Home
Đăng nhập
Đăng ký
Nội dung học
Loading...
Luyện nghe
Luyện nghe
/
Video
/
The Infographics Show
/
Nvidia Just Bought The US Government For $4.95 Million
Nvidia Just Bought The US Government For $4.95 Million
Chọn chế độ học:
Xem phụ đề
Chọn từ
Viết lại từ
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
Phụ đề (244)
0:00
America sold China the most powerful AI chips on Earth… for a cut of the profits.
0:05
That’s not a conspiracy. It’s official policy. After years of promising to cripple China’s AI
0:11
industry Washington reversed course and allowed Nvidia’s chips for direct
0:15
sale to Beijing. No espionage or cyberattacks. Just a lobbyist and ink drying on a contract.
0:22
With a flick of a pen, the most strategically important technology of the 21st century landed
0:26
back in the hands of America’s biggest rival. Because the AI Cold War was never really
0:31
about stopping China. It was about who gets paid.
0:35
Chapter 1: The Beijing Reversal The press release that announced the H200 deal in
0:40
December 2025 reads like ordinary trade paperwork. At first glance.
0:45
Read it again, and something starts to feel different. Buried inside what is basically
0:49
a one-page memo is one of the larger policy U-turns in American technology history. The
0:54
kind of pivot that would normally roll out over months. Instead, it landed almost out
0:59
of nowhere. The U.S. government would now take a cut of every Nvidia H200 chip shipped across
1:05
the Pacific. A quarter on every dollar. The H200 isn’t something you’d find in a
1:10
gaming PC. This is different. It’s the kind of hardware built for one job..
1:15
Training advanced AI systems at scale. There’s so much raw data passing through
1:20
it that it behaves less like a chip and more like a small factory for intelligence.
1:24
Stack a few thousand together and you get an industrial compute cluster that can
1:28
train elite AI models in weeks rather than years. Under U.S. export controls, that kind of system
1:33
was never meant to end up in places like China. At least, that was the rule.
1:38
But before we go any further, imagine this.
1:40
You’re online every single day. You check your email, open a few apps, look things up,
1:44
stream videos, maybe even do all of that while traveling. And to you, it feels totally normal.
1:49
But behind the scenes, your internet provider, advertisers, network admins, and sometimes
1:54
even governments can build a surprisingly detailed picture of what you’re doing online.
1:58
Now, to be clear, no VPN can protect you from everything. You still have to be smart online.
2:02
Don’t click suspicious links, don’t hand over personal information to sketchy emails,
2:07
and definitely don’t trust the so-called Prince of Nigeria.
2:10
But a good VPN is an important layer of privacy, because privacy shouldn’t be something you only
2:14
think about after something goes wrong. It should be the default.
2:17
And that’s exactly what Proton VPN is built for.
2:20
Proton VPN helps keep your browsing private wherever you are, whether you’re at home,
2:24
traveling, or just trying to stop your online activity from
2:27
being tracked. Their no-logs policy has been verified by independent auditors,
2:31
and Proton is backed by a foundation dedicated to privacy, transparency, and user rights.
2:36
Proton VPN is also fully open source, meaning its code is publicly available for inspection. So
2:41
instead of just asking you to trust them, Proton gives people a way to verify what they’re doing.
2:46
And privacy doesn’t have to mean slow. Proton VPN
2:48
offers high-speed connections with VPN Accelerator, plus NetShield,
2:52
which blocks malicious ads and trackers before they ruin your browsing experience.
2:56
It also gives you more freedom online. You can connect through servers around the world,
3:00
access content more securely while traveling, and use it for streaming,
3:03
gaming, and P2P file sharing, all while keeping your online activity more private.
3:08
So whether you’re trying to stop advertisers from following you, keep your browsing private
3:12
from your ISP, or just take back control of your online life, Proton VPN makes privacy simple.
3:18
Click the link in the description or go to protonvpn.com/theinfographicsshow
3:22
to get up to 70% discount when you sign up to Proton VPN 2-year plan.
3:27
Back in 2023, the Bureau of Industry and Security drew a hard line. They are the
3:32
ones who decide what technology can cross which borders. The concern was simple.
3:36
China’s AI labs were making inroads in the AI arms race, and Washington was desperately
3:41
trying to slow So officials started limiting how much computing power could legally be exported.
3:46
The problem? The best chips were already way beyond those limits.
3:51
So companies like Nvidia did what companies do… they adapted. They built a downgraded version
3:56
designed to slip just under the rules, keeping the market open without crossing the line.
4:01
But even that workaround didn’t hold for long. Instead of a clean ban, Washington shifted
4:05
to something more flexible, a one-year waiver system with individual approvals. In practice,
4:11
the government itself would decide - case by case - who in China was allowed to buy advanced chips.
4:16
The press called it the “China Chip Review.” The list of who qualified is the real story.
4:21
Because the approved buyers weren’t obscure startups or low-risk firms. They included Alibaba,
4:26
ByteDance, and Tencent. There were also research institutions with links to the People's Liberation
4:31
Army (PLA), and state-backed cloud providers tied into China’s defense infrastructure.
4:36
The same organizations American policy had spent two administrations trying to starve
4:41
of compute were suddenly eligible buyers. Once the hardware leaves, the rules and
4:46
regulations don’t leave with it. Compute doesn’t stay supervised. A data center
4:50
doesn’t announce what it’s training. The United States was now collecting revenue
4:54
from selling hardware it had publicly called a national security threat just years before.
4:59
What’s most telling is how the people who built the original blockade found out it
5:04
had changed. According to later reporting, many of them found out the same way the
5:08
public did, through news agencies. Beijing didn’t sit still either.
5:12
Chinese regulators told state-linked firms to favor Huawei chips over American ones. But
5:17
Alibaba, ByteDance, and Tencent moved fast on the H200 anyway, with reported orders running
5:22
into the hundreds of thousands of units. So how did this move from unthinkable to
5:27
operational without a single congressional vote? Just follow the money trail.
5:32
Chapter 2: The Tithe In August 2025, an agreement was
5:36
already taking shape between Nvidia, AMD, and the executive branch. In exchange for export licenses,
5:41
the chipmakers agreed to send 15% of their Chinese chip revenue back to the U.S. government.
5:46
This wasn't a tax or a tariff. It was something genuinely new in
5:50
American trade history. A voluntary revenue share, negotiated directly
5:55
between private companies and the executive branch, all tied to approval for exports.
6:00
Access, in exchange for a cut. For Nvidia, the numbers are staggering. China
6:04
revenue from the H20 alone is measured in tens of billions. The 15% share, plus other commitments,
6:11
comes in at $12 billion. That’s in the same range as the entire annual economies of places
6:16
like Madagascar or the Bahamas. And none of it went through the usual political machinery.
6:22
What this really created was a precedent. Export controls stopped behaving like
6:26
hard rules and started behaving like pricing terms. Something you could adjust, negotiate,
6:32
trade against revenue. The December deal pushed it even further. With the more powerful chips,
6:36
the surcharge reportedly climbed to 25%. The pricing model became obvious.
6:41
Pay more, ship more sensitive hardware. The fine print added a 50% cap. Shipments to China
6:47
can’t exceed half of what’s sold in the U.S. And the whole arrangement runs on a one-year clock,
6:52
expiring in December 2026. After that, it gets renegotiated or rolled forward.
6:57
But by then, the precedent will already be set. A government that funds itself partly from a
7:02
corporation's exports has a financial interest in those exports continuing. And once that happens,
7:07
the incentive structure shifts. Blockades stop acting like a deterrence. They start
7:12
behaving like checkpoints that collect revenue. What was sold to the public as smart policy looks,
7:17
on closer look, more like a protection racket. To pull this off, Nvidia needed more than slides
7:22
and a corporate affairs team. It needed people who could rewrite the rules from inside the building.
7:28
And in 2025, it found them. Chapter 3: The Gatekeepers
7:32
Nvidia, for most of its corporate life, was almost an afterthought in Washington. In 2024,
7:37
its federal lobbying filings totaled roughly $640,000. By the standards of a corporation with
7:42
a market value north of $3 trillion, that figure was nothing. Pharmaceutical firms one-tenth its
7:48
size routinely spent 10 times that. In 2025, something changed.
7:53
Nvidia's lobbying spend climbed to $4.95 million, a jump of more than 600% in a single year. The
8:00
main goal wasn’t tax policy or antitrust. It was the Bureau of Industry and Security and
8:05
the export rules that decided which Chinese addresses could receive American silicon.
8:10
A key hire landed at Brownstein Hyatt Farber Schreck, the Denver-based lobbying firm that
8:15
has steadily embedded itself in Washington’s corridors. They created a dedicated team for
8:20
the Nvidia account led by Ed Royce. Royce isn’t a peripheral figure.
8:25
He spent years as chair of the House Foreign Affairs Committee, one of the key places where
8:30
U.S. policy on technology and rival states is shaped. Which means he didn’t just understand
8:35
export controls on paper. He understood how they actually move. Who writes the language before
8:40
it becomes policy. Which offices quietly steer decisions. And where the real pressure points sit.
8:45
It’s the kind of experience that turns regulation into something
8:48
you can work with, not just something you read. But companies hire former officials all the time,
8:53
lobbying numbers go up every year, so why is this any different?
8:57
The answer comes down to timing. The Brownstein hire, the lobbying surge,
9:01
and the policy shifts didn’t unfold across a multi-year arc. It was a matter of months.
9:06
Export categories that had been closed in May 2025 became unclear by August
9:11
and were basically open by December. There’s a version of this story where it's a coincidence.
9:16
But it doesn’t really read that way anymore. Because the record shows something too targeted
9:20
to ignore. Specific pressure on specific rules. The exact levers that end up moving.
9:26
What they delivered was something more valuable than meetings. A foothold inside the rooms where
9:31
licensing exceptions were decided. Once Nvidia's policy team had that foothold, the next part of
9:36
the machine did most of the work on its own. Chapter 4: The Corporate Loyalty Scorecard
9:42
According to Axios, a rating system exists within the West Wing, one that tracks corporate America.
9:47
It covers 553 companies and trade associations, scored on how hard each one worked to support
9:52
the administration's signature bill, the One Big Beautiful Bill. Companies are rated
9:57
as strong, moderate, or low supporters. The factors include social media posts,
10:01
press releases, ads, video testimonials, and attendance at White House events. Senior
10:06
officials told reporters the document is evolving to cover other priorities.
10:10
Those who score well are said to receive what staff describe as fast-track treatment.
10:15
From everything visible on the outside, Nvidia's score is exceptional.
10:19
Their separate $15 billion domestic commitment landed on the right side of every metric the
10:24
system appears to track. The company announced datacenters in politically meaningful states,
10:29
gave executives plenty of face time at administration events, and tied their massive
10:34
supply partnerships to the Stargate buildout. The connecting thread runs through
10:38
the chief of staff’s office. Susie Wiles, who managed the 2024
10:42
campaign before taking the role, came to the White House after a long career in political consulting
10:47
and lobbying work That background has shaped how some inside the system describe the current setup:
10:52
a West Wing where senior staff are closely familiar with how corporate and policy
10:57
interests move through federal decision-making. For Nvidia, the advantage doesn’t come from a
11:01
single approval or decision. It builds in the background, through how applications
11:05
are categorized before they’re even reviewed. Over time, that classification determines how
11:10
smoothly requests move through the system. A competitor who shows up to the licensing
11:15
window with a stronger technical case but a weaker scorecard rating does not
11:19
lose because their argument is wrong. They lose because the scorecard has pre-sorted the queue.
11:25
People familiar with the process describe it in simple terms:
11:27
by the time you’re debating the decision, the decision has often already been framed.
11:32
Chapter 5: The Green Channel Nvidia doesn’t just lead
11:35
the AI chip market. It dominates it. In the commercial sector used for training
11:40
and running advanced AI systems, Nvidia controls roughly 94% of the market. AMD holds 5%, and Intel
11:47
takes most of the rest. So when export approvals become case-by-case decisions, the system isn’t
11:52
really choosing between equal options. And that matters.
11:56
Because modern AI isn’t just about hardware, it’s about what that hardware runs on. Customer demand
12:02
is overwhelmingly for Nvidia silicon. That is what the Chinese labs have built their software stacks
12:07
around. Switching it out isn’t like swapping brands. It’s closer to rebuilding your entire
12:12
infrastructure while the system is still running. So even though, on paper, multiple suppliers
12:18
exist, in practice the demand funnels toward one. Nvidia did not need a clear policy granting it
12:23
preferred treatment. Its market dominance, combined with how deeply its software is
12:27
embedded in the industry, meant that any export pathway for advanced chips
12:31
would almost automatically route through it. So when a revenue-share system was introduced,
12:36
it didn’t need to explicitly pick winners. It just followed the path the market had already
12:41
carved. On paper, the arrangement looked neutral, something any supplier could,
12:45
in theory, participate in. In practice, there was only one meaningful option.
12:49
And everyone else was left outside watching the terms get set.
12:53
The knock-on effect makes the structure feed itself. Each chip shipped pushes Nvidia’s
12:58
products further inside Chinese datacenters and generates revenue. A percentage flows to
13:02
the U.S. Treasury, which strengthens the political case for continued shipments.
13:07
But the rulebook still officially rests on technical benchmarks written into law. S
13:12
o how, exactly, do you legally ship a chip that, by the explicit limits
13:16
set in 2023, should be banned? Chapter 6: The Loophole Factory
13:22
This is where the system gets clever, and where the technical details start to matter.
13:26
Inside the Bureau of Industry and Security, rules don’t only change through formal policy shifts.
13:31
They also evolve through technical questions and guidance notes. These are documents written to
13:36
explain how existing rules apply in practice. No hearings. No votes. Just clarifications that
13:42
shape how the rules work. It’s where policy gets adjusted in real time… without ever looking like
13:47
it was changed at all. And over time, those adjustments tend to move in one direction.
13:52
As senior meetings between Nvidia and U.S. officials take place, new guidance keeps
13:56
refining what counts as acceptable under export limits. Each update, on its own, looks minor. But
14:02
together, they don’t just interpret the rules. They move the boundary, just enough for
14:07
the chips to keep moving through. None of this is technically illegal.
14:11
Most of it is the system working as intended. The U.S. Department of Commerce
14:15
has the authority to interpret its own export rules. That interpretation happens constantly
14:20
through technical updates and guidance. The problem is what the combined effect
14:24
of dozens of small clarifications looks like once you step back. Stacked together,
14:29
they start to change what the rules actually are in practice. A fixed limit turns into a
14:34
negotiated threshold. Technical specs that were meant to be objective limits start reflecting
14:39
input from the very industry they govern. When a company can influence how the rules
14:43
governing its exports are interpreted, it’s no longer just being regulated in
14:47
a simple sense. It’s helping shape the rulebook it operates inside.
14:52
The memory in Nvidia’s H200 chip is more than just a memory spec. In frontier model training,
14:57
it’s the difference between a system that can hold a top-tier language model in active
15:02
context and one that can’t. It’s also the line between a chatbot and a cyberwarfare engine.
15:08
By engineering chips right up to the very edge of the rules, Nvidia is now writing the working
15:13
version of American export policy in real time. Chapter 7: The Shadow State Department
15:18
Jensen Huang's 2025 calendar reads like a study in parallel diplomacy.
15:23
The Nvidia CEO held high-profile international meetings in Taiwan, Japan, the United Arab
15:28
Emirates, Saudi Arabia, and the United Kingdom. At each stop, he announced compute commitments and
15:33
buildout deals at a scale that would historically have been worked out by trade representatives or
15:38
cabinet secretaries. In several cases, the agreements the CEO announced set the agenda
15:43
for the nation-to-nation conversations that followed, rather than the other way around.
15:48
The Gulf compute deals are perhaps the clearest example.
15:51
Saudi Arabia's announcement of a sovereign AI initiative, anchored by Nvidia hardware, set
15:56
the terms of the American foreign policy meetings that followed with the Kingdom. Nvidia built the
16:01
policy and the State Department adapted to it. The same pattern played out in Abu Dhabi. The
16:06
G42 partnerships and tied chip commitments came before, not after, the formal U.S.
16:11
rulebook for AI cooperation with the UAE. It’s not how foreign policy is supposed to work.
16:16
The Department of State exists to coordinate how the United States approaches strategic
16:20
technology and its transfer abroad. But now, a company can announce commercial
16:25
commitments first, and the government is left reacting afterward. They have to
16:29
adjust policy to match what’s already in motion. What emerges is an unusual kind of sovereignty.
16:35
Nvidia is not a state in the traditional sense. It has no army and no treasury. But
16:40
it does sit at the point where AI capability is actually determined through the chips that
16:44
define what systems can and cannot do. And it has enough political access to ensure its
16:50
commercial choices don’t just operate inside policy, they actively shape it.
16:55
That’s a different relationship than regulation alone.
16:57
It starts to look less like a company operating under rules,
17:00
and more like a company operating alongside the formation of those rules. Not replacing the state.
17:07
But, in key areas, setting the terms the state then responds to.
17:11
Chapter 8: The Reckoning The first H200s under the new waiver are
17:15
starting to land inside Chinese datacenters, with reported orders already running into hundreds of
17:21
thousands of units. The training runs will follow. The models that come out will end up applied to
17:26
use cases spanning the full range, from commercial chatbots to military targeting
17:31
systems. The compute, once delivered, becomes whatever the operator chooses to do with it.
17:35
What makes this hard to unwind is that it’s no longer just policy,
17:39
it’s income. Once the revenue starts flowing into federal planning, it stops being a clean
17:44
on/off switch. Federal budget projections include it. Domestic buildout programs partly funded by
17:50
it have constituencies that will defend it. The bureaucratic machine of the U.S. government has,
17:55
in effect, taken a financial stake in continued exports of the very hardware it once labeled a
18:00
controlled good. Reversing the deal would require not just a political decision but the removal of
18:06
revenue already locked into agency planning. When Nvidia raises its revenue projections,
18:11
the government's projected share rises with them. It only strengthens the political case
18:15
for expanding the waiver. The privatization of foreign policy stops being a metaphor and
18:20
starts being a balance sheet relationship. Once a company has shown that exemptions to
18:25
national security policy can be bought at a fixed price, the precedent becomes
18:29
permanent. What’s emerging here looks less like a one-off deal and more like a template
18:33
for turning regulatory pressure into a negotiated cut of revenue. And other industries are already
18:39
watching closely. Pharma, satellites, advanced biotech… sectors where the same product can
18:44
be both commercial and strategic. For decades, the rule was simple:
18:48
some technologies stay out of open markets. That rule hasn’t been rewritten.
18:52
It’s just… stopped being enforced in the same way. Nvidia is just one player in the AI race. And no
18:59
one wants to be the one that slows down. But an arms race only works if it’s sustainable. Right
19:04
now, the spending is starting to look less like growth and more like pressure building
19:07
in the system. And when it breaks, everyone feels it. Find out what happens next in “$115
19:13
Billion Burn Rate. The AI Bubble Just POPPED.” Or watch this video instead.
Nvidia Just Bought The US Government For $4.95 Million - Video học tiếng Anh
