In
1996, Nvidia had roughly 30 days of cash left and a failed chip. A deal with
Sega bought six more months — and a decade-long software bet later made it the
world’s most valuable company.
In 1996, Nvidia had roughly 30 days of
cash left. Its first chip had bet on the wrong technical standard, a follow-on
project with Sega had just been cancelled, and the company was staring at
collapse. Founder Jensen Huang flew to Japan and persuaded Sega to convert
money it already owed Nvidia into a $5 million equity investment instead of a
cash payment — buying the company about six more months of runway. Three
decades later, Nvidia is a roughly $5 trillion company, the most valuable
public company in the world. The rescue is a good story. The decade that came
after it explains the trillion-dollar outcome.
Three
Engineers, a Diner, and a Bet on 3D Graphics
Jensen Huang, Chris Malachowsky, and
Curtis Priem met at a Denny’s outside San Jose in 1993 and founded Nvidia there
with roughly $40,000 in initial capital, later adding a $2 million seed round
from Sequoia Capital’s Don Valentine and Sutter Hill Ventures. Personal
computers at the time had no dedicated hardware for 3D graphics — every polygon
rendered on the CPU — and the three founders bet the PC gaming market would
eventually justify a chip purpose-built for the job. Their first product, the
NV1 (1995), used a quadratic-surface rendering approach that became
incompatible once Microsoft standardized DirectX around triangle-based
rendering, the direction the industry actually went. That miscalculation, plus
a cancelled follow-on project with Sega, is what put the company down to 30
days of cash in 1996. The Sega equity conversion bought time; Nvidia’s next
chip, the RIVA 128 (1997), built on the conventional architecture the market
had settled on, and sold over a million units in four months. (The name, incidentally,
was a fallback — the founders wanted “NVision,” but a toilet-paper manufacturer
already held the trademark.)
The
Decade-Long Bet Almost Nobody Noticed
The real foundation of today’s Nvidia
wasn’t a chip — it was software. In 2006, Nvidia launched CUDA, a
parallel-computing platform letting developers run general-purpose code on
GPUs. For the better part of a decade, CUDA was a niche, largely unprofitable
investment. It took until 2012, when AlexNet — trained on Nvidia GPUs — won the
ImageNet competition and helped spark the modern deep-learning boom, for the
bet to pay off. Today CUDA has more than 4 million developers building on it,
and every optimized library and trained engineer makes the ecosystem stickier.
Rewriting a CUDA-based system for a rival platform like AMD’s ROCm can take
months of engineering time. That’s the moat: not just faster silicon, but
nearly two decades of accumulated software lock-in.
From $27
Billion to $216 Billion in Four Years
The scale of Nvidia’s recent growth is
almost hard to process in raw numbers: revenue went from $26.9 billion in
fiscal 2022 to $215.9 billion in fiscal 2026, with data-center demand from the
generative-AI boom doing most of the work. But the growth wasn’t a straight
line — fiscal 2023 was a genuine down year, not a footnote: gross margin fell
from 64.9% to 56.9%, and net income was cut by more than half, to $4.4 billion,
during the crypto and gaming downturn. That detail matters because it shows
even the category leader can have a rough year inside a larger growth story.
The valuation picture has actually
improved as earnings caught up: Nvidia’s trailing P/E has compressed from
119.8x in fiscal 2023 to roughly 31–38x today, because earnings grew faster
than the share price — a re-rating from “expensive growth story” toward “richly
priced but profitable compounder.” Return on equity above 100% in fiscal 2025
and 2026 looks extraordinary, but buybacks have shrunk the equity base in the
denominator, so the raw percentage overstates things slightly even though
profitability is genuinely excellent. Capital expenditure also grew 58.9%
year-over-year in fiscal 2026, just under revenue growth of 65.5%, worth
watching against future free cash flow.
The Risk
That Erased an Entire Market Overnight
Two direct customers represented 36%
of Nvidia’s total fiscal 2026 revenue, according to the company’s own 10-K — an
unusually concentrated base for a company this size. Several of those largest
customers — Google, Amazon, Microsoft, Meta — are simultaneously investing
billions in their own custom AI silicon (TPU, Trainium, Maia, MTIA), meaning
Nvidia’s biggest buyers are also its most capable long-term competitors. And
geopolitics has already demonstrated how fast a market can vanish: U.S. export
controls took Nvidia’s China AI-chip market share from roughly 95% to
effectively zero, forcing a $4.5 billion charge in the first quarter of fiscal
2026 tied to unsellable H20 inventory. Nvidia’s own 10-K states that exclusion
from China is helping rivals build competing ecosystems. As fabless
manufacturer dependent on TSMC’s leading-edge process nodes, Nvidia also
carries supply-chain risk the company itself has flagged around
demand-forecasting mismatches.
Watch the Full Breakdown
Nvidia’s story is really two stories
stitched together: a near-death hardware bet in the 1990s, and a decade-long
software bet in the 2000s that nobody outside the company thought would matter
until it suddenly did. For the full margin charts, the China exposure numbers,
and how Nvidia stacks up against Broadcom, AMD, and Cerebras, watch the
complete deep-dive video on the Company Narratives YouTube channel.
Comments
Post a Comment