How Trump’s AI Chip Deal Undermines National Security

Written By Leeja Miller

Further Reading:

Empire of AI by Karen Hao

More Everything Forever by Adam Becker

The AI Con by Emily M. Bender and Alex Hanna

Sources:

Transcript:

As Trump attends talks with Putin in Alaska, today I want to focus on a different major world power, China, and on the deal that Trump just struck with chip providers Nvidia and AMD to allow them to export chips used for AI to China, so long as the US gets a 15% cut of the exports. We’re gonna talk about why this is an unprecedented move, what it means for the AI race, but also what these chips even are, why they matter, and what to expect from this global AI race moving forward.

First, the latest on the Nvidia chip saga. Nvidia’s got the world’s most advanced chip needed for AI, it’s called the H100. They started shipping out those chips in October 2022, right as the ChatGPT launch revolutionized how we use the internet forever. Demand for those chips soared, Nvidia regularly ranks now as the most valuable publicly traded company in the world. During the Biden administration, Nvidia was banned from exporting the H100 chips to China in an attempt to thwart China’s access to advanced AI capabilities–something that could help them with everything from advanced weapons manufacturing to curing disease and modeling climate change, across the Chinese economy. This was actually building off of policy from the first Trump administration that took aim at Chinese semiconductor companies like Huawei to try to thwart the advancement of 5G wireless infrastructure that could be used by China for spying or further human rights abuses. Basically, semiconductors and chips could pose a national security threat in the US if China developed the tech faster than we did.

AD

Often, the way news outlets report on a story differs depending on the bias of the publication. And it’s hard to know how YOUR own perception of bias influences your interpretation of the news. This headline from PBS Newshour caught my eye “Under new, unusual agreement, U.S. will get a 15% cut of Nvidia and AMD chip sales to China.” Using the Ground News browser extension, I can see that in the US PBS is considered “left leaning.” To get a fuller picture of the story on all sides of the political spectrum I can click on Full Coverage, which will show me coverage of the same story from publications across the political spectrum.

My partner on today’s video, Ground News, uses 3 independent monitoring organizations to assess the bias rating for each publication. Each news monitoring organization has their own methodology - including editorial reviews, blind bias surveys, independent reviews, and third party research. These are legit INDEPENDENT organizations, but they are limited by the data available to them: their analysis is done in the context of the U.S. political system. Our overton window has shifted so far right that PBS is considered “left leaning.” That tells you a lot about the context within which the media you’re consuming operates. And thanks to Ground News, I get even more context, including how factual each publication is and who owns it, giving me a well-rounded idea of the motivations, biases, and accuracy of the news I’m consuming. Wouldn’t it be great if everyone in America had that much information about the news they consume??

Because depending on where you or your family members get the news, you’re going to get very different takes on the same story.

This is where Ground News comes in - and why I've been using them for over a year. Today’s partner Ground News is an app and website that offers tools to help you critically analyze the news you read, providing context to understand the full picture. I feel better equipped to make sense of what’s happening in the world without being influenced by just one perspective. And listen I’m not the only one who loves Ground News, the Nobel Peace Center even called it "an excellent way to stay informed, avoid echo chambers, and expand your worldview.”

I’m always really impressed with Ground News and genuinely think they’re a great resource. If you want to stay informed on US Politics and more Subscribe through this QR code or our link below at ground dot news slash leeja for 40% off this unlimited access Vantage Plan. It comes out to $5 a month for unlimited access to ALL the features I’ve mentioned. Thanks Ground News!

So, to further the Trump-era policy, Biden banned NVIDIA exports to China of H100s. In response, NVIDIA created H20 chips to try to get around the ban, saying they’re less powerful. And in some ways they are, they can’t train AI as well as the H100s. BUT H20s are actually MORE powerful in some regards–they are actually faster at processing AI queries. So when Trump regained the presidency he blocked exports of H20s, because they’re yes technically not H100s but they’re still giving advanced tech to China. Now that’s not to say that China has ZERO access to H100s or H20s, they just have to get them through illicit means, which makes it much more expensive and time consuming, and not really a realistic way to develop their AI infrastructure. China is smuggling the chips in through waystations like Malaysia and Thailand where the chips are legal.

Then in July Trump flip flopped after trade talks with China and with Nvidia CEO Jensen Huang, one of the richest men on earth, and now Trump will allow the export of H20s, not the H100s but the H20s, in exchange for Chinese exports of rare earth metals. And now this month Trump announced that the US wants to take a 15% cut of the exports of those chips from NVIDIA, as well as from AMD, another chipmaker, in exchange for allowing those exports to move forward. This is likely in violation of the constitution’s ban on export taxes–under the constitution the government cannot impose taxes on exports of goods between states or from the US abroad. It is also likely a violation of the US Export Control Reform Act which prohibits the government from charging a fee in exchange for an export license. Now typically Trump flip flopping on things, Trump seeking more money as leverage in exchange for getting what he wants, tariffs and taxes, even blatantly breaking the law, this has become frankly old news during this Trump administration. But this latest flip flop on the Nvidia chips and agreement to play ball with Nvidia in exchange for allowing exports to China is particularly concerning and unprecedented because of the essential nature of these chips for AI, and because of the potential vast and largely unknown consequences for our national security should China get these chips and use them to surpass us in AI innovation.

According to Vox, quote “When Trump chickens out on export controls on Nvidia chips, he made a decision with massive global ramifications, altering the likely future balance of US and Chinese power.”

Many China hawks are voicing fierce dissent, saying that this poses a threat to our national security, and that selling national security risks for a price means that national security in the US can be bought and sold for the right fee. The reason this poses a security threat is that export controls like this are specifically designed to keep other countries from gaining access to technology that could be used militarily against the United States. As relations between the US and China change or disintegrate, Chinese access to the power that AI gives them–everything from military weapons to science and technology innovations, could create a power imbalance that impacts the entire world.

"You either have a national security problem or you don't," said Deborah Elms, head of trade policy at the Hinrich Foundation.

"If you have a 15% payment, it doesn't somehow eliminate the national security issue," she added.

Former US trade official Stephen Olson told Politico quote ““To call this unusual or unprecedented would be a staggering understatement. What we are seeing is in effect the monetization of U.S. trade policy in which U.S. companies must pay the U.S. government for permission to export. If that’s the case, we’ve entered into a new and dangerous world.”

Now let me just add a little caveat here that the messaging we’re getting from the officials within the federal government are based on decades-old anti-communist, anti-China rhetoric. And while there are certainly things to critique about the Chinese government, not least of which their vast human rights abuses, I want to at least point out that we have collectively been fed vast amounts of anti-Chinese propaganda in this country over decades that all may or may not be true but often paints a very black and white picture of us versus China that likely isn’t 100% based in reality, ok. Not to say this isn’t incredibly important, and US relations with China ARE deeply important because of the global power both countries hold, this is all just to say that anti-Chinese rhetoric from our Government likely doesn’t paint the full picture.

Ok. To understand why this is important, I think it’s important to cover some basics about how AI models work and the role chips play in all of this. You probably already know that they are trained by being fed astronomical amounts of man-made art, writing, etc. We’re talking pretty much all recorded human writing in the English language has been fed through AI models at this point. That's a ton of information and it must be processed in and out of thousands of graphics processing units or GPUs. Nvidia and AMD are the two major GPU creators in the US. The computer you use likely has either Nvidia or AMD tech inside it to power everything your computer does. When we talk about NVIDIA chips used in AI, we’re talking about GPUs and the circuits in those chips that are designed for the kinds of calculations used in AI. Nvidia has invested billions of dollars to improve the performance of its GPUs literally 4000 fold over the past 10 years alone. Chip supply cannot keep up with chip demand, and each chip like the H100s costs around $35,000. Nvidia has a corner on the market, a virtual monopoly, because it was the first to develop the best technology. Like the railroad and oil barons of old who made their wealth because they were the first and the fastest to exploit an industry, Nvidia is the chip baron of the 21st century. So the reason why chip production cannot keep up with demand is because it takes a while to build production facilities, and then there is just a small number of producers that have monopolies in the market, so there are only so many facilities with the capabilities to produce the chips. On top of that, these chips require rare earth metals and other scarce materials in their production. Silicon is an essential building block in all semiconductors. China is the largest producer of raw and refined silicon. High quality quartz is needed to create the crucibles where silicon is melted into a single-crystal structure needed for semiconductors. Most of that quartz is mined in North Carolina in mines that are vulnerable to climate change, like when Hurricane Helene disrupted operations in the two mines that operate there. Gallium and germanium are rare earth metals that are also used in these chips, China has a bunch of both and recently banned exports of those metals to create an AI production bottleneck. Chinese control over these metals means they can drive up global prices, indirectly impacting the US’s ability to create chips using those resources. Ukraine also has some of the largest rare earth deposits in Europe, including gallium, which is why a few months ago Trump was so keen on forging a mineral deal with Ukraine and one of the reasons why Russia wants Ukraine as well.

And while we’re talking about natural resources, let’s talk about the resources that AI needs to exist beyond the precious metals. Or at least what little information we know about this, because there are very few regulations around AI at this point so transparency is notably lacking, meaning it’s impossible to know exactly the resources being used by any AI company or model. Arguably, while chip scarcity is creating a global race to the top, it is energy production that will be the main thing holding back AI advancement. By 2030, AI data centers, the places where your answer is generated when you ask ChatGPT something, they are projected to use 8% of total available power in the US. By 2034, annual global energy consumption by AI data centers is expected to be triple what it is today, to about 1500 terawatt hours, which is about the energy used to power all of India. Data centers require a constant output of energy, so renewable resources like wind and solar aren’t viable solutions until battery technology catches up with needs. So coal, which is still burned to generate a third of electricity supplies, and natural gas which fuels 20% of power and also creates emissions, are necessary to power the needs of AI. One chatGPT search uses the equivalent power of 10 Google searches. And while major companies are turning to renewable energy to help fuel the demand, many others are turning to nuclear energy, including Microsoft, Amazon, and Google. But then the energy needs to be transmitted through overtaxed and old power lines and substations, energy infrastructure that takes years to update. And because we don’t tax businesses or billionaires nearly enough, it will be the 90% who have to foot the bill to update the depleted infrastructure while AI founders make billions.

Water is also necessary to ensure that servers don’t overheat. AI data centers use more than one BILLION litres of water per day. Your conversation with ChatGPT uses about a bottle of water to generate. And then there is the pollution generated by the data centers themselves–noise pollution, light pollution, and air pollution, that are affecting communities where these data centers are being built, usually over the objections of most residents for the benefit of the tax revenue it will generate because the same officials that want the tax revenue are unwilling to tax the rich. It is fucking maddening to me how many of our problems would be solved by just taxing the rich. You’re benefiting from our infrastructure, you’re exploiting the labor of workers that use our roads and infrastructure to get to work, that require our social services to cover the health care and other costs that your abysmal wages don’t cover. You should have to pay taxes for those benefits. But I digress.

The reality is that AI uses tons and TONS of energy to exist, and while it is heartening to hear about some AI developers who are attempting to figure out ways to make AI more energy efficient and sustainable, I’ll link an article in Nature that I read laying out some of those initiatives in the sources for this video, AI is developing at a far faster rate that the sustainable energy alternatives needed to keep it from completely ravaging the planet.

And once the AI tech is created, US officials are concerned with making sure the most advanced tech is kept in US control and out of the hands of China because of the economic value it could generate as well as because of the potential abuse it could allow in the form of advanced high tech weapons manufacturing. Basically AI is not possible without GPUs, Nvidia creates the fastest, best chips on the market, and the more advanced the chip, the faster we can move towards further AI development, and we are in a race to develop AI as fast as possible basically because weapons. Like yes productivity, human development, AI as a tool for furthering mankind blah blah blah but when we talk about the race to develop it, to block other countries from having it, we’re talking in terms of economic development and, ultimately from a national security standpoint, weapons.

How could AI be used militarily by our enemies? So glad you asked. Our enemies would likely learn from our own weaponized tech innovations. America loves to set the standard on violence. And with increasing global conflict especially the genocide in Gaza and the war in Ukraine, there is a LOT of money to be made off war. In June it was revealed that Spotify founder Daniel Ek had recently invested half a billion dollars in an AI defense startup company called Helsing, betting big that AI development in the weapons sector offered a major ROI opportunity. Palmer Luckey is a tech billionaire who founded Anduril, yes that’s a lord of the rings reference, which is a california based defense products company, and he is betting that AI will revolutionize how the federal government buys and uses weapons. And for those of you listening to the podcast who can’t see the picture I’m showing Palmer Luckey is a 32 year old white man with a mullet and a goatee with a penchant for hawaiian tshirts, he’s from Palm Beach California, loves wearing flip flops, and helped develop the Oculus VR headset that was acquired by Facebook for $2 billion dollars. He’s a major Trump donor, a self-described “radical Zionist”, has a net worth of 3 and a half billion dollars and his sister Ginger is married to Matt Gaetz. Matt Gaetz is his brother in law, god I hate it here. Anyway he’s making a line of autonomous weapons that operate using AI that he hopes to sell to the federal government. So that’s great.

A letter last month written by a group of 20 security specialists and sent to US Commerce Secretary Howard Lutnick warned that “Chips optimized for AI inference will not simply power consumer products or factory logistics; they will enable autonomous weapons systems, intelligence surveillance platforms and rapid advances in battlefield decision-making." In a report earlier this month, Wired warned that weapons, AI, and nuclear experts all believe that AI integrating with nuclear weapons is a matter of when, not if. A virtual inevitability. This is on top of the potential AI presents for mass surveillance. Last year, the Bulletin of the Atomic Scientists reported quote “These systems offer cash-strapped autocracies and weak democracies the deterrent power of a police or military patrol without needing to pay for, or manage, a patrol force. … The decoupling of surveillance from costly police forces also means autocracies “may end up looking less violent because they have better technology for chilling unrest before it happens.” And back in March the US State Department reported that they would use AI to scan the social media accounts of immigrants in order to revoke the visas of immigrants who quote “appear to support Hamas or other designated terror groups.” And ICE has entered a $30 million dollar contract with AI firm Palantir, also a lord of the rings reference, founded by certified insane billionaire Peter Thiel, to build ImmigrationOS, a surveillance platform that will allow ICE to more efficiently surveil, catch, and deport people on US soil. These AI developments in surveillance, whether used by our military abroad or, more likely and concerningly, used on US soil, will have a detrimental impact on democracy. According to the bulletin of the atomic scientists quote “A recent study in The Quarterly Journal of Economics suggests that fewer people protest when public safety agencies acquire AI surveillance software to complement their cameras. The mere existence of such systems, it seems, suppresses unrest.” And China is a key exporter of surveillance technology across the globe, especially given its closer ties to autocratic regimes. So advanced tech that Trump has now agreed to allow to flow into China not only could help develop weapons used against the US, it will also allow other autocratic regimes greater access to tech they can turn against their own citizens. That’s not to say that the US doesn’t export surveillance AI to autocratic nations, just that China does so more than we do. For now, I guess. And the bulletin of atomic scientists recommends certain protections for the use of AI as surveillance that indicate Trump is taking this country in exactly the wrong direction. It recommends, quote “Democracies should establish ethical frameworks, mandate transparency, limit how mass surveillance data is used, enshrine privacy protections, and impose clear redlines on government use of AI for social control. Export controls and investment screenings, which would scrutinize and potentially restrict investments in entities or countries engaged in rights abuses, could cut off rights-violating regimes. … Crucially, policymakers should factor in societal impact when setting international standards on artificial intelligence technology—much like accounting for the negative externalities of unethically sourced and polluting products. Otherwise, the loss of civil liberties won’t inform discussions on regulating AI exports.”

And there are opportunities for AI to be developed that would actually further and uphold democratic values like privacy and human rights and protect against surveillance and control. Tools like anonymous browsing or communication enablers, facial recognition blockers, and anti-tracking software. Global efforts could also prioritize developing AI technologies that uphold democratic values like privacy and human rights, rather than enabling oppressive surveillance and control. Many leaders across the globe are calling for AI development to be rooted in democratic values like equality under the law, public accountability, and advancing the greater good. There is a world where that would be possible. Trump, it seems, is intent on leading us in the opposite direction.

If you’re looking to dive deeper, one book I’m really looking forward to reading is the New York Times bestseller Empire of AI by investigative reporter Karen Hao which takes you into the underbelly of OpenAI and how growth, greed, and global influence corrupted the initially altruistic goals of its founders.

If you want to support my work, all these episodes are ad free over on Patreon, patreon dot com slash Leeja Miller.

Thank you to my multi-platinum patrons Marc, Sarah Shelby, Art, David, R_H, L’etranger (Lukus), Thomas Johnson, and Tay. Your generosity makes this channel what it is, so thank you!

And if you liked this episode, you’ll like the one from Wednesday about why you should care about the Bureau of Labor Statistics.

Leeja Miller
Next

Trump’s Attack On Truth At The BLS