No single stock has defined this era of markets more than NVIDIA. It is the centrepiece of the largest corporate investment theme in history — the AI infrastructure build-up — and for the better part of a decade, it has been less a company than a referendum on the future of computing. How an investor responded to NVIDIA — whether to buy, hold, sell, or look away — has become, in hindsight, the single most consequential investment decision of a generation.
What follows is a tale told from the inside — through six investors, each with a distinct temperament and playbook, who watched the same opportunity unfold over the better part of two decades. Their names are withheld until the end. For now, they are defined only by how they think.
The Holder — patient, contrarian, thinks in decades. Buys platforms, not products. Has held positions through 90% drawdowns without flinching. When the thesis is intact, the price is irrelevant.
The Visionary — bold, talks in revolutions. Sees the future before anyone and bets billions on it. But the fund has bills to pay each quarter, and conviction competes with liquidity.
The Prophet — identifies the thesis before anyone and says it publicly. Evangelical, media-savvy, builds a following around the calls. But cannot sit still when the price moves against the conviction.
The Macro Trader — waits for inflections, sizes aggressively, takes profit. Discipline is the edge and the cage. Does not hold for years. Holds for the move.
The Institution — one of the most respected names in investing. Fundamentals-driven, values consumer franchises with brand moats and undervalued large-caps with durable earnings. Comfortable with technology only when it resembles a consumer business.
The Sovereign — manages a nation’s wealth across generations. Moves slowly, accumulates quietly, never explains. No headlines. The longest time horizon in the room.
2007 — Big things have small beginnings.
For a long time, there is a mid-cap semiconductor company, NVIDIA, specialised in making GPUs that have been widely popular amongst video gamers and graphic designers. This year it releases CUDA — a software toolkit that makes its graphics chips programmable for general-purpose computation. Jensen Huang, the CEO, had funded the project against board resistance, believing parallel computing would eventually find uses beyond video games.
None of our six investors notice. NVIDIA’s market capitalisation is $20 billion. Its revenue comes from selling graphics cards to gamers, mostly. The word “artificial intelligence” does not appear in its annual report.
2012 — An academic breakthrough.
A PhD student and two professors enter an image recognition competition using a neural network trained on two NVIDIA gaming GPUs. Their model, AlexNet, wins by a margin so wide it effectively ends the previous era of computer vision. Within two years, every serious machine learning lab in the world has purchased NVIDIA hardware.
The Institution has little reason to follow computer vision research. The annual reports it reads say gaming, gaming, gaming. The Institution has built its reputation on a simple rule: invest in what you understand, and ignore what you don’t. Graphics chips for academics do not meet the threshold. The Institution does nothing.
The Prophet, running a smaller research-focused fund, is paying attention — reading the technology press obsessively, looking for exponential curves in their infancy. The Prophet files this away. Not yet investable, but the curve is there.
The others — the Holder, the Visionary, the Macro Trader, the Sovereign — are elsewhere. The Holder is holding an e-commerce company through another year of the market questioning whether it will ever turn a profit. The Visionary is building a massive technology fund. The Macro Trader is trading currencies. The Sovereign is buying government bonds and blue-chip equities across forty countries. None of them has reason to think about GPUs.
2015-2016 — The customer list.
Something changes in NVIDIA’s products, but more noticeably in its customers.
Google open-sources a machine learning framework called TensorFlow. Facebook, Baidu, and Microsoft all publicly disclose large GPU cluster purchases for artificial intelligence research. NVIDIA responds by launching the Tesla P100 — the first chip designed explicitly for deep learning, not a gaming card repurposed, but silicon built from the ground up for matrix multiplication at scale. Jensen Huang begins saying “AI computing” on every earnings call.
NVIDIA’s data centre revenue is growing roughly 145% year-over-year — from $339 million in fiscal 2016 to $830 million in fiscal 2017. From a small base, but the trajectory is vertical.
The Holder sees this. The firm is known for one thing above all: patience. It held the same e-commerce company when it lost 90% of its value. It held an electric car maker when it nearly went bankrupt. The Holder does not think in quarters. The Holder thinks in decades.
The Holder looks at the customer list — Google, Facebook, Microsoft, Baidu — and sees a pattern seen before. When the four largest technology companies on earth are all buying the same product from the same supplier, it usually means the supplier is building a platform, not selling a component. The Holder builds a position of over 55 million shares.
The Prophet acts at almost the same time, making a first purchase in Q4 2016. The Prophet is among the first to publicly articulate the investment thesis: this is not a gaming business. It is the infrastructure layer for machine intelligence. The Prophet says this on television, in white papers, in conference keynotes. Early, right, and loud about it.
The Visionary arrives in 2017 with characteristic boldness, calling NVIDIA “the core of the AI revolution” and building a $4 billion position. The Visionary sees what the Holder sees — the customer list, the revenue curve — but through a different lens. Where the Holder sees a decade-long hold, the Visionary sees a proof point for the thesis that artificial intelligence will reshape every industry on earth.
The Macro Trader is not here yet. The Macro Trader trades inflections, not narratives. The narrative is interesting but the inflection has not arrived. Watch and wait.
The Sovereign has a small allocation through index weight. Its mandate does not permit concentrated bets on mid-cap semiconductor companies. It is patient by design, not by choice.
The Institution is not here. It has owned technology companies before — but only those it could frame as consumer franchises with brand moats. A semiconductor company selling infrastructure to other companies does not fit that frame. The analysts file a note. It goes nowhere.
2018-2019 — The fog and the first exit.
The cryptocurrency boom of 2017-2018 drives GPU demand to unsustainable levels. Miners are buying every GeForce card they can find, and NVIDIA’s gaming revenue surges — up 52% year-over-year in one quarter. Management attributes the growth to “strong organic demand from gamers.” It is not entirely true. Internally, as much as 60-70% of GeForce revenue in China is flowing to miners, and crypto demand accounts for roughly 83% of GPU revenue growth during this period. The SEC will later fine NVIDIA $5.5 million for failing to disclose the extent of crypto-driven sales.
Then the crypto market collapses. NVIDIA’s gaming revenue falls 45% in a single quarter. The stock drops 50% in four months — from roughly $7.20 to $3.30 in split-adjusted terms. The guidance cut in January 2019 — from $2.7 billion to $2.2 billion, against Wall Street expectations of $3.4 billion — sends the stock down another 28% in two days.
This is the fog that tests every investor in this story. The AI thesis has not changed — data centre revenue continues to grow. But the crypto crash contaminates the narrative. If management obscured crypto exposure within gaming revenue, what else might be overstated? Is the AI demand real, or is it the next transient bubble dressed in a better story? No one can separate the signal from the noise with certainty.
The Visionary sells. All of it. The entire $4 billion position, liquidated in January 2019 for approximately $3.6 billion. The fund has taken a significant paper loss from the peak, the stock is in freefall, and there are other commitments to fund. The long-term thesis is abandoned for tactical reasons.
The Visionary does not yet know that the position just sold will be worth over $250 billion.
Just weeks after the sale, OpenAI publishes GPT-2 — a language model trained on GPUs that it declares “too dangerous to release.” The paper makes headlines in the technology press. It does not make headlines in the financial press.
The Holder does not sell. Does not even consider selling. The customer list has not changed. Google is buying more GPUs, not fewer. The data centre revenue line, briefly obscured by the crypto hangover, continues to grow. The crypto revenue was never part of the Holder’s thesis — the thesis was AI infrastructure, and that line is intact. The Holder’s process does not include “the stock went down” or “management had a disclosure problem” as reasons to sell a position whose fundamental demand drivers remain in place.
The Prophet does not sell. Holds and continues to evangelise. The thesis is intact. The crypto noise is exactly that — noise.
The Institution does not notice. It is buying back its own stock and collecting dividends.
The Sovereign’s allocation has grown slightly with the index weight. It makes no active decision in either direction.
March 2020 — The crash.
The world shuts down. Markets collapse. NVIDIA’s stock falls to around $6 per share (split-adjusted). For a brief window — days, not weeks — you can buy the picks-and-shovels supplier to the entire artificial intelligence industry for less than it traded a year earlier, with four years of additional evidence that the thesis is correct.
The Holder holds. The Prophet holds. The Visionary, having sold the year before, begins quietly re-acquiring shares without announcement. The Macro Trader watches the data centre revenue line — still growing through the pandemic — and files another note. The Sovereign continues its scheduled rebalancing, buying mechanically into the drawdown as it does with every asset in its portfolio.
The Institution buys bank stocks.
2022 — The inflection.
In March, NVIDIA’s newest chip, the H100, ships with a component the company calls the “Transformer Engine” — named after the architecture that will soon change everything. Eight months later, in November, OpenAI releases a chatbot called ChatGPT built on that architecture. It reaches 100 million users in two months. Every CEO on earth suddenly needs an AI strategy. Every board suddenly needs to understand GPU supply chains. NVIDIA’s stock, beaten down through the 2022 bear market, trades at around $14 per share — roughly 25 times forward earnings.
The Macro Trader moves. This is the moment — a discontinuous narrative shift, where what was speculative becomes consensus overnight. The Macro Trader builds a position worth roughly $400 million — 14% of the portfolio — over the quarter. The position grows as more shares are added through Q1 and Q2 2023.
The Holder is still here. The position from 2016 has been compounding quietly for six years. Nothing new is done. The Holder simply continues to hold.
The Prophet is also still here — right about NVIDIA for six years, having said it publicly, repeatedly, with conviction. The fund holds a meaningful position.
And then, in Q4 2022, the Prophet begins to sell. Millions of shares across the fund’s portfolios. Then more in Q1 2023. Then more in Q2 2023. The valuation is declared “too high.” The capital rotates into smaller names. By mid-2023, the position is largely gone.
The stock is at $14-17 per share when the selling begins. It will reach $230.
The Prophet saw the future before anyone. Articulated the thesis years before the market caught up. Held through the doubt, the crypto winter, the COVID crash. And sold at the beginning of the move that had been predicted for an entire career. The missed gains will exceed $1 billion. It is the most visible mis-call of the AI era — made not by a skeptic, but by the truest of believers.
The Sovereign’s allocation has been growing steadily as NVIDIA’s weight in global indices increases. Active additions begin — not just index tracking.
The Institution watches the same ChatGPT demonstration as the rest of the world. The analysts prepare a briefing. The response is the same as it was in 2012, 2016, and 2020. No position is taken.
2024-2025 — The winners start leaving.
NVIDIA has tripled from the Macro Trader’s entry point. The position, built at roughly $400 million, has grown substantially. The process says: take profit. Concentration risk is real. The Macro Trader begins selling through 2023 and 2024, fully exiting by Q3 2024.
The Holder does not sell. Eight years in, the thesis is more right than it has ever been. The customer list has expanded from four companies to every Fortune 500 company on earth. Data centre revenue is more than doubling year after year. The Holder begins to trim for portfolio management reasons — reducing exposure where a single position has grown to dominate the portfolio — but the core position remains. NVIDIA stays the firm’s largest holding.
The Visionary, having re-acquired NVIDIA shares after the 2019 exit, now holds roughly $5.8 billion. In, out, and back in. The conviction in the AI thesis is genuine. But in October 2025, the entire position is sold again — this time to fund a $23 billion investment in a different AI company. The “core” thesis has moved from the hardware layer to the application layer.
The Sovereign now holds over 320 million shares. It has never sold. It has never published a thesis. It simply accumulated, quarter after quarter, year after year, while the others debated, exited, re-entered, and debated again.
2026 — The ledger.
NVIDIA’s market capitalisation today exceeds $5 trillion. The $700 billion the hyperscalers are spending this year on AI data centres flows predominantly through its chips. The stock trades above $220 per share.
Let us open the ledger — and the names.
Reveal the names ↓
The Holder is Baillie Gifford, the Edinburgh-based investment firm founded in 1908. They built a position of over 55 million shares starting in 2016 and held. Their position compounded from roughly $1 billion to tens of billions. They have trimmed approximately 40% of the stake in 2024 as the position grew to dominate the portfolio, but NVIDIA remains their largest holding. When asked why they trimmed, the answer was not that the thesis had changed — it was that portfolio concentration had exceeded internal limits.
The Visionary is Masayoshi Son of SoftBank. He bought $4 billion worth of NVIDIA in 2017, sold the position in January 2019 for approximately $3.6 billion, re-acquired through 2020-2025, built back to $5.83 billion, and sold everything in October 2025 to fund OpenAI. Across two full cycles, he was early, correct about the thesis, and unable to hold the position through either one. At today’s price, the 4.9% stake he sold in 2019 would be worth over $250 billion.
The Prophet is Cathie Wood of ARK Invest. She identified the thesis in 2016, articulated it publicly before anyone, held through years of skepticism, and began liquidating in Q4 2022 at around $14 per share — months before the stock began a run past $230. She bought back a small number of shares in November 2025 on a dip. The missed gains exceed $1.2 billion. It remains the most painful call of the AI era — not because the thesis was wrong, but because conviction could not survive the short-term price action.
The Macro Trader is Stanley Druckenmiller. He bought in Q4 2022 at roughly $15, built a position of approximately $400 million, added through H1 2023, and sold gradually through late 2023 and 2024 — fully exiting by Q3 2024. He publicly acknowledged the exit was a mistake: “I’ve made so many mistakes in my investment career; one of them was I sold all my Nvidia. It was a big mistake.” His process — wait for the inflection, size aggressively, take profit — produced an exceptional absolute return and an exceptional opportunity cost simultaneously.
The Institution is Berkshire Hathaway. It never purchased a single share. It sat on a $397 billion cash pile while the information was in every earnings call, every 10-K, and every hyperscaler’s capital expenditure disclosure.
The Sovereign is Norway’s Government Pension Fund Global — the world’s largest sovereign wealth fund at $1.78 trillion. It accumulated over 324 million shares worth approximately $43.5 billion as of early 2025, quietly, methodically, without headlines. NVIDIA became one of the fund’s most valuable single-stock holdings, built through index-weight increases and active allocation by Norges Bank Investment Management. It never sold. It simply compounded on behalf of a nation.
And then there is a seventh story — one that was not in the original cast because it is not an investor’s story. It is a founder’s story.
Curtis Priem co-founded NVIDIA alongside Jensen Huang and Chris Malachowsky in 1993. At the 1999 IPO, he held approximately 12.8% of the company. In the same year, he donated roughly two-thirds of his stake to a family foundation. He sold the remainder by 2006 to fund science education. It was a deliberate choice to use the present differently.
Those shares, had they been held, would today be worth over $600 billion.
Jensen Huang, the only co-founder who held through the entire run, owns approximately 860 million shares worth over $190 billion. He has sold roughly $2.9 billion through pre-arranged plans since mid-2024 — less than 2% of his holdings.
What the chronicle tells us.
Six investors. One stock. Nearly two decades. Same information, same earnings calls, same public filings. Six different outcomes.
The Holder held. The Visionary left twice. The Prophet called it first and abandoned it at the worst moment. The Macro Trader timed the entry perfectly and the exit too early. The Institution never arrived. The Sovereign simply accumulated.
None of them lacked information. The thesis took nineteen years to fully express itself — from a toolkit no one asked for in 2007 to a $5 trillion company in 2026. The hardest part was never identifying the opportunity. It was staying in the room long enough.
Note on data
NVIDIA share prices are split-adjusted for the 10:1 split of June 2024 and the 4:1 split of July 2021 where applicable. Institutional holdings are sourced from SEC 13F filings and financial press reports. SoftBank transaction data from CNBC, Fortune, and TechCrunch. Baillie Gifford holdings from FundSelector Asia, Livewire Markets, and CNBC. ARK Invest transactions from StockCircle, Yahoo Finance, and Benzinga. Curtis Priem biographical data from Fortune (Feb 2026). Stanley Druckenmiller transactions from StockCircle and Fortune. Norway GPFG data from MarketBeat and CNBC. This article does not constitute investment advice.