젠슨 황 퇴진, AI 패권 전쟁의 서막! 엔비디아가 주인이 아니었다… 앤트로픽·스페이스X가 준비하는 새로운 AI 시대 — Transcript
Full transcript
- 0:00There are two companies that NVIDIA TV
- 0:02has been focusing on intensively lately
- 0:04. They are Anthropic and SpaceX. I
- 0:07believed they were companies creating AI
- 0:16models and showing us where we are
- 0:23headed. And at this very moment, a
- 0:26surprising event occurred. Jensen Huang
- 0:28has suddenly stepped down from the CEO
- 0:30position. NVIDIA appointed its CFO as
- 0:33interim CEO and began searching for new
- 0:35leadership.
- 0:36Now, Jensen Huang was not just a typical
- 0:38CEO of a semiconductor company.
- 0:41For the past few years, NVIDIA has been
- 0:43at the center of the AI industry, and
- 0:46Jensen Huang was at its core. Whether
- 0:48OpenAI announced a new model or Google
- 0:50expanded its AI data centers, NVIDIA's
- 0:52GPUs were essential, and NVIDIA remained
- 0:55in the spotlight.
- 0:56Jensen Huang appeared every time the
- 0:58future of the AI industry was discussed
- 1:00, and he recently went around personally
- 1:01explaining this future in detail.
- 1:03It was as if he were the leader guiding
- 1:06the entire AI industry.
- 1:07But suddenly, that figure has
- 1:09disappeared. Now, will he be able to
- 1:10return? Of course, it’s not that
- 1:12NVIDIA’s GPU technology has vanished.
- 1:14CUDA is still there, and their
- 1:15world-class semiconductor development
- 1:17talent remains in place. However,
- 1:19watching this event, one question came
- 1:21to mind. Was NVIDIA truly the master of
- 1:24the AI industry? And will AI companies
- 1:26have no choice but to continue relying
- 1:28on NVIDIA in the future? Let’s examine
- 1:30this issue from a broader perspective.
- 1:33Let's look at how the AI industry is
- 1:35actually moving at the moment. They say
- 1:38the US and China are currently engaged
- 1:39in a war for AI supremacy.
- 1:41In the US, world-class AI models are
- 1:43being developed, creating new markets
- 1:45through AI services.
- 1:46There are the companies creating those
- 1:49AI models, and NVIDIA sits at the
- 1:51hardware core, the GPU. They supply the
- 1:53GPUs, networking, and software ecosystem
- 1:56needed to train and run AI models.
- 1:58But who actually makes NVIDIA's chips?
- 2:02It is Taiwan’s TSMC. They produce
- 2:04advanced semiconductors and even handle
- 2:06high-performance packaging.
- 2:08And the places that assemble these
- 2:10semiconductors into actual AI servers,
- 2:12like Foxconn and Quanta, play a vital
- 2:14role.
- 2:16And recently, the importance of Korean
- 2:17companies has also grown. It is the HBM
- 2:19supplied by SK Hynix and Samsung
- 2:21Electronics.
- 2:23There is even talk that even if a GPU is
- 2:25fast, if it cannot receive data quickly
- 2:27enough, HBM is what truly determines the
- 2:29actual performance.
- 2:31It has become a core component.
- 2:33Now, while the US is leading in AI
- 2:35models and semiconductor design, the
- 2:37hardware that actually runs the AI
- 2:39depends on a global manufacturing supply
- 2:41chain, including Taiwan and South Korea
- 2:44.
- 2:44Now, on the other side of this, there is
- 2:46China.
- 2:48China is developing its own AI
- 2:49semiconductors centered around Huawei,
- 2:51while expanding servers, data centers,
- 2:52and power infrastructure.
- 2:54China is responding by combining its
- 2:56proprietary technology, manufacturing
- 2:58capabilities, and low infrastructure
- 3:00costs.
- 3:01However, Jensen Huang has been a very
- 3:03special figure in this complex supply
- 3:04chain.
- 3:05This is because he has long connected
- 3:08U.S. AI companies with semiconductor and
- 3:10server manufacturers in Taiwan. The
- 3:12relationship with TSMC was not built
- 3:14over a short period, either. For decades
- 3:16, they have developed technology
- 3:18together, secured production capacity,
- 3:20and pioneered new markets. Such
- 3:22relationships are vital, especially in
- 3:24markets like the AI industry where
- 3:26demand is surging rapidly.
- 3:29This is because having just a GPU does
- 3:31not mean you can build an AI data center
- 3:32. You must secure semiconductor
- 3:34production, packaging, memory, servers,
- 3:36and even network equipment on time. The
- 3:38ability to coordinate all of these is
- 3:40necessary. Jensen Huang has been
- 3:42performing that very role. But now,
- 3:44things are set to change. Of course,
- 3:46technical agreements and cooperation
- 3:48will continue. The question is how
- 3:50naturally the trust and network built by
- 3:53Jensen Huang personally will continue
- 3:56into the future. Now, there will clearly
- 3:58be some changes. This means it is a more
- 4:01significant issue than a simple CEO
- 4:03replacement.
- 4:04From the perspective of AI supremacy,
- 4:06Nvidia was not just a company selling
- 4:09GPUs, but a company that connected the
- 4:11global AI ecosystem into a single system
- 4:14.
- 4:15And at the core of this, there is Jensen
- 4:17Huang.
- 4:18Now, interestingly, a new company has
- 4:20suddenly appeared here.
- 4:22It is Anthropic. Most AI companies have
- 4:25competed over how many of the latest
- 4:27Nvidia GPUs they could secure. However,
- 4:30Anthropic's strategy was a little
- 4:31different. They also use Amazon's
- 4:34Trainium and Google's TPU. Of course,
- 4:36they use Nvidia GPUs as well. But on top
- 4:39of that, they added AMD's AI
- 4:41accelerators. In particular, they
- 4:43decided to adopt 4 th-generation AI
- 4:45computing infrastructure from AMD on a
- 4:47scale of up to 2 GW. 2 GW is not a small
- 4:50scale. It is a power capacity level
- 4:53equivalent to several large power plants
- 4:55. Anthropic did not say they would stop
- 4:57using Nvidia. But they did say they
- 4:59would break free from situations where
- 5:01they*had*to use it. Why is this
- 5:03important? Until now, one of Nvidia's
- 5:05strongest competitive edges was CUDA,
- 5:08which is software. This is because the
- 5:10software ecosystem for developing and
- 5:13running AI models was built around
- 5:15Nvidia GPUs.
- 5:17But recently, AI itself has begun to
- 5:19help with the task of running and
- 5:22optimizing software on other
- 5:23semiconductors.
- 5:26Anthropic is also utilizing the cloud to
- 5:28push for software development and
- 5:30optimization for AMD GPUs.
- 5:32They are using AI to develop software
- 5:34for AI semiconductors. Now, the power of
- 5:37this AI is emerging one by one through
- 5:40agents, as I said.
- 5:43All software is becoming a target, and
- 5:45especially the software that causes
- 5:47bottlenecks and critical software will
- 5:49all be touched by AI.
- 5:51As these technologies advance, the cost
- 5:53and time to switch semiconductor
- 5:55suppliers are likely to decrease.
- 5:57Of course, this alone does not mean
- 5:59NVIDIA's technological edge will vanish
- 6:01. NVIDIA won't just sit still either.
- 6:03But from the perspective of AI companies
- 6:05, their options are increasing. There
- 6:06are many bottlenecks right now.
- 6:09For AI firms, hardware is difficult, but
- 6:10they are actually starting to enter that
- 6:12space, and all software-related things
- 6:14are becoming targets.
- 6:17Tasks favorable to Amazon's chips can
- 6:18run on Amazon EC2, and tasks efficient
- 6:20on Google TPUs can run on Google.
- 6:23They should also be able to select
- 6:24between tasks where AMD or NVIDIA have
- 6:26the advantage. If this happens, the
- 6:28negotiation structure between
- 6:29semiconductor suppliers and AI model
- 6:31firms could change.
- 6:33Here, the figure Dario Amodei is also
- 6:35important. Dario previously worked at
- 6:38the U.S. AI lab of China's Baidu.
- 6:41At the time, Baidu led an incredible
- 6:43group of AI talent, but they have faded
- 6:45a bit now.
- 6:46From the early days of AI, China was
- 6:48investing significantly. He studied
- 6:51speech recognition in a team led by
- 6:53Andrew Ng, and after passing through
- 6:55Google and OpenAI, he founded Anthropic
- 6:58. For Dario, the sibling structure is
- 7:00also important. Essentially, he is
- 7:02someone who has directly experienced
- 7:04China's AI research ecosystem. So, the
- 7:06AI industry has a quite complex
- 7:07structure, or should I say, a limited
- 7:10cast of characters?
- 7:11Now, Dario has shown a strong stance on
- 7:14controlling exports of advanced AI chips
- 7:16to China.
- 7:18On the other hand, Jensen Huang has
- 7:19emphasized the global spread of U.S. AI
- 7:21technology and the importance of access
- 7:22to the Chinese market. It is a somewhat
- 7:24opposing stance. This difference cannot
- 7:26be explained solely by political
- 7:28background, as the business structures
- 7:30of the two companies are different.
- 7:32It is the difference between
- 7:33semiconductors and AI models.
- 7:36NVIDIA may want a wider semiconductor
- 7:38market, while Anthropic may judge it
- 7:41strategically advantageous to limit
- 7:43rival nations 'access to advanced
- 7:45computing.
- 7:46Here, a new face of the AI supremacy war
- 7:48is being revealed.
- 7:50AI companies are no longer just
- 7:52customers buying NVIDIA GPUs, but are
- 7:54becoming entities that choose
- 7:55semiconductor supply chains, influence
- 7:57tech policy, and build their own
- 7:59computing strategies.
- 8:02So then, how is China responding? One of
- 8:05China's most powerful weapons is cost.
- 8:09We know that the recent cost of building
- 8:11AI data centers in China is much lower
- 8:13than in the U.S., but just how much
- 8:15lower?
- 8:16The recent comparison region is Inner
- 8:18Mongolia. It is a lower-cost area in
- 8:21China, and they say it's about 1/5th the
- 8:23cost. Of course, that's excluding GPUs.
- 8:27Since it mainly compares construction
- 8:28and infrastructure costs, China’s
- 8:30ability to leverage cheaper construction
- 8:32, power infrastructure, and supply
- 8:33chains is a key competitive edge.
- 8:36However, there is a decisive problem
- 8:37here.
- 8:39No matter how cheaply you build data
- 8:41centers, you can't run state-of-the-art
- 8:43AI models efficiently without the latest
- 8:45AI semiconductors.
- 8:47Now, evidence of this is emerging from
- 8:50various sources. Jensen Huang's recent
- 8:52issue is also due to high-end GPUs
- 8:54leaking into China. If Huawei chips were
- 8:56superior, they wouldn't need Nvidia GPUs
- 8:58, right? From China's perspective. But
- 9:00that is not the case.
- 9:02That’s why Chinese companies have been
- 9:04keen on Nvidia's latest GPU computing
- 9:06resources even while developing their
- 9:07own chips.
- 9:09That is one of the reasons why regions
- 9:11like Malaysia and Singapore have become
- 9:13important recently.
- 9:15Now, what happens if the U.S. starts
- 9:17restricting Chinese companies' use of
- 9:19overseas cloud services?
- 9:20Chinese firms may find it difficult to
- 9:22access GPUs secured abroad. The bigger
- 9:25problem lies with the operators who
- 9:26built the data centers.
- 9:28While Southeast Asian data centers
- 9:30recently gained attention, sentiment is
- 9:32shifting rapidly toward them being at
- 9:34risk due to this event.
- 9:36The situation will depend on reliance on
- 9:38Chinese clients, contract terms, and the
- 9:40ability to secure other customers. This
- 9:42shows that the competitiveness of an AI
- 9:44data center isn't decided solely by
- 9:45construction costs. Access to
- 9:47semiconductors, power prices, financial
- 9:49structures, client mix, and
- 9:51international regulations all play a
- 9:53part.
- 9:54This is exactly why we need to examine
- 9:56Elon Musk’s SpaceX strategy.
- 9:59Musk is developing AI models, but he is
- 10:02even more focused on building
- 10:03large-scale data centers.
- 10:05Tesla is developing its own AI
- 10:07semiconductors for autonomous driving
- 10:09and robotics. While these are separate
- 10:11businesses, they share a common
- 10:12direction.
- 10:13They are trying to secure the necessary
- 10:16computing, energy, networks, and
- 10:18hardware for AI more directly.
- 10:21This explains why Elon Musk is obsessed
- 10:24with building data centers quickly,
- 10:26securing power, and investing in chip
- 10:28design and manufacturing.
- 10:31Improving AI model performance alone is
- 10:33not enough. The cost of training and
- 10:35running those models is what matters.
- 10:38And in the future, AI will operate in
- 10:40cars, robots, and industrial facilities
- 10:43—meaning it will be integrated with
- 10:45physical hardware.
- 10:47The AI race no longer ends with who
- 10:49builds the smarter model.
- 10:52It’s becoming about who can operate
- 10:54those models more affordably and
- 10:55reliably.
- 10:57I believe this is exactly why we should
- 10:59look at Anthropic and Musk’s companies
- 11:01together.
- 11:03Anthropic is trying to reduce its
- 11:05reliance on suppliers by utilizing
- 11:06various semiconductors, while Musk is
- 11:08directly securing more infrastructure
- 11:10and physical systems to run AI.
- 11:13Both strategies stem from the
- 11:15realization that AI models alone are not
- 11:17enough.
- 11:19However, Nvidia has also been moving
- 11:20into the territory of AI model companies
- 11:22.
- 11:23Beyond building faster GPUs, they have
- 11:26been specifically emphasizing the
- 11:28development of AI platforms for robots.
- 11:30Physical AI. They have approached
- 11:33software from a more advanced level.
- 11:34They have been heavily investing in
- 11:36simulation and training environments for
- 11:38autonomous driving, known as "world
- 11:40models." World models are AI
- 11:41technologies that understand and predict
- 11:43the movements and physical changes of
- 11:44the real world. Autonomous driving is
- 11:46also a technology contained within world
- 11:48models. These are important parts of
- 11:50Physical AI.
- 11:51What does it mean that Nvidia has
- 11:53developed this technology?
- 11:55Up until now, their core role was
- 11:57supplying semiconductors, but moving
- 11:59forward, they intended to provide
- 12:01software platforms that help AI
- 12:02understand and operate in the real world
- 12:05. This is where their business areas
- 12:06begin to overlap with existing AI
- 12:08companies.
- 12:09Let’s imagine a robot company adopting
- 12:13AI.
- 12:13They could use Nvidia’s robot AI
- 12:15platform, or they could utilize
- 12:16technology from other AI model companies
- 12:18. Whenever a new AI model is released, a
- 12:22major concern in the robotics industry
- 12:25is how well it makes the robot operate.
- 12:28Unitree’s Wang Xingxing recently proved
- 12:31this. Such choices will become even more
- 12:33important in the future.
- 12:34If that happens, can Nvidia and existing
- 12:37AI companies continue to only cooperate
- 12:39? That might not be the case. They will
- 12:41end up competing with each other.
- 12:43Conversely, AI model companies are
- 12:45developing their own semiconductors and
- 12:47moving toward utilizing various types of
- 12:50chips.
- 12:51Now, the boundaries between companies
- 12:52are breaking down.
- 12:53I think this is a more significant
- 12:55long-term change than Jensen Huang’s
- 12:58eventual retirement.
- 13:00Why have I been keeping a close eye on
- 13:02Anthropic and SpaceX recently?
- 13:04It is the strengths in software and AI
- 13:06models versus hardware between Anthropic
- 13:08and SpaceX, as well as the respective
- 13:10strengths of China and the US.
- 13:13The common thread among all these
- 13:15companies and nations is that they are
- 13:17all transcending their existing business
- 13:19domains.
- 13:21When China steals Anthropic’s latest
- 13:23models to build their own, it is
- 13:25ethically problematic, to say the least
- 13:27.
- 13:29They are essentially taking something
- 13:30that required massive investment and
- 13:31getting it for free.
- 13:33However, the reason behind these actions
- 13:36is hegemony; while there is a war for
- 13:38hegemony between nations, the core
- 13:41technology here is AI, and as this AI
- 13:43becomes physical AI, it begins to
- 13:45influence hardware and actual physical
- 13:48situations.
- 13:50Things are no longer happening just
- 13:52within networking. In the future, it
- 13:54will be possible to have a direct
- 13:56physical impact or influence.
- 13:58Manufacturing companies are trying to
- 14:00change actual industries using AI. This
- 14:02means that AI competition is going to
- 14:05change. Even if you secure many GPUs, if
- 14:07power costs are too high,
- 14:08competitiveness drops, and even if an AI
- 14:10model has excellent performance, it
- 14:12could be limited if it is difficult to
- 14:13apply to actual industries. It might not
- 14:16be chosen by the market.
- 14:18Conversely, if you combine appropriate
- 14:19semiconductors, affordable power, and
- 14:21efficient AI models, you can provide
- 14:23services at a much lower cost.
- 14:26And here, in the era of physical AI,
- 14:28actual hardware like robots, automobiles
- 14:30, and factories is being added.
- 14:32Now, AI is evolving from a technology
- 14:34that simply answers questions into a
- 14:36technology that moves the real world. In
- 14:39that process, the integration of brains
- 14:41and hardware becomes even more important
- 14:43. Now, AI and hardware exist separately
- 14:46, but how do we combine them?
- 14:50I expect that cooperation between
- 14:51companies that develop chips and
- 14:53companies that develop AI models will
- 14:55increase.
- 14:57Now, Jensen Huang stepping down doesn't
- 14:59mean Nvidia is over, and he could even
- 15:01return.
- 15:02But the point is that there is a reason
- 15:04why this situation has arisen.
- 15:06It shows the current AI industry
- 15:08evolving from a single center called
- 15:09Nvidia into a more complex and fierce
- 15:11competitive structure.
- 15:14Even if Jensen Huang's retirement did
- 15:16not directly cause this change, it is
- 15:18clearly an important event that makes us
- 15:20rethink the existing order.
- 15:23Nvidia did, in fact, open the door to
- 15:25the AI era, but the master of that era
- 15:28does not necessarily have to be Nvidia.
- 15:31I think this is one of the most
- 15:33important changes that will take place
- 15:34in the AI industry going forward.
- 15:37The real AI hegemony war is just
- 15:39beginning now.
About this transcript
This page contains the full transcript of 젠슨 황 퇴진, AI 패권 전쟁의 서막! 엔비디아가 주인이 아니었다… 앤트로픽·스페이스X가 준비하는 새로운 AI 시대 by 엔지니어TV, generated from the public captions YouTube serves with the video. The transcript has 2,537 words across 454 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.