How Nvidia Grew From Gaming To A.I. Giant, Now Powering ChatGPT
#CNBC#business#news#finance stock#stock market#news channel#news station#breaking news#us news#world news#cable#cable news#finance news#money#money tips#financial news#Stock market news#stocks#nvidia#technology
2.4M จำนวนครั้งที่เข้าชม｜8 สรุปแล้ว｜2 ปีที่แล้ว
💫 สรุป
Nvidia 從遊戲圖形處理器起家，現在成為 AI 巨頭，推動 ChatGPT 等大型語言模型的發展，儘管面臨中美貿易緊張和晶片短缺的挑戰，其依賴台積電製造的策略仍使其具有風險。該公司曾在多次瀕臨破產的情況下，通過創新和與軟體社群的合作，成功轉型並擴展至數據中心和自駕車等領域。儘管面臨出口管制和地緣政治風險，Nvidia 仍在透過 AI 和新技術如 RTX 光線追蹤技術，持續引領市場。
✦
Nvidia 從遊戲領域轉向 AI 巨頭，現在為 ChatGPT 提供支持。
00:00Nvidia 的 GeForce 芯片在過去近 30 年中一直受到遊戲玩家的追捧，主導圖形處理市場。
Nvidia 現在的芯片支持大型語言模型如 ChatGPT，並因此獲得 AI 投資的回報。
依賴台灣半導體製造公司（TSMC）生產幾乎所有芯片，使其面臨美中關係的風險。
Nvidia 在歷史上曾多次接近破產，創始人兼 CEO 黃仁勳曾冒險押注於看似不可能的項目。
✦
Nvidia的創立和早期發展專注於加速計算。
02:58Nvidia於1993年在加州Fremont成立，名稱靈感來自於拉丁文的“envy”。
主要業務為GPU，佔公司收入的80%以上，提升中央處理器的運算能力。
Nvidia在早期面臨競爭，但因為與軟體社群合作而成功生存下來。
1997年設計首款高效能圖形晶片，並選擇無廠半導體模式以降低成本。
✦
Nvidia在計算方法和市場策略上經歷了重大變化。
05:57Nvidia的計算方法類似於擁有數千名士兵的軍隊，能夠同時處理多個任務。
在2010年代初，Nvidia在智能手機市場的嘗試失敗，隨後退出。
Nvidia成功收購Mellanox，卻在收購Arm的過程中因監管挑戰而失敗。
儘管面臨挑戰，Nvidia如今擁有26,000名員工，並在AI和雲計算領域占有一席之地。
✦
Nvidia 在醫療、藝術和加密貨幣領域的影響力。
08:55利用基因組測序技術，Nvidia AI 幫助診斷病人並提供治療。
在藝術方面，Nvidia AI 支援的創作如 Rafik Anadol 的作品覆蓋整棟建築。
加密貨幣熱潮使 Nvidia 的 GPU 成為挖礦的熱門工具，導致遊戲卡片供應短缺與價格飆升。
儘管遊戲收入下降，AI 需求仍使 Nvidia 超出預期，特別是在大型語言模型的訓練中。
✦
Nvidia面臨出口管制和地緣政治風險的挑戰。
11:49美國禁止向中國出口先進的AI晶片，影響Nvidia的收入。
Nvidia必須遵守新規定，並重新設計產品以符合要求。
依賴台灣的TSMC使Nvidia面臨更大的地緣政治風險。
美國通過晶片法案以促進本土晶片製造，減少對外依賴。
✦
Nvidia的技術在自動駕駛和數據中心方面不斷擴展。
14:50Nvidia的Tegra晶片被用於亞馬遜的倉庫機器人和數位雙胞胎模擬。
Nvidia Drive平台支援從簡單的駕駛輔助系統到全自動的Robotaxi。
Nvidia推出了新的數據中心CPU，Grace，並強調AI和物理模擬的重要性。
RTX技術透過光線追蹤提升了電腦圖形和遊戲的表現。
00:00This is what hundreds of
millions of gamers in the
00:03world plays on. It's a
GeForce.
00:06This is the chip that's
inside.
00:08For nearly 30 years.
00:09Nvidia's chips have been
coveted by gamers shaping
00:12what's possible in graphics
and dominating the entire
00:15market since it first
popularized the term
00:17graphics processing unit
with the GeForce 256.
00:20Now its chips are powering
something entirely
00:23different.
00:23ChatGPT has started a very
intense conversation.
00:27He thinks it's the most
revolutionary thing since
00:28the iPhone.
00:29Venture capital interest in
AI startups has skyrocketed.
00:33All of us working in this
field have been optimistic
00:35that at some point the
broader world would
00:38understand the importance
of this technology.
00:41And it's it's actually
really exciting that that's
00:43starting to happen.
00:44As the engine behind large
language models like
00:46ChatGPT, Nvidia is finally
reaping rewards for its
00:50investment in AI, even as
other chip giants suffer in
00:53the shadow of U.S.-China
trade tensions and an ease
00:56in the chip shortage that's
weakened demand.
00:58But the California-based
chip designer relies on
01:01Taiwan Semiconductor
Manufacturing Company to
01:03make nearly all its chips,
leaving it vulnerable.
01:06The biggest risk is really
kind of U.S.-China relations
01:09and the potential impact to
TSMC.
01:11That's, if I'm a
shareholder in Nvidia,
01:14that's really the only
thing that keeps me up at
01:16night.
01:17This isn't the first time
Nvidia has found itself
01:19teetering on the leading
edge of an uncertain
01:21emerging market.
01:22It's neared bankruptcy a
handful of times in its
01:25history when founder and
CEO Jensen Huang bet the
01:27company on impossible
seeming ventures.
01:30Every company makes mistakes
and I make a lot of them.
01:33And some of them, some of
them puts the company in
01:38peril. Especially in the
beginning, because we were
01:41small and and we're up
against very, very large
01:44companies and we're trying
to invent this brand new
01:46technology.
01:47We sat down with Huang at
Nvidia's Silicon Valley
01:50headquarters to find out
how he pulled off this
01:52latest reinvention and got
a behind-the-scenes look at
01:54all the ways it powers far
more than just
01:56gaming.
02:07Now one of the world's top
ten most valuable companies,
02:09Nvidia is one of the rare
Silicon Valley giants that,
02:1130 years in, still has its
founder at the helm.
02:15I delivered the first one of
these inside an AI
02:18supercomputer to OpenAI
when it was first created.
02:2260-year-old Jensen Huang, a
Fortune Businessperson of
02:25the Year and one of Time's
most influential people in
02:272021, immigrated to the U.S
.
02:29from Taiwan as a kid and
studied engineering at
02:32Oregon State and Stanford.
02:34In the early 90s, Huang met
fellow engineers Chris
02:36Malachowsky and Curtis
Priem at Denny's, where they
02:39talked about dreams of
enabling PCs with 3D
02:42graphics, the kind made
popular by movies like
02:44Jurassic Park at the time.
02:47If you go back 30 years, at
the time, the PC revolution
02:50was just starting and there
was quite a bit of debate
02:54about what is the future of
computing and how should
02:56software be run.
02:58And there was a large camp
and rightfully so, that
03:03believed that CPU or
general purpose software was
03:06the best way to go.
03:07And it was the best way to
go for a long time.
03:10We felt, however, that
there was a class of
03:12applications that wouldn't
be possible without
03:15acceleration.
03:17The friends launched Nvidia
out of a condo in Fremont,
03:20California, in 1993.
03:22The name was inspired by N
.V.
03:24for next version and
Invidia, the Latin word for
03:26envy. They hoped to speed
up computing so much,
03:29everyone would be green
with envy.
03:31At more than 80% of
revenue, its primary
03:33business remains GPUs.
03:35Typically sold as cards that
plug into a PC's
03:38motherboard, they
accelerate - add computing
03:40power - to central
processing units, CPUs, from
03:43companies like AMD and
Intel.
03:45You know, they were one
among tens of GPU makers at
03:48that time. They are the
only ones, them and AMD
03:51actually, who really
survived because Nvidia
03:55worked very well with the
software community.
03:57This is not a chip
business.
04:00This is a business of
figuring out things end to
04:03end.
04:04But at the start, its future
was far from guaranteed.
04:07In the beginning there
weren't that many
04:08applications for it,
frankly, and we smartly
04:11chose one particular
combination that was a home
04:15run. It was computer
graphics and we applied it
04:19to video games.
04:20Now Nvidia is known for
revolutionizing gaming and
04:23Hollywood with rapid
rendering of visual effects.
04:26Nvidia designed its first
high performance graphics
04:28chip in 1997.
04:30Designed, not manufactured,
because Huang was committed
04:33to making Nvidia a fabless
chip company, keeping
04:35capital expenditure way
down by outsourcing the
04:38extraordinary expense of
making the chips to TSMC.
04:41On behalf of all of us,
you're my hero.
04:45Thank you. Nvidia
04:50today wouldn't be here if
and nor nor the other
04:54thousand fabless
semiconductor companies
04:56wouldn't be here if not for
the pioneering work that
05:01TSMC did.
05:03In 1999, after laying off
the majority of workers and
05:06nearly going bankrupt to do
it, Nvidia released what it
05:08claims was the world's
first official GPU, the
05:11GeForce 256.
05:13It was the first
programable graphics card
05:15that allowed custom shading
and lighting effects.
05:17By 2000, Nvidia was the
exclusive graphics provider
05:20for Microsoft's first Xbox.
05:22Microsoft and the Xbox
happened at exactly the time
05:26that we invented this thing
called the programable
05:28shader, and it defines how
computer graphics is done
05:31today.
05:32Nvidia went public in 1999
and its stock stayed largely
05:36flat until demand went
through the roof during the
05:37pandemic. In 2006, it
released a software toolkit
05:41called CUDA that would
eventually propel it to the
05:43center of the AI boom.
05:45It's essentially a
computing platform and
05:47programing model that
changes how Nvidia GPUs
05:50work, from serial to
parallel compute.
05:53Parallel computing is: let
me take a task and attack it
05:57all at the same time using
much smaller machines.
06:01Right? So it's the
difference between having an
06:04army where you have one
giant soldier who is able to
06:08do things very well, but
one at a time, versus an
06:12army of thousands of
soldiers who are able to
06:16take that problem and do it
in parallel.
06:19So it's a very different
computing approach.
06:22Nvidia's big steps haven't
always been in the right
06:24direction. In the early
2010s, it made unsuccessful
06:27moves into smartphones with
its Tegra line of
06:29processors.
06:30You know, they quickly
realized that the smartphone
06:33market wasn't for them, so
they exited right from that
06:38.
06:38In 2020, Nvidia closed a
long awaited $7 billion deal
06:41to acquire data center chip
company Mellanox.
06:44But just last year, Nvidia
had to abandon a $40 billion
06:47bid to acquire Arm, citing
significant regulatory
06:50challenges. Arm is a major
CPU company known for
06:53licensing its signature Arm
architecture to Apple for
06:55iPhones and iPads, Amazon
for Kindles and many major
06:59carmakers.
07:05Despite some setbacks, today
Nvidia has 26,000
07:08employees, a newly built
polygon-themed headquarters
07:11in Santa Clara, California,
and billions of chips used
07:13for far more than just
graphics.
07:15Think data centers, cloud
computing, and most
07:18prominently, AI.
07:19We're in every cloud made by
every computer company.
07:22And then all of a sudden
one day a new application
07:25that wasn't possible before
discovers you.
07:27More than a decade ago,
Nvidia's CUDA and GPUs were
07:30the engine behind AlexNet,
what many consider AI's Big
07:33Bang moment. It was a new,
incredibly accurate neural
07:36network that obliterated
the competition during a
07:39prominent image recognition
contest in 2012.
07:42Turns out the same parallel
processing needed to create
07:45lifelike graphics is also
ideal for deep learning,
07:47where a computer learns by
itself rather than relying
07:50on a programmer's code.
07:51We had the good wisdom to go
put the whole company behind
07:55it. We saw early on, about
a decade or so ago, that
07:59this way of doing software
could change everything, and
08:02we changed the company from
the bottom all the way to
08:04the top and sideways.
08:06Every chip that we made was
focused on artificial
08:09intelligence.
08:10Bryan Catanzaro was the
first and only employee on
08:13Nvidia's deep learning team
six years ago.
08:15Now it's 50 people and
growing.
08:17For ten years, Wall Street
asked Nvidia, why are you
08:20making this investment and
no one's using it?
08:23And they valued it at $0 in
our market cap.
08:26And it wasn't until around
2016, ten years after CUDA
08:30came out, that all of a
sudden people understood
08:32this is a dramatically
different way of writing
08:35computer programs and it
has transformational
08:38speedups that then yield
breakthrough results in
08:40artificial intelligence.
08:42So what are some real world
applications for Nvidia's
08:44AI? Healthcare is one big
area.
08:47Think far faster drug
discovery and DNA sequencing
08:50that takes hours instead of
weeks.
08:51We were able to achieve the
Guinness World Record in a
08:55genomic sequencing
technique to actually
08:57diagnose these patients and
administer one of the
09:00patients in the trial to
have a heart transplant.
09:02A 13-year-old boy who's
thriving today as a result,
09:05and then also a
three-month-old baby that
09:08was having epileptic
seizures and to be able to
09:11prescribe an anti-seizure
medication.
09:14And then there's art powered
by Nvidia AI, like Rafik
09:16Anadol's creations that
cover entire buildings.
09:19And when crypto started to
boom, Nvidia's GPUs became
09:22the coveted tool for mining
the digital currency.
09:24Which is not really a
recommended usage, but that
09:27has created, you know,
problems because, you know,
09:31crypto mining has been a
boom or bust cycle.
09:33So gaming cards go out of
stock prices, get bid up and
09:38then when the crypto mining
boom collapses, then there's
09:41a big crash on the gaming
side.
09:44Although Nvidia did create a
simplified GPU made just for
09:47mining, it didn't stop
crypto miners from buying up
09:50gaming GPUs, sending prices
through the roof.
09:53And although that shortage
is over, Nvidia caused major
09:56sticker shock among some
gamers last year by pricing
09:58its new 40-series GPUs far
higher than the previous
10:02generation. Now there's too
much supply and the most
10:05recently reported quarterly
gaming revenue was down 46%
10:08from the year before.
10:09But Nvidia still beat
expectations in its most
10:12recent earnings report,
thanks to the AI boom, as
10:15tech giants like Microsoft
and Google fill their data
10:17centers with thousands of
Nvidia A100s, the engines
10:21used to train large
language models like
10:23ChatGPT.
10:24When we ship them, we don't
ship them in packs of one.
10:29We ship them in packs of
eight.
10:31With a suggested price of
nearly $200,000.
10:34Nvidia's DGX A100 server
board has eight Ampere GPUs
10:38that work together to
enable things like the
10:40insanely fast and uncannily
humanlike responses of
10:43ChatGPT.
10:45I have been trained on a
massive dataset of text
10:47which allows me to
understand and generate text
10:49on a wide range of topics.
10:50Companies scrambling to
compete in generative AI are
10:53publicly boasting about how
many Nvidia A100s they have.
10:56Microsoft, for example,
trained ChatGPT with 10,000.
11:00It's very easy to use their
products and add more
11:04computing capacity.
11:06And once you add that
computing capacity,
11:08computing capacity is
basically the currency of
11:11the valley right now.
11:12And the next generation up
from Ampere, Hopper, has
11:14already started to ship.
11:16Some uses for generative AI
are real time translation
11:19and instant text-to-image
renderings.
11:21But this is also the tech
behind eerily convincing and
11:23some say dangerous deepfake
videos, text and audio.
11:27Are there any ways that
Nvidia is sort of protecting
11:30against some of these
bigger fears that people
11:32have or building in
safeguards?
11:34Yes, I think the safeguards
that we're building as an
11:38industry about how AI is
going to be used are
11:41extraordinarily important.
11:42We're trying to find ways
of authenticating content so
11:45that we can know if a video
was actually created in the
11:47real world or virtually.
11:49Similarly for text and
audio.
11:53But being at the center of
the generative AI boom
11:55doesn't make Nvidia immune
to wider market concerns.
11:58In October, the U.S.
12:00introduced sweeping new
rules that banned exports of
12:02leading edge AI chips to
China, including Nvidia's
12:05A100. About a quarter of
your revenue comes from
12:07mainland China. How do you
calm investor fears over the
12:10new export controls?
12:12Well Nvidia's technology is
export controlled, it's a
12:16reflection of the
importance of the technology
12:17that we make. The first
thing that we have to do is
12:19comply with the
regulations, and it was a
12:23turbulent, you know, month
or so as the company went
12:27upside down to re-engineer
all of our products so that
12:31it's compliant with the
regulation and yet still be
12:34able to serve the
commercial customers that we
12:36have in China. We're able
to serve our customers in
12:39China with the regulated
parts and delightfully
12:43support them.
12:43But perhaps an even bigger
geopolitical risk for Nvidia
12:46is its dependance on TSMC
in Taiwan.
12:49There's two issues.
12:50One, will China take over
the island of Taiwan at some
12:53point? And two, is there a
viable, you know, competitor
12:57to TSMC?
12:59And as of right now, Intel
is trying aggressively to to
13:03get there. And you know,
their goal is by 2025.
13:07And we will see.
13:09And this is not just an
Nvidia risk.
13:10This is a risk for AMD, for
Qualcomm, even for Intel.
13:15This is a big reason why the
U.S.
13:16passed the Chips Act last
summer, which sets aside $52
13:19billion to incentivize chip
companies to manufacture on
13:22U.S. soil. Now TSMC is
spending $40 billion to
13:26build two chip fabrication
plants, fabs, in Arizona.
13:29The fact of the matter is
TSMC is a really important
13:32company and the world
doesn't have more than one
13:34of them. It is imperative
upon ourselves and them for
13:39them to also invest in
diversity and redundancy.
13:42And will you be moving any
of your manufacturing to
13:44Arizona?
13:45Oh, absolutely. We'll use
Arizona.
13:46Yeah.
13:47And then there's the chip
shortage.
13:49As it largely comes to a
close and supply catches up
13:51with demand, some types of
chips are experiencing a
13:53price slump. But for
Nvidia, the chatbot boom
13:56means demand for its AI
chips continues to grow, at
13:59least for now.
14:00See, the biggest question
for them is how do they stay
14:04ahead? Because their
customers can be their
14:07competitors also.
14:09Microsoft can try and
design these things
14:12internally. Amazon and
Google are already designing
14:14these things internally.
14:16Tesla and Apple are
designing their own custom
14:18chips, too. But Jensen says
competition is a net good.
14:21The amount of power that the
world needs in the data
14:23center will grow. And you
can see in the recent trends
14:26it's growing very quickly
and that's a real issue for
14:29the world.
14:34While AI and ChatGPT have
been generating lots of buzz
14:38for Nvidia, it's far from
Huang's only focus.
14:40And we take that model and
we put it into this computer
14:43and that's a self-driving
car.
14:47And we take that computer
and we put it into here, and
14:50that's a little robot
computer.
14:52Like the kind that's used at
Amazon.
14:53That's right. Amazon and
others use Nvidia to power
14:57robots in their warehouses
and to create digital twins
15:00of the massive spaces and
run simulations to optimize
15:03the flow of millions of
packages each day.
15:05Driving units like these in
Nvidia's robotics lab are
15:08powered by the Tegra chips
that were once a flop in
15:10mobile phones. Now they're
used to power the world's
15:13biggest e-commerce
operations. Nvidia's Tegra
15:16chips were also used in
Tesla model 3s from 2016 to
15:192019. Now Tesla uses its
own chips, but Nvidia is
15:23making autonomous driving
tech for other carmakers
15:25like Mercedes-Benz.
15:26So we call it Nvidia Drive.
15:28And basically Nvidia D
rive's a scalable platform
15:30whether you want to use it
for simple ADAS, assisted
15:34driving for your emergency
braking warning,
15:36pre-collision warning or
just holding the lane for
15:39cruise control, all the way
up to a robotaxi where it is
15:42doing everything, driving
anywhere in any condition,
15:46any type of weather.
15:47Nvidia is also trying to
compete in a totally
15:49different arena, releasing
its own data center CPU,
15:52Grace. What do you say to
gamers who wish you had kept
15:56focus entirely on the core
business of gaming?
16:01Well, if not for all of our
work in physics
16:05simulation, if not for all
of our research in
16:09artificial intelligence,
what we did recently with
16:13GeForce RTX would not have
been possible.
16:16Released in 2018, RTX is
Nvidia's next big move in
16:19graphics with a new
technology called ray
16:21tracing.
16:22For us to take computer
graphics and video games to
16:25the next level, we had to
reinvent and disrupt
16:28ourselves, basically
simulating the pathways of
16:31light and simulate
everything with generative
16:34AI. And so we compute one
pixel and we
16:38imagine with AI the other
seven.
16:42It's really quite amazing.
16:43Imagine a jigsaw puzzle and
we gave you one out of eight
16:47pieces and somehow the AI
filled in the rest.
16:51Ray tracing is used in
nearly 300 games now, like
16:54Cyberpunk 2077, Fortnite
and Minecraft.
16:57And Nvidia Geforce GPUs in
the cloud allow full-quality
17:00streaming of 1500-plus
games to nearly any PC.
17:04It's also part of what
enables simulations,
17:06modeling of how objects
would behave in real world
17:09situations. Think climate
forecasting or autonomous
17:12drive tech that's informed
by millions of miles of
17:15virtual roads. It's all
part of what Nvidia calls
17:17the Omniverse, what Huang
points to as the company's
17:20next big bet.
17:21We have 700-plus customers
who are trying it now, from
17:25the car industry to
logistics warehouse to wind
17:29turbine plants. And so I'm
really excited about the
17:32progress there. And it
represents probably the
17:35single greatest container
of all of Nvidia's
17:38technology: computer
graphics, artificial
17:40intelligence, robotics and
physics simulation all into
17:42one. I have great hopes for
it.