To The Moon

​Wall Street is terrified that the AI bubble is about to burst.

​But Gavin Baker (the former Fidelity legend who famously beat 99% of his peers) just dropped a contrarian masterclass on why this cycle is fundamentally different.

​The tl;dr? The brakes of this bull market aren't controlled by the Fed. They are controlled by TSMC.

​If you are trading the AI infrastructure boom, your thesis is dangerously incomplete without these 3 structural realities:

​1️⃣ The Ultimate Anti-Bubble: TSMC’s "Stubborn" Guardrails

​The market's biggest fear is a repeat of the 2000 telecom crash or the 2018 memory glut: supply massively outrunning demand, leading to a catastrophic collapse in pricing.

​Baker’s reality check: That requires unhinged overbuilding. And the only company capable of overbuilding is TSMC—run by battle-hardened, conservative veterans who have survived decades of brutal semiconductor cycles. They know overexpansion leads to blood in the streets.

​Even if Jensen Huang flies to Taiwan every quarter demanding a 200% capacity increase, TSMC's management won't budge. They are pacing growth at a highly restrained 5-10%.

​The Alpha: Software can scale infinitely overnight; advanced fabs take years and billions in CapEx. By bottlenecking supply at the physical layer, TSMC is artificially extending the life of this bull market. This enforced scarcity is the ultimate long-term bull case for Nvidia’s pricing power.

​2️⃣ The Invisible Moat: Chips are Useless Without the "Highway"

​Everyone is obsessed with the chips. Baker is looking at the network.

​Frontier AI models (like GPT-4 and Claude) run on Mixture-of-Experts (MoE) architecture. Instead of firing up the entire brain for every prompt, they only wake up specific "experts." This requires a "Switched Scale-Up Network"—a hyperspeed nervous system allowing thousands of GPUs to talk to each other with near-zero latency.

​Building this nervous system is brutally difficult. Baker notes only two companies have it fully operational at scale:

​Nvidia: The undisputed gold standard (NVLink + InfiniBand).

​Amazon: The sleeper agent. Their custom network built for Trainium chips is the most underestimated infrastructure in the market today.

​The Alpha: The barrier to entry for the network is actually higher than the silicon itself. Keep a close eye on Amazon’s Trainium 3 scaling up late this year. If Baker is right, AWS is sitting on a massive, mispriced advantage.

​3️⃣ Neoclouds: The F1 Teams of Silicon

​The consensus is lazy: “Once legacy hyperscalers (Microsoft, Google) get enough GPUs, they will crush Neocloud upstarts like CoreWeave.”

​Baker’s response: Operating a massive GPU cluster isn't like running a rental car fleet; it’s like racing a Formula 1 car.

​Anyone can buy an Nvidia GPU. But top-tier operators like CoreWeave squeeze 2-3x higher utilization rates out of the exact same hardware. Building the software stack, thermal management, power routing, and task scheduling around these chips requires an elite, specialized engineering team.

​The Alpha: Legacy clouds are built to optimize for lowest cost. AI tasks demand extreme, redline performance. The Neoclouds aren't just arbitrageurs; they possess a massive operational moat that will yield compounding growth long after the hyperscalers catch up on hardware.

@TigerEvents  @Tiger_comments  @Daily_Discussion  @TigerWire  @TigerStars