bUt wE cAn"T lEt cHinA WiN tHe AI aRmS rAcE!!

0:00 Major AI companies have quickly and quietly

0:02 become some of the biggest lobbyists across America,

0:05 spending over $100 million to influence policy in the last year alone.

0:09 In what might be their first major

0:11 investment to actually generate a financial return,

0:13 OpenAI has massively increased its lobbying efforts year-over-year.

0:16 In addition to the usual suspects that have

0:19 also uped their political contributions and Nvidia,

0:21 who has typically been quiet in Washington,

0:24 quintupling its spending to make sure that their voice was heard.

0:28 And well, all of their well-funded

0:30 voices have overwhelmingly been saying one thing.

0:33 I mean, we can't let the Chinese beat us in AI.

0:37 We are going to be the dominant player

0:39 or China's going to be the dominant player.

0:41 But it's nothing compared to what will happen if

0:44 China beats the US on the ultimate goal of AGI, artificial general intelligence.

0:49 Yeah, that's right.

0:50 If we overregulate AI, China will win.

0:53 If we underregulate AI, China will win.

0:56 If the government doesn't support and financial

0:58 backs stop AI development, China will win.

1:01 If we sell cuttingedge chips to China, they will win.

1:04 But if we don't sell cutting edge chips to China, they will also win.

1:09 AI lobbying has become some of the most

1:12 effective in Washington because of China.

1:14 And well, you know what?

1:15 As dumb as it sounds, they might actually have a point.

1:18 Despite all of the talk around how

1:20 indispensable Nvidia is to the current AI ecosystem,

1:23 they themselves have admitted that their market

1:25 share in China is now effectively zero.

1:27 As local suppliers have developed their own

1:29 alternatives and that homegrown Chinese models are matching

1:31 or even exceeding American models while using

1:34 supposedly inferior hardware and using far fewer resources,

1:37 why would they admit to that if it wasn't true, right?

1:41 At the same time, it's also becoming increasingly clear that current AI models

1:45 might fall short in a lot of the applications they have been pitched on.

1:48 But they are becoming incredibly effective tools for mass surveillance,

1:52 war fighting, cyber warfare, and even just simple propaganda.

1:55 So, if this really is an existential threat

1:57 that we simply can't let China gain the upper

2:00 hand in, it is worth at least understanding

2:02 the hypocrisies around how we are dealing with it.

2:05 Citizens will be on their best behavior

2:07 because we're constantly recording and reporting everything.

2:10 For example, active cyber attacks attack a whole country.

2:13 Do it until everybody's dead.

2:15 And you can imagine that scenario.

2:16 You can also imagine the scenario where you say,

2:19 "I want to kill a million people.

2:20 Show me a biological path to do it." New

2:22 companies cannot comply with can't afford to comply with.

2:25 It's too complicated.

2:25 It's too hard to get to from a standing start.

2:27 And then what happens is that big company

2:29 now basically has a permanent government supported monopoly.

2:32 We believe that excessive regulation of the AI sector

2:35 could kill a transformative industry just as it's taking off.

2:40 Late last week, Kevin Olirri,

2:42 the method acting movie star who occasionally plays a businessman on TV,

2:45 announced that he would be building the world's largest data center in Utah,

2:49 a project that is slated to use more energy than the rest

2:52 of the state combined and cost tens of billions of dollars to build out.

2:56 Now, despite wearing two very expensive watches at the same time,

2:59 Kevin Olirri does not have tens of billions of dollars.

3:02 So most of his work has

3:04 been dedicated to securing favorable government approvals,

3:06 regulatory fasttracks,

3:07 and stateisssued tax exemptions with the vague hope that these can be

3:11 offered up to a consortium of investors who will ultimately fund this project.

3:15 Last week's announcement also came around 16 months after a similar announcement

3:19 by Olyri that he would be building the world's largest data center,

3:23 but this time in Alberta, albeit with a similar pitch around

3:27 an accommodating local government that would cut red

3:29 tape and hook them up to a practically unlimited supply of cheap natural gas.

3:34 Now, as a shock to basically nobody, to date,

3:37 there has been effectively no tangible progress on either of these projects.

3:40 Now, we have already made a video about how a large share

3:43 of data center construction sites have

3:44 been quietly cancelled over the last year.

3:46 And out of all of the data centers that will never get built,

3:50 these are surely the least buildiest of all.

3:52 But there is a chance, and I can't believe I'm going to say this, that Olyri

3:56 may not be as silly of a sausage as he seems.

3:59 Now, the only reason why these proposals are really worth

4:01 mentioning in the first place is because of the messaging

4:04 that he has been using to make several governments bend over

4:07 backwards to roll out the regulatory red carpet for a project.

4:10 they surely knew was never going to see the light of day.

4:13 Not in this case.

4:14 We're building power from scratch from the pipeline.

4:16 And by the time the pipeline is natural gas,

4:18 natural gas and we're going to burn it with turbines clean,

4:21 but we can also give back to the grid.

4:23 That's good for the community, but for the country,

4:25 we need to compete with China.

4:27 So yeah, it's clearly an effective message and for what it's worth,

4:30 we probably don't want China to take a commanding lead on this technology.

4:34 But there are three big dumb ironic problems

4:37 to how we are currently addressing that threat.

4:40 The first is that we are really only fighting this supposedly

4:43 existential arms race when it is profitable to do so.

4:46 The test of any threat narrative is whether the people

4:50 pushing it actually act like the threat is real.

4:53 And as it turns out, they really don't.

4:55 The same companies warning that a Chinese AI takeover is a generational

4:59 emergency are also the ones lobbying to sell their best chips into China,

5:02 the ones telling regulators their products are

5:04 too harmless to need a liability framework,

5:06 and the ones actively fighting any regulation

5:08 that would limit how fast they can scale.

5:11 So, these tools exist in some kind of techbro

5:13 superp position of simultaneously being the most existential threat

5:16 to American dominance since the Cold War while also

5:19 being too harmless to require any regulation within America itself.

5:23 It's a dumb message, but it's one that the industry is

5:27 spending an ungodly amount of money on amplifying.

5:29 Beyond the nine figures of direct political lobbying that require official

5:33 disclosures and major uh podcast acquisitions to promote their talking points,

5:37 certain companies in the space have also been exposed for running what are

5:41 referred to as dark money campaigns

5:42 that sit just outside of official regulations.

5:44 As an example, OpenAI founders and early investors have been

5:48 funneling millions of dollars into a group called Build American AI,

5:52 which is in turn redirecting that funding to paid

5:55 influencers to promote effectively the same narrative as Mr.

5:58 Wonderful, just with a little bit less transparency.

6:00 A AI is something that needs to be kept in America.

6:04 And B, we can't afford to let China win.

6:07 The end goal is pretty openly to pave the way

6:10 for a government bailout should open AI need it hypothetically.

6:14 Broadly speaking, they understand the AI is not politically popular.

6:18 Government bailouts are not particularly popular either,

6:20 but they are hoping that Chinese AI is even less popular.

6:24 Now, the channel Wall Street Millennial did

6:26 a great video on this earlier this week,

6:28 literally as I was putting this video together,

6:30 that went into detail about how they

6:32 are channeling this money and crafting this message.

6:34 So, I will leave a link to his video in the description alongside Taylor Loren,

6:39 the journalist influencer who originally broke the story when

6:41 she herself was offered money to repeat their talking points,

6:44 but decided to expose them instead.

6:46 Either way, it's pretty clear that this is happening,

6:49 and this is the message they are running with.

6:52 So, while we might not have a 9-figure lobbying budget to push back against

6:55 this, we do have basic logic to point out the flaws in their argument.

6:59 The first irony is the most basic one.

7:01 The same companies warning about Chinese dominance are also the ones

7:04 lobbying hardest to sell their cuttingedge technology directly into China.

7:07 For the better part of 2 years, the headline messaging out of Nvidia, OpenAI,

7:11 and most of the major AI labs has been that letting China

7:15 get ahead is a civilizational risk that the US simply cannot allow.

7:19 And then late last year, when it started to look like the AI

7:22 sector might be headed for a market correction,

7:25 uh that messaging got a little bit more flexible.

7:28 In December, the government announced that Nvidia would

7:32 once again be allowed to sell its H200

7:34 chips into China with the regulation formally taking

7:36 effect around the Department of Commerce in January.

7:39 The H200 is, by Nvidia's own description,

7:41 one of the most advanced production AI chips on the planet.

7:45 China bound shipments are technically capped at 50%

7:47 of US sales and routed through US territory for inspection.

7:50 But in practice, the cap on the H200 alone could allow up to 1

7:54 million units to flow into China with potentially another million H100s on top.

7:58 According to the Council on Foreign Relations,

8:00 that is enough silicon to build the largest AI

8:03 data center on Earth and to add roughly 2

8:05 and 1/2 times Chinese AI compute compared to a scenario

8:08 where they only had domestic chips to work with.

8:11 So the same year we were told that China

8:13 overtaking the US and AI was a generational threat.

8:16 We also decided that selling them the actual chips required to do exactly

8:19 that was perfectly acceptable as long

8:21 as the right American company was getting paid.

8:24 Now that Nvidia is allowed to sell into China again,

8:27 they are not exactly thanking everyone and getting on with it.

8:30 They are now lobbying to sell even more advanced chips on the basis

8:34 that the H200 isn't competitive enough on its

8:36 own to actually win them the Chinese market, which is impressive,

8:41 confusing, both when you remember that this is the same Nvidia that has spent 2

8:46 years explaining to investors that CUDA makes

8:48 their hardware essentially irreplaceable across the entire AI stack.

8:51 One of those two stories has to be

8:54 wrong because they directly contradict one another.

8:57 Either Nvidia chips are uniquely irreplaceable,

8:59 in which case Chinese labs will be paying through the nose

9:02 to use them no matter how stripped down they are, or they aren't.

9:05 In which case, the entire valuation of every AI company downstream of Nvidia

9:10 is built on a much shakier foundation than the market has been pricing in.

9:14 The contradiction does not stop with chips either.

9:16 The same companies that will on a Tuesday tell a Senate committee

9:20 that their products are essentially harmless

9:21 and don't need any kind of liability framework

9:24 will on a Wednesday tell a podcast host that these are the most

9:27 powerful creations in human history with a real

9:30 chance of wiping out the species.

9:32 Do you have a bunker?

9:34 I have like underground concrete heavy reinforced basement,

9:37 but I don't have anything I would call Hold on.

9:40 Hold on.

9:41 Hold on.

9:41 Dude, look, I'll let you I'll let you keep

9:43 me on the ropes in a lot of this conversation,

9:45 but I am going to call that out of the dang bunker, dude.

9:47 The harmless version comes out in front of regulators,

9:50 where the goal is to keep liability frameworks off the table.

9:52 The species ending version gets saved

9:54 for investor decks and government grant committees,

9:56 where bigger stakes mean bigger checks.

9:59 Both can't be true, but both are extremely useful depending

10:02 on which room the lobbyist happens to be standing in.

10:05 And that's just irony number one.

10:06 The next two are even harder to explain away and they

10:10 reveal a lot more about what this lobbying campaign is actually buying.

10:14 So, it's time to learn how many works to find

10:16 out why the loudest voices warning about losing this arms

10:18 race might be the exact same voices guaranteeing that we

10:22 do and what their version of winning actually looks like.

10:25 The biggest companies in the world are spending hundreds of millions of dollars

10:28 on lobbyists to make sure the AI race plays out in their favor.

10:32 And we are going to get into all of that.

10:34 But while we are doing that, a lot

10:36 of regular Americans are quietly going in the other direction,

10:39 building something for themselves that does not

10:41 depend on what way a senator votes.

10:43 New business applications have stayed near record highs

10:45 in the United States for 4 years straight.

10:48 And a lot of those start the same way

10:50 with somebody who has a design or an audience trying

10:52 to put it on a hoodie or a mug

10:55 or a tote bag without becoming a logistics company to do it.

10:57 This video is sponsored by Printify,

10:59 a print on demand platform that handles every part of merch.

11:02 Nobody actually wants to deal with the inventory,

11:05 the packaging, the shipping, and the returns.

11:07 You design what you want in their editor,

11:09 mock it up on the actual product, and that becomes your product.

11:13 Nothing exists physically until a customer buys it,

11:16 so there is no inventory order,

11:17 no warehouse, and no boxes piling up in your living room.

11:20 When the order comes in, Printify routes it

11:22 to one of their print partners near the customer,

11:25 prints it, and ships it directly.

11:26 It connects directly to Shopify, Etsy, Tik Tok Shop,

11:29 eBay, Wix, and a handful of other platforms.

11:32 So, your products show up wherever your audience already shops,

11:34 and you keep the profit between Printify's cost and what your customer pays.

11:38 You can even determine what you could make using their profit calculator.

11:41 So, if you have a design idea, a podcast, a community, or a brand,

11:45 this is about as low friction as it gets

11:47 to start a real business and an income on the side.

11:49 Let's start today building a business with Printify.

11:52 Use my link in the description to sign up for free

11:55 and use code how many works for 15% off your first order.

11:58 Only the first 100 people get this offer.

12:01 Okay, so the people giving the loudest warnings about

12:05 losing the AI race to China are also the people

12:08 lobbying hardest to sell China the chips and tools

12:10 that would in turn let them actually win it.

12:13 Not a great start, but the second irony is bigger because

12:17 it really isn't about who is selling what to who anymore.

12:20 It's about what kind of regulation is actually

12:22 getting pushed through under the banner of beating them.

12:25 Even if you accept that an AI arms race is a real and imminent threat,

12:28 almost none of the deregulation that has been pushed

12:31 through under that banner is actually aimed at winning it.

12:34 Most of the recent deregulatory effort has gone into two areas.

12:38 Environmental protections and energy prioritization,

12:40 both designed to get private sector data centers online faster.

12:44 The vast majority of the compute that those data

12:46 centers will produce is not going to advance defense applications,

12:50 secure cyber infrastructure,

12:51 or any of the surveillance and war fighting systems that actually

12:54 would matter in a real strategic competition with another superpower.

12:57 It is going to add targeting,

12:59 AI customer service replacements, image generators,

13:01 the 10th iteration of a chat product, and the fifth season of Fruit Love Island.

13:06 If this really were a national security priority,

13:08 the simple version of policy is straightforward.

13:11 narrow the scope of what counts as a strategic AI project,

13:15 fasttrack those, and treat the rest of the industry

13:17 like the consumer product business it actually is.

13:20 Instead, the policy has been to treat the entire

13:22 industry as if it were the Manhattan project.

13:25 It might not be the best analogy,

13:27 but something like steel clearly has civilian and military applications.

13:30 So, maintaining a ready supply of the stuff is a matter of national security.

13:34 But we aren't going to let a car manufacturer ignore safety standards just

13:38 because their products are made out of the same stuff as our artillery shells.

13:42 Again, I know it sounds dumb,

13:44 but the parts of regulation that have been most fiercely fought against

13:47 are the parts that have nothing to do with national security at all.

13:51 Liability for inappropriate imagery generated by these models,

13:54 mental health protection for kids being exposed to AI companions,

13:57 and actual responsibility for crimes these tools have been helped to facilitate,

14:01 if not directly encourage.

14:02 None of those rules would slow down

14:05 America's ability to build a defensegrade AI system.

14:07 If anything, forcing companies to actually clean up

14:10 the worst behavior of their models would probably

14:12 make them better at the things they are

14:14 pitching to the Pentagon in the first place.

14:15 The fact that those are the rules being lobbied against

14:18 the hardest tells you what this campaign is actually about,

14:21 and it has very little to do with China.

14:24 Now, if you genuinely thought you were

14:26 in an existential civilization ending arms race,

14:28 the obvious template is right there in the history books.

14:32 We have done arms races before and we have rules for them.

14:35 Industries with the kind of national security exposure these companies claim

14:39 to have are usually regulated like

14:41 weapons manufacturers or nuclear energy operators,

14:43 meaning strict licensing on supply chains,

14:46 mandatory tracking of where the technology ends up,

14:48 background checks on consumers,

14:49 and end use restrictions with criminal penalties for violations.

14:52 It might shock you to learn that none of the people warning that AI could end

14:56 humanity have actually asked to be regulated like

14:59 people who already built things that could end humanity.

15:01 Of course, they don't actually want that because that level

15:05 of oversight would massively limit their ability to scale,

15:08 take on capital from foreign sovereign wealth funds,

15:10 and throw a lot of impressive but ultimately unfounded

15:12 numbers around in the press to drive a hype.

15:15 The arms race framing, in other words, is really just a convenient line,

15:18 so long as it begins and ends with the government clearing the runway

15:22 for them and never actually putting them

15:24 under the kind of scrutiny it would imply.

15:26 Which is why, going back to where this video started,

15:29 Kevin Oolirri may be cooking something here.

15:31 His actual plan, as far as anyone can tell,

15:33 isn't really to build the world's largest AI data center in Utah,

15:37 or before that in Alberta.

15:38 it is to get on television, sell a story about competing with China to as many

15:43 state and local governments as he can find,

15:45 and then use that uh personality to stack up regulatory allowances,

15:49 tax exemptions, and lock in cheap natural gas contracts.

15:52 And then if a real consortium with actual

15:55 capital wants to come along and use those allowances,

15:58 he flips the package to them, potentially with some kind of perpetual royalty.

16:02 In the very likely event that nobody actually shows up,

16:05 he has been very careful to keep

16:08 his side of all of this remarkably non-committal.

16:10 Heads, he wins.

16:11 Tails, the taxpayers of two countries lose.

16:14 It's the Mr.

16:15 Wonderful metaphor for the whole industry.

16:17 Get the governments to bend over backwards to clear

16:19 the path under the banner of beating China,

16:21 pocket the value of those allowances,

16:23 and worry about whether the underlying project ever ships later.

16:26 It sounds bad, but it gets dumber.

16:29 The China narrative has been effective at getting

16:32 governments on board to cut red tape,

16:34 but so far this is ignoring the biggest irony of all.

16:37 China is regulating its AI sector like crazy.

16:40 China's amended cyber security law,

16:42 which took effect on January 1st of this year,

16:45 included a dedicated AI compliance provision

16:47 that puts firms on the hook for ethics, risk monitoring, and safety assessments.

16:51 The US has no equivalent federal law and a real chunk of the current lobbying

16:56 effort is dedicated specifically to making sure

16:58 no state level law fills that gap either.

17:01 China requires mandatory algorithm registration for any

17:03 service offering recommendations or generative content.

17:05 Companies have to file the name of the service,

17:08 the application domain, the algorithm type, and a self- assessment report.

17:13 The US has no equivalent.

17:15 China made AI generated content labeling mandatory across text, images, audio,

17:19 and video with rules that came into force on September 1st last year.

17:23 The US has only proposed it.

17:25 China has published a set of national standards on data security,

17:29 pre-training data security,

17:30 and basic security requirements for generative AI services,

17:33 all of which became enforceable last November.

17:36 Last month, a Chinese court in Hao

17:38 ruled that companies cannot fire workers solely

17:40 to replace them with AI and that any

17:43 restructuring on those grounds requires retraining offers,

17:46 fair reassignment, and proper severance.

17:47 The US has had no comparable ruling in part because there

17:52 is essentially no statutory basis from which one could even be made.

17:55 It is also worth pointing out that overall investment from China

17:59 into AI is considerably lower than it is here in America.

18:03 So, if this is really a race,

18:05 spending alone isn't going to be the deciding factor.

18:07 Now, to be clear, China is not exactly a model citizen of any of this.

18:12 And there is plenty to criticize about how these laws actually get applied.

18:16 Enforcement is real, but selective and politically motivated,

18:19 with rules getting deployed against companies that fall out

18:22 of favor and conveniently ignored for companies that haven't.

18:25 Many provisions are written vaguely enough that they can really

18:28 only be enforced when the party feels like enforcing them.

18:31 And most of the financial penalties by tech

18:34 company standards are small enough to be priced in as a cost of doing

18:37 business rather than treated as a serious deterrent.

18:40 So no, China is very obviously not running a civics class on AI ethics here.

18:45 A lot of these rules are functionally about state control

18:48 of information rather than user safety in any meaningful sense.

18:51 But that is sort of the point.

18:54 China is strongarmming its tech sector into developing AI in a way

18:58 that for better or worse forwards the interests of the state.

19:01 America is told that the way to beat this is to let its tech sector

19:05 do whatever they want in the vague

19:06 hope that private interest and geopolitical interest align.

19:09 From the market's point of view, this is already working exactly as intended.

19:13 And if you want further proof of that, go watch this video

19:17 next to find out how OpenAI's problems are already becoming our problems.

19:21 And don't forget to like and subscribe to keep on learning how money works.