The ledger remembers what the hype forgets. On a Tuesday morning when Bitcoin liquidity was bleeding out of order books and DeFi TVL had dropped 12% in a week, ARK Invest filed its 13F. The filing showed a 5% increase in NVIDIA and a 3% increase in TSMC. The market yawned. The crypto Twitter argued about tariffs. But I saw something else: a capital allocation strategy that treats semiconductor fabrication as the ultimate smart contract.
Let me rewind. In 2017, I spent 400 hours auditing the Zcash v1.0.0 bridge to Ethereum. I found a timestamp manipulation vulnerability that allowed infinite minting under specific block timing conditions. My colleagues were busy chasing ICO marketing hype. I learned that the real value in crypto isn’t the token—it’s the underlying infrastructure that enables trust. Today, that infrastructure is shifting from code to silicon.
Context: ARK is not a semiconductor fund. It’s a thematic investor that bets on disruptive innovation. The firm’s decision to increase positions in NVIDIA (a fabless GPU designer) and TSMC (a pure-play foundry) during a period of market uncertainty signals a conviction that the physical layer of computation is becoming the bottleneck for both AI and crypto. The common narrative treats these as separate industries. ARK’s portfolio says otherwise.
Core insight: The convergence of AI and crypto is not happening in the application layer—it’s happening in the fab. NVIDIA’s Hopper and Blackwell architectures are the engines for training large language models, but they also power zero-knowledge proof generation, blockchain validator nodes, and high-frequency trading bots. TSMC’s 3nm and 2nm processes are the only production lines capable of delivering the transistor density required for next-generation ASICs used in Bitcoin mining and Ethereum staking. The market treats these as separate supply chains. They are the same supply chain.
Let me walk through the technology. TSMC’s 3nm (N3) process has been in mass production since late 2022. The enhanced N3E and N3P variants are now standard for high-performance computing. NVIDIA’s Blackwell GPU, built on a custom 4NP process, packs two chiplet dies in a single package using TSMC’s CoWoS-L advanced packaging. CoWoS (Chip-on-Wafer-on-Substrate) is the bottleneck that everyone pretends doesn’t exist. In 2024, TSMC’s CoWoS capacity was around 40,000 wafers per month. By 2025, it’s expected to double to 80,000. But demand from NVIDIA, AMD, Broadcom, and AI startups has already oversubscribed the entire capacity. This is not a supply chain—it’s a single point of failure dressed as a monopoly.
From my experience at a hedge fund during DeFi Summer 2020, I identified that 15% of total value locked in Uniswap V2 was artificially inflated by impermanent loss harvesting bots. The bots exploited the constant product formula. Today, the same logic applies to semiconductor allocation: arbitrage is just a mispricing of scarcity. The market is pricing TSMC and NVIDIA as if their capacity is elastic. It is not. A single fab takes three to five years and $20 billion to build. The capital expenditure is irreversible. The depreciation cycle is five years. The yield curve takes another two to three years to stabilize. This is the opposite of the agile, forked world of crypto. It’s hard, physical, and unforgiving.
The core of the matter is simple: the cost of compute is the new cost of trust. Every DeFi protocol, every L2 rollup, every zk-proof requires computation. That computation requires chips. Those chips require TSMC. And the software that orchestrates them requires NVIDIA’s CUDA ecosystem. ARK is not betting on a stock. It’s betting on the gatekeepers of the physical layer.
Now, the contrarian angle. The market is obsessed with the “AI bubble” narrative, especially after Meta’s earnings miss in late 2024. The argument: if Meta—the biggest spender on AI infrastructure—can’t monetize its investment, then the whole AI capex cycle is a waste. The data says otherwise. Meta’s capex for 2024 was $37 billion, and it guided for $40-45 billion in 2025. The so-called “miss” was a 20% increase in revenue, not a decline. The narrative that AI adoption is slowing is a misreading of the numbers. What’s actually happening is that the benefits of AI are accruing to the infrastructure layer first—just like how the benefits of the internet accrued to Cisco and Qualcomm before they reached Yahoo and Amazon.
But here’s the crypto-specific twist: the same narrative applies to the blockchain space. When DeFi summer ended, the market said “DeFi is dead.” But the infrastructure—Ethereum, L2s, bridges—continued to build. The ledger remembers. The hype forgets. The same is happening now. The market is bearish on crypto because of regulatory uncertainty and macroeconomic headwinds. Yet ARK is loading up on the back-end of the crypto stack: the chips that power the validators, the miners, and the zk-proof generators.
Liquidity is just confidence dressed as code. In the crypto market, liquidity is measured in TVL and order book depth. But the real liquidity is the ability to execute computation at scale. That liquidity is concentrated in the hands of two companies. ARK is betting that this concentration will persist, and that the value of compute will continue to rise relative to the value of tokens.
Let me give you a concrete example. The Bored Ape Yacht Club liquidity trap in 2021 taught me that 80% of floor price stability relied on a single whale wallet. The NFT market was an illusion of decentralization. The same is true for AI compute. The market assumes that anyone can buy chips. But the reality is that NVIDIA’s top customers—Microsoft, Amazon, Google, Meta—already have long-term allocation agreements that lock out smaller players. The secondary market for H100s is a graveyard of broken promises. The price of a used H100 on eBay is still $25,000, down from $40,000 in 2023, but that’s not a sign of weakness—it’s a sign that the bottleneck is shifting from the chip itself to the interconnects and the software stack.
Now, let’s dive into the geo-political dimension. The Ukraine war and the Taiwan strait are the two most dangerous geopolitical flashpoints for the global economy. TSMC is headquartered in Taiwan. If the strait freezes, the entire world’s advanced semiconductor supply stops. That’s not a risk—it’s a certainty that hasn’t happened yet. The market prices this risk at zero. ARK is pricing it at a premium, but not by hedging—by doubling down. Why? Because the only way to mitigate the risk is to build fabs elsewhere. TSMC is building fabs in Arizona, Japan, and Germany. Those fabs will take years to reach yield parity. In the meantime, TSMC’s Taiwan fabs are the only game in town. The market is ignoring the geographic concentration, but the advanced packaging and 2nm processes are staying in Taiwan for at least the next five years.
Smart contracts execute; they do not feel remorse. The same can be said about semiconductor supply chains. They execute regardless of human sentiment. The capacity is fixed. The demand is growing. The price will adjust upward until the market clears. That’s the investment thesis in a nutshell.
But let’s be precise. The technology analysis reveals that TSMC’s 2nm (N2) process, using GAA (Gate-All-Around) nanosheet architecture, is expected to enter production in the second half of 2025. This is a significant leap from FinFET. The early yield ramp is expected to take 12-18 months to reach commercially viable levels. This means that the supply of the most advanced chips will be even more constrained in 2025-2026 than it is today. NVIDIA’s next-generation architecture, codenamed Rubin, is expected to use TSMC’s 3nm or 2nm derivatives. The combination of N2 + CoWoS-L + HBM4 will create a new class of supercomputers that are 10x more powerful than Blackwell. The demand for these systems is not just AI—it’s also for proof-of-stake consensus, zk-SNARK verification, and high-frequency trading. The crypto industry is a silent but growing consumer of these chips.
From my crisis analysis during the Terra/LUNA collapse, I spent 600 hours modeling the UST de-pegging mechanism. I found that the withdrawal limits on Curve pools were the critical point of failure. If the caps had been enforced within 12 hours, $2 billion in liquidity could have been preserved. The lesson: liquidity is not infinite. In the semiconductor world, the same applies. The supply of advanced chips is finite, and the withdrawal limits are not set by code but by physics. The ARK thesis is that the physics of Moore’s Law is slowing, but the demand for compute is accelerating. The only way to reconcile the two is for the price of compute to rise. That’s a macro bet on the cost of trust.
Now, let’s examine the market demand side. The AI training chip market is estimated to be $60-80 billion in 2025, growing at 40-60% annually. AI inference is growing even faster, at 80%+ annually, because every deployed model needs to run inference repeatedly. The crypto sector’s demand for inference chips is often overlooked. Every time a user interacts with a DeFi protocol, the protocol runs a smart contract. That smart contract is executed on a validator node, which is a computer running a CPU or GPU. The migration to zk-rollups has increased the demand for prover hardware, which is essentially a GPU farm. The leading zk-prover implementations use NVIDIA GPUs for parallelized computation. The more L2s scale, the more GPUs they need. The market is pricing this as negligible. It is not.
We don’t buy history; we buy the memory of it. The memory of the 2021 NFT bull run is still fresh. The memory of the 2022 crypto winter is fading. But the memory of the computing power required to build those systems is stored in the billions of dollars spent on chips. ARK is buying the memory of the next bull run, before the market remembers that it already happened.
Let me address the elephant in the room: Tether’s reserves. I have always maintained that USDT’s dominance is a ticking time bomb. The industry pretends that Tether’s reserves are fine without an independent audit. That’s a liquidity risk that could trigger a systemic shock. How does that relate to semiconductors? If Tether collapses, the crypto market would lose a significant amount of liquidity. But the physical infrastructure—the miners, the validators, the zk-provers—would still need to run. They would switch to other stablecoins or fiat. The demand for chips would not disappear. It would shift. That’s why ARK’s bet on the infrastructure is more resilient than a bet on any token. The chips are the ultimate reserve asset.
Now, the contrarian take: the market is bearish on crypto because of regulatory uncertainty. The SEC is suing exchanges. The EU’s MiCA is imposing compliance costs that will kill small projects. But the infrastructure companies—TSMC, NVIDIA—are not directly exposed to those regulations. They sell to the highest bidder, regardless of jurisdiction. The regulatory pressure on crypto will actually increase the demand for compliant, regulated infrastructure. The institutions that are entering the space will need to run their own validators and miners. They will buy chips from TSMC and NVIDIA. The regulatory crackdown is a tailwind for the infrastructure layer, not a headwind.
The ledger remembers what the hype forgets. The hype is about memecoins and AI agents. The ledger is the physical world of silicon and copper. The market is currently pricing the hype at a premium and the ledger at a discount. ARK is arbitraging that gap.
Let me conclude with a forward-looking judgment. The next crypto cycle will not be driven by retail speculation or DeFi yields. It will be driven by institutional adoption of AI-crypto hybrid systems. The infrastructure for those systems is already being built. The bottleneck is not capital—it’s compute. The companies that control the compute will capture the lion’s share of the value. ARK is positioning for that reality. The question is: are you?
In five years, we will look back at the 2025 sell-off as a buying opportunity for the two companies that underpin the digital economy. The market will realize that the semiconductor supply chain is the most important infrastructure in the world. The ledger will show that ARK was early. The hype will have moved on. But the data will remain.