle both core business tasks and AI workloads.
Ackman believes this infrastructure is an engine for growth, not just a cost. The earnings call proved this point. AI tools for ad ranking delivered a revenue impact nearly 4x larger than simply increasing ad loads. The company reported Q4 revenue of $59.9 billion, which is up 24% year-over-year. This proves the AI spending is already paying off through higher engagement and better ad performance.
The Invisible Moat: PyTorch and Software Supremacy
Beyond the physical steel and chips, Meta holds another critical advantage. It owns the “operating system” of modern AI. Meta created and maintains PyTorch, the deep learning framework used in over 80% of research papers. Competitors like Tesla and OpenAI use it as well. This gives Meta deep expertise in optimizing software to get the best performance from its hardware.
While others struggle with standard tools, Meta uses PyTorch 2.0. This utilizes the TorchInductor compiler to turn code into high-performance machine language. It allows Meta to reach 100% GPU utilization in training scenarios where other frameworks might top out at 90%. Furthermore, Meta developed ExecuTorch. This enables them to run complex AI models directly on mobile devices. It moves the cost to the edge, or the user’s phone, rather than the server. This is vital for bringing “Personal Superintelligence” to 3.5 billion users without ruining the company’s finances.
This software mastery helps with hardware design too. Meta’s custom MTIA chips are optimized specifically for their ranking models. By designing the silicon to work with the PyTorch software stack, Meta achieved a 44% reduction in Total Cost of Ownership compared to commercial GPUs. Advanced software inventions like the Distributed Shampoo optimizer allow Meta to train huge neural networks more efficiently than standard methods. In a world where computing power is scarce, Meta’s ability to write code that makes hardware run faster multiplies the value of every dollar spent.
The LeCun Divergence: A Philosophical Schism
While the financial numbers look good, a deep intellectual debate has ended in a departure. Chief AI Scientist Yann LeCun has left Meta. While the rest of the company races to scale Large Language Models, LeCun has signaled he wants no part of that path. At a seminar at NYU, he declared that auto-regressive LLMs are “doomed” when it comes to finding true Artificial General Intelligence.
LeCun’s disagreement stems from the hard limits of text-based learning. He argues that LLMs, which just predict the next word in a line, cannot truly reason, plan, or understand the physical world. He points to Moravec’s paradox. Computers can easily pass a bar exam but cannot match the physical intuition of a domestic cat. LeCun suggests shifting toward “World Models” and Joint Embedding Predictive Architecture, or JEPA. These systems learn from sensory inputs like video rather than just text. This approach aims to avoid the errors and hallucinations that are common in LLMs.
The Helen Keller Rebuttal: Why Language Is Enough
LeCun’s doubts about LLMs might be scientifically wrong. The history of human thought offers a strong counter-argument. You do not need to be immersed in the physical world to possess intelligence. The proof lies in the life of Helen Keller.
Keller was blind and deaf from 19 months old. She was cut off from the visual and sound data that LeCun thinks is essential for a “world model.” Yet, she became a prolific author, a political activist, and a graduate of Radcliffe College. If LeCun were right, and intelligence required high-bandwidth sensory data like video, Keller’s development should have been impossible.
Instead, Keller’s life supports the “Strong Minimalist Thesis” of language. This idea says that syntax and symbols can lead to a deep understanding of reality. When Keller connected the finger-spelled letters “W-A-T-E-R” to the liquid on her hand, she had a realization. She could organize her mental concepts. She used language to understand things she could never physically experience, such as art or the abstract idea of “thought.” As Keller herself wrote, “The bulk of the world’s knowledge is an imaginary construction.”
This challenges LeCun’s view that text is not enough. Keller had access to far less data than a modern LLM, which trains on trillions of tokens. Still, she achieved human-level reasoning. She proved that language is not just a copy of thought. It is a tool that can encode the structure of reality. If a human brain can reconstruct the world through simple tactile signs, an LLM trained on all human knowledge has the materials needed for deep understanding.
The Investment Verdict
Bill Ackman’s bet is not just about ad revenue. It is a wager that Meta will win the race to AGI through superior engineering.
If LeCun is right, Meta is ready. It is funding his JEPA research and has access to the world’s largest library of video data from Instagram and Facebook to train these World Models. If the counter-argument holds, and the Helen Keller example proves LLMs work, then Meta is also winning. Its massive investment in Llama and infrastructure will create the “Personal Superintelligence” Zuckerberg promises.
Meta is trading at a discount because the market fears the cost of infrastructure. But the company is building gigawatt-scale centers. It optimizes them with PyTorch software. It is bringing AI into the daily lives of 3.5 billion users. Meta is creating a reality where it owns the foundation of digital intelligence. It does not matter if the future lies in LeCun’s sensory models or the power of language that freed Helen Keller. Meta is forging the path forward. Ackman sees the discount. The rest of the market will eventually see the reality.
Disclaimer:
All views expressed are my own and are provided solely for informational and educational purposes. This is not investment, legal, tax, or accounting advice, nor a recommendation to buy or sell any security. While I aim for accuracy, I cannot guarantee completeness or timeliness of information. The strategies and securities discussed may not suit every investor; past performance does not predict future results, and all investments carry risk, including loss of principal.
I may hold, or have held, positions in any mentioned securities. Opinions herein are subject to change without notice. This material reflects my personal views and does not represent those of any employer or affiliated organization. Please conduct your own research and consult a licensed professional before making any investment decisions.