Earlier this week I spoke with VOA China about export controls (Google Translate Link) being implemented to prevent the sale of specific types of semiconductor / chips / finished hardware products between the United States and China.
I think it’s important for policy makers and the public to understand the importance of these chips, and the complexities of these discussions. In this post, I’ve attempted to frame a few core concepts, provide a little recent history, and included a few suggestions for ways to create a more proactive national defense strategy around chip production, so that we can be better prepared for any future supply chain disruptions.
Advanced AI chips are the equivalent of a specialized brain, which can be networked into something much more powerful than its individual components. When these chips are integrated into a larger computer or network these are oftentimes then referred to as “AI accelerator chips” – and these chips are programmed to solve specific problems.
For all intents and purposes, most sophisticated policy makers are only focused on restricting specific types of advanced chips and finished products being sold between United States corporations and Chinese corporations – the specific products that serve this broad machine learning purpose and have been apparently used by the CCP for predictive policing and a wide variety of automated censorship tasks.
And unfortunately for policy makers, many of the chips that are used for IOT/mobile usecases, the more advanced versions of those same chips can be used as AI accelerator chips.
There are a wide variety of consumer semiconductors and chips so it’s important to not broadly categorize all chips as creating the same types of risks when sold – options range from ARM-designed chips for mobile and IOT devices, to more classic intel x86 chips for desktop computers and servers. The new open source RiskV architecture is likely to play a big role in the future of chip designs. Some of these consumer chip designs are shared safely as a course of the technical supply chains between the U.S. and China, and many of the “legacy chips” are not part of discussions for bans.
It’s important for China and the United States to not create blank technical policies which restrict the production of consumer products, while also crucially important to acknowledge that some high tech manufacturing products are too sensitive to be traded between the countries. We must acknowledge that all countries relying on chips and semiconductor manufacturing largely rely on products built in Taipei, Taiwan, while also admitting that raw materials are being mined elsewhere, and the consumers purchasing those products are also typically located across the world – we have a global hardware supply chain, that if-disrupted, would lead to worldwide recessions and a rolling back of technical progress on a wide range of important fronts.
For the last few years, dozens of Chinese companies have been added to technology ban lists in the U.S., rug pulling US companies who have their revenue and clients removed without clarity about new customers available to purchase those same products domestically or outside China. The bans against these Chinese companies were not accidental, so we should assume that the U.S. Congress or other entities pushing the bans had a seemingly ‘good reason’ to believe that specific hardware products were being purchased by businesses in China but not being used for their intended purposes, and being diverted into the CCP’s military-business fusion programs. There’s seemingly nothing that can be done about national security investigations that lead to “surprise bans” on specific hardware purchase arrangements – and we should expect more of this to happen in the future, not less.
One of the biggest challenges with regulating AI accelerator chip use cases is that the current regulation thresholds have seemingly been around “how fast” the chips are – how many concurrent tasks they can conduct and the speed of those calculations. And the proprietary nature of the semiconductor manufacturing process is filled with R&D teams testing new methods. So due to this, can a company expect to be able to wall off innovations within a company so that “slower/less modern tech” could be sold to specific customers, whereas “faster/more modern tech” both couldn’t be sold to specific customers, but also specific employees at that company couldn’t even use or know about that technology? How would you literally break up the company to do this?
What happens when someone working on the “slower chips” suddenly has a breakthrough that allows them to compete with the “faster chips” being produced by that same company? What kind of ridiculous idea-monitoring architecture would be needed to properly firewall a business that operated across jurisdictions where some of their own employees would need to openly work on less sophisticated technology than some of their other counterparts? Do you just deliberately hire non rockstar developers to manage this failing downwards consumer development process?
Arm Ltd + Arm China make everything more difficult
These hypothetical scenarios are all too real, due to the extremely complex arrangement between Arm Ltd and Arm China driving the vast majority of the complex politics around chips and semiconductor manufacturing.
In 2020, Softbank claimed that Arm China’s intellectual property powered 95% of the sensitive computer chips being designed in China – yet Arm China was sold for only $775 million by SoftBank in 2018 to their Chinese Joint Venture, a price so low that it’s been surprising more investor lawsuits didn’t crop up around this undervalued sale. (https://www.reuters.com/article/us-softbank-group-china-jointventure/softbanks-arm-cedes-control-of-china-ops-to-consortium-for-775-million-idUSKCN1J10RM)
Furthermore, Arm China accounted for 1/5th of the Arm Ltd revenue in 2018, and Softbank paid $32 billion for Arm Ltd in 2016 (https://www.reuters.com/article/us-softbank-group-china-jointventure/softbanks-arm-cedes-control-of-china-ops-to-consortium-for-775-million-idUSKCN1J10RM) – so at a minimum – if Arm retained it’s value from 2016, the 2018 sale of Arm China should have been worth at least $6.4 billion based on the original 2016 purchase price. Yet, SoftBank sold the assets for merely $775 million, a somewhat shockingly low number.
The challenging history of Arm China extended to a multi-year battle just recently resolved between Softbank, local Chinese authorities, and the Arm China CEO who largely “went rogue” by creating investment conflicts of interest and making it nearly impossible to audit the financials during this period. (https://www.reuters.com/technology/arms-chinese-joint-venture-replaces-ceo-2022-04-29/) This entire episode with Arm China and Softbank broke the trust between Chinese business interests and non-Chinese chip and semiconductor manufacturers, and CCP authorities should be blamed for allowing a rogue CEO to undermine these crucial relationships for multiple years.
Now, Arm China is looked at with extreme skepticism, and the entire Western world has been forced to create new policies to prevent semiconductor intellectual property from leaking through the next CCP-ARM China joint venture.
Another way to explain the current CCP conundrum – the CCP always intended to become self-sufficient in chip manufacturing through investments in new chip design protocols like RiskV, which is open source and has been receiving massive investments in China, all while China was still largely using Arm designs for the vast majority of their domestic chip production. But the timing of these shifts needed to be carefully considered, and CCP officials associated with Arm China overplayed their hands, and now they are years away from RiskV being scaled to production and performance, while also now losing access to more and more sophisticated chips due to Defense embargoes in the West.
It’s 2022 and the U.S. only weeks ago passed a significant law to invest in domestic semiconductor manufacturing – we’re decades behind China and the rest of the world in building production supply chains that reach from raw materials to nanomanufacturing – and our reliance on China for crucial aspects of this supply chain should be an embarrassment to everyone in the U.S. national defense establishment.
We also must acknowledge that all semiconductors are unique, and maybe in 3 to 5 years RiskV open source architecture will power more consumer devices, enterprise servers, and also AI accelerator use cases – but right now, that’s not the case. The vast majority of sophisticated AI accelerator microprocessors are not being created with open source RiskV architecture, and the current non-riskV proprietary AI accelerator chips need to be treated like the limited natural resource that they are – production limits which rely on global partners are squeezing supply chains of the most sophisticated chips, and the alternative open source AI accelerator chip market won’t be able to meet proprietary markets demands – so current advanced chips are a limited resource, literally created by combining expensive raw materials with complex manufacturing processes only possible in a few specific locations worldwide. So U.S. policy makers could either acknowledge this reality and create export policies which adequately acknowledge the scarcity and national security priority to maintain a national reserve of proprietary AI accelerator chips, or we could allow businesses to sell as many as they wanted to Chinese interests, even though apparently it’s been documented that many of these sales were diverted for military purposes.
For the immediate future, Arm chip/processor designs will be the foundation for intellectual property both inside and outside China for entire hardware verticals. But it’s important to recognize that Arm does not make chips itself – and Arm China also doesn’t make chips itself – the entire global Arm operation is a highly sophisticated licensing operation grounded in the intellectual property of Arm and their expertise in chip design. For instance, Apple/Google and numerous big tech companies uses Arm for the design of their proprietary chips – and for instance the Apple-Arm chips are being designed in Taiwan, but no other company will have Arm chips just like the Apple-Arm chips. These types of customized Arm designs are the foundation for proprietary hardware created by companies like Apple.
For Arm China, there are dozens of core partners licensing their tech within China, and a murky flow of money through investment funds and startups (heavily documented in the media) – this creates potential challenges for any future Arm IPO and public reporting requirements, but the larger threat to Arm LTD revenue from Arm China will not be through financial shenanigans, but it will be through any eventual mandate by the CCP to switch to open source RiskV architecture, or any proprietary CCP-RiskV architecture that is designed in the future for specific AI accelerator use cases.
Sometime in the next 3-10 years, Arm (and Intel designed-chips for that matter) will start to see real competition from the open source RiskV architecture, and China is investing more in RiskV than any other country, specifically in the RiskV AI accelerator use cases.
It should be assumed that China does not want to rely on the Softbank-owned Arm designs (RISC) indefinitely, beyond the required time horizons needed before an adequate replacement can be created independently of Western intellectual property, and eventually there will be countrywide shifts in China to RiskV chip designs — this is their only current path to independent chip manufacturing.
And let’s be clear: 100% of the Arm China technology is in the hands of the CCP and those designs are in the hands of any and all engineers that the CCP has working on priority efforts around RiskV. If we can’t all acknowledge this is how the national security of China works, and how Arm China was setup to operate, and why the CCP would both demand this code/architecture for security investigations and also for product development, then it will be hard to plan for the future.
There is some potential future time horizon where Arm China is not providing revenue to Arm Ltd due to some future CCP demand to use CCP-RiskV chip designs, which are coming in the next decade and would empower China to cut off external payments for chip designs while ideally creating a new cutting edge path to technology competition with existing Arm designs for mobile/IOT/advanced AI use cases.
For the near future, Arm Ltd should restrict Arm China’s access to *new* designs while not restricting access to many legacy designs, or potentially creating a timeline that leads to no new Arm designs for Arm China and forces the adoption of CCP-RiskV architecture in some amount of years.
The entire chip market will be significantly stronger if CCP-RiskV eventually becomes the dominant mobile chip design out of China over Arm, so that both China and the West can coordinate and collaborate on their own versions of the open source RiskV architecture, while China and the CCP-aligned chip makers can build whatever proprietary versions of RiskV (CCP-RiskV) end up meeting their own intellectual property and lack-of-privacy needs. The West would also create their own proprietary chips, yet still be able to collaborate through the opensource RiskV designs. Collaboration and competition, without muddying up why a firewall needs to exist in this design process.
Where was the CCP leadership when ARM China was operating like legitimate semiconductor pirates, with a CEO who had been fired retaining control for nearly 2 years, while hiring his own security, R&D department and working on ARM-China-only intellectual property?
How well did that process work out for everyone? Shouldn’t the West be concerned that it took the CCP years to take action against a rogue CEO who took over the most important semiconductor joint venture between China and the West? Has the CCP done anything to indicate that they don’t fully have control of the Arm China entity?
What happens if the U.S. is unable to access our current monthly demand for advanced semiconductors?
I’m calling on Congress to pass a law that would create a new “Strategic AI Accelerator Chip Reserve” with a goal to sell/refresh the reserve every ~12-24 months and encourage all innovative states to develop their own physical semiconductor reserves similar to current “Strategic Petroleum Reserves” – but with an eye towards ensuring this hardware ends up in consumer products developed and manufactured domestically, and in the hands of domestic researchers.
We need to develop a “Strategic AI Accelerator Chip Reserve” also to ensure that in the event of future conflicts, the U.S. has significant backup resources of advanced chips for military and civilian research purposes, and we can’t keep pretending like the biggest threat to our national security is the scarcity of PlayStation 5 consoles, or the other consumer shortages that are always better covered in the news than the advanced AI chip shortages.
And let’s be clear, just because the CCP is using some chips for predictive policing across China, no one is suggesting that the U.S. government keeps these chips for ourselves “so we can do more predictive policing” – hopefully our government starts to develop a much more sophisticated appreciation for both the hardware needed by authoritative states, but also guardrails that can be put on private businesses to minimize harm and prevent the creation of algorithmic blackholes without public feedback loops.
The U.S. government also can’t keep on haphazardly restricting commercial sales (https://www.nytimes.com/2022/08/31/technology/gpu-chips-china-russia.html) of US semiconductor companies without also finding a replacement for that revenue – it’s tantamount to shooting ourselves in the foot so we don’t kick our own ass.
Some of the most sophisticated semiconductor chips used for AI use cases are developed in Taiwan, and for far too long the United States ignored that we could both support Taiwan politically and economically while also supporting the growth of our own domestic semiconductor manufacturing. We can buy their chips and make them too.
We don’t need an annual trade war with semiconductors – we need the U.S. government to appreciate that innovators and US businesses struggle with getting access to sophisticated AI accelerator chips, and the US government can support purchasing and donation programs so that more people are able to use these crucial technologies in real products and research programs.
The U.S. government had historically attempted to thread an imaginary needle where they allow U.S companies to conduct some narrowly defined sales of sensitive high-end chips to China and Russia, while claiming that those sales were protected by agreements not to share it with the military-civil research schemes. We now know these agreements failed to predict reality, and those chips sold under these schemes were used by oppressive militaries for predictive policing schemes and other harmful activities.
If China wants to increase cooperation with the United States on our current semiconductor production, we must get commitments that the CCP’s military-civil research schemes developing open source RiskV architecture also open source their production methods. Otherwise, if the CCP continues to bluster about U.S. defensive efforts around our own sensitive proprietary semiconductor supply chains, while the CCP pushes their typical public-private hegemonism, it will only become more clear that the CCP is lying and playing the victim on behalf of their military-commercial industrial complex.
The U.S. government needs to both fund new chip supply chains, but also fund a strategic reserve of these advanced chips. Then we should be creating programs to empower universities and entrepreneurs to take advantage of this influx of high-tech domestic chips. We can not keep treating the most sophisticated hardware on the planet like a renewable resource, and the US must also ensure that we’re not putting up new walls of innovation while trying to protect our own national defense interests.
The vast majority of companies building sophisticated high-end semiconductors have historically been manufacturing products in Taiwan, all while the West was largely ignoring the tightening grip of power by the CCP on all aspects of Taiwanese life, business and politics. Ongoing debates about the history of Taiwan change nothing about the current realities of the semiconductor supply chain, and the literal products the world relies upon which are mostly being produced there. To reduce the pressure cooker of Taiwan, we must produce semiconductors in the United States, while ensuring that everyone in the world can continue scientific advancements.
Unfortunately for Chinese businesses, the CCP does not play fair on the global stage with non-Chinese businesses, so those same Chinese businesses can’t be trusted members of sensitive hardware supply chains – period.
As domestic US production of semiconductors and chips used for defense and critical manufacturing industries ramps up, should the federal government or specific states with high tech manufacturing interests consider investing in a “strategic semiconductor & chip reserve” which is purposefully acquired in bulk to ensure 24-36 month reserves of this hardware for cutting edge industries?