Deep Dive: China’s Industrial Might and Domestic Policy

About 6 months ago, after an eye-opening dinner with two CEOs building an American supply chain for EV battery materials and rare earths, I posted on X:

“The difficulty in beating China won’t come down to tariffs but mercantilism. The Chinese have 1-2 ‘National Champions’ for critical parts of the global economy.”

With further research, I want to expand on what China’s industrial playbook has been. Beijing doesn’t simply pick “National Champions” and back them with unlimited capital. That would be highly inefficient - the Soviets tried it and, broadly speaking, failed.

Part of what separates modern-day China’s state-backed model from the Soviets is decentralization.

Beijing designates an industry a “national priority”. It then instructs its 31 provincial governments to nurture that industry by backing over 300 prefecture-level governments. These 300+ prefectural governments compete against each other in what you could think of as a national tournament.

An illustrative but imperfect analogy could be to imagine the prefectural governments as state-guided VCs. They deploy a lot of capital and most of the companies they back fail, but some winners emerge too.

This tournament-style competition creates some national champions (like BYD), but it also generates problematic overcapacity.

What I Read This Week…

Berkshire Hathaway has been a net seller of stocks for ten consecutive quarters, trimming or exiting positions, including Apple, Bank of America, and Chevron, while fully exiting holdings like HP and Paramount Global.

At the 2025 annual shareholder meeting, Buffett explained that the firm is waiting for “extraordinarily attractive” opportunities, preferring to hold cash rather than overpay in an overheated market.

As a result, Berkshire’s equity portfolio has become more concentrated, with Apple (still its largest holding), Coca-Cola, American Express, and Occidental Petroleum anchoring Warren’s long-term convictions.

Meanwhile, the company’s cash pile has increased to an unprecedented $348 billion, which is roughly equal to the total reserves held by all U.S. commercial banks at the Federal Reserve.

Warren also announced he will step down as CEO by the end of 2025.

Under his leadership, Berkshire Hathaway delivered a return of 5,502,284% since 1965, compared to just 39,054% for the S&P 500 over the same period. Google’s rollout of AI Mode marks a shift in its core search strategy, transforming Search from a results engine into a task-oriented AI assistant.

This mode offers conversational inputs, including voice and images, and delivers visually rich answers with product cards, local business info, and session memory similar to Search’s current UI.

AI Mode works by combining Gemini with structured real-time data from across its services, including Search index, Maps, Shopping Graph, and local business databases.

When a user enters a query, such as “best foldable camping chair under $100,” AI Mode generates a clear, concise answer formatted in visual cards, which are clickable, tappable, and linked to actions like visiting a store, calling the business, or placing an order.

Users can ask follow-up questions without starting over, and a memory panel on the left-hand side (on desktop) shows previous sessions, letting users revisit or refine their queries over time.

This launch is central to Google’s ecosystem play: by integrating shopping, trip planning, and local discovery directly into the AI layer, Google keeps users inside its platform as they compete with OpenAI and other platforms for more and more tasks that traditionally people searched for.

Deep Dive: A Primer on Stablecoins

Last year, Visa announced the launch of the Visa Tokenized Asset Platform, a new infrastructure designed to help banks issue and manage stablecoins and tokenized deposits.

This week, Stripe began testing a stablecoin pilot, aiming to help companies outside the U.S., and E.U. access U.S. dollars more easily through stablecoin payments.

And just today, Mastercard announced new features that allow consumers to spend stablecoins and enable merchants throughout the world to receive them.

Stablecoins weekly transaction volumes have already exceeded Visa’s, making it the second killer app in the crypto ecosystem, besides Bitcoin. So what’s going on?

This deep dive will explain how stablecoins work, as well as the use cases and implications of more of these tokens flowing through the global economy.

We wanted to communicate the ‘a-ha’ moments which make it clear why players like Visa, Mastercard, and Stripe have allocated substantial resources towards stablecoins, and why one of the current administration’s priorities this year is to establish a clear legal and regulatory framework for them.

What I Read This Week…

The Federal Reserve cut interest rates by 25 basis points on Wednesday, lowering the federal funds rate to 4.25%-4.50%.

Jerome Powell described the decision as a "closer call" and indicated only two rate cuts for 2025, down from four rate cuts projected earlier this year.

Powell emphasized they are entering "a new phase" where they will be "cautious about further cuts" after having reduced rates by a full percentage point since September, reflecting the Fed's growing concerns about persistent inflation.

Public markets responded negatively to the Fed's more cautious approach to future cuts, with the S&P 500 decreasing 3% within hours of the announcement. OpenAI announced their latest reasoning model, o3, demonstrating another step function advance in AI capabilities.

The model sets new performance records across multiple benchmarks, achieving 96.7% on the American Invitational Mathematics Exam, 87.7% on graduate-level science questions, and a Codeforces rating above the 99th percentile.

Using a "private chain of thought" approach, o3 can adjust its reasoning time across three settings (low, medium, high compute), taking longer to respond but providing more reliable answers.

However, o3 still has limitations in handling basic tasks despite its advanced capabilities. The model also requires higher computational costs, at $20 per task in low-compute mode.

o3's approach to problem-solving differs from previous models: instead of retrieving memorized information, it searches through possible solutions and reasons about them step by step, though this process takes more time and computing power.

This addresses a limitation of previous LLMs because it can recombine existing knowledge in new ways to solve novel problems rather than just applying memorized patterns.

Deep Dive: A Primer on Cryptocurrencies

x.com/chamath/status…

Check the date.

In July, President Trump spoke at a Bitcoin conference and told the audience he would make the United States "the crypto capital of the planet". The crowd went wild.

Since then, President Trump won the presidential election with an unprecedented sweep of all seven battleground states, Gary Gensler announced his departure from the SEC next month, and earlier today, Bitcoin passed the $100,000 milestone.

The cryptocurrency market now commands $3.5 trillion in value, with over 15 million daily active users across all decentralized networks. Yet many people are still puzzled by what Bitcoin and other cryptocurrencies actually are, and some people still can't see past all the speculation and scams to understand the real use cases driving this ecosystem.

Moreover, I've read many articles and flipped through many slide decks on crypto, but they all miss something crucial – those key "a-ha" moments that help someone really understand what's going on in this space.

For people still skeptical, I've realized that people don't necessarily need more information – they need the right guide to navigate the crypto rabbit hole.

Our fundamental goal with this deep dive is to deliver those "a-ha" moments in an accessible way. We start from first principles and review crypto's evolution from Bitcoin to smart contract platforms, to stablecoins and "P'nut the Squirrel" memecoins.

If we've done our job right, by the end of the deck, you should have a more visceral understanding of why cryptocurrencies command trillions of dollars in value, what projects in the ecosystem have achieved varying levels of product-market fit, as well as what are some of the takeaways from failed experiments.

For people that have already experienced those "a-ha" moments, I think this deep dive is a great framework to zoom out from the fractal edges and review everything that's happened since Satoshi's white paper was published.

What I Read This Week…

Alibaba recently released QwQ, an open-source AI model that competes with OpenAI's o1 in reasoning capabilities.

What is QwQ and how does its performance compare with OpenAI's o1?

QwQ is an open-source, 32-billion-parameter model with a 32,000-token context window that is strong in mathematical and scientific reasoning.

It outperforms o1-preview on mathematical reasoning benchmarks AIME and MATH, and it beats o1-mini on GPQA for scientific reasoning tasks.

While it performs worse than o1 on LiveCodeBench coding tests, it still surpasses other leading models like GPT-4o and Claude 3.5 Sonnet.

QwQ shows that open-source models continue to rival the capabilities of closed-source models and that Chinese AI models continue to rival the capabilities of U.S. AI models. Ethereum, despite being the second-largest cryptocurrency by market capitalization, faces challenges to its growth model.

The network's heavy reliance on Layer-2 solutions, which are separate networks built on top of Ethereum for faster and cheaper transactions, has created unexpected complications.

While Layer-2 adoption has seen an increase with popular networks like Base and Arbitrum, this success has undermined Ethereum's main network.

Users increasingly prefer the more efficient Layer-2 networks, which have reduced activity and fee generation on Ethereum's base layer.

This shift has two implications.

First, it weakens Ethereum's deflationary economic model, which relies on network activity to reduce token supply, and second, it has fragmented the ecosystem's liquidity and user base, leading to a more convoluted user experience.

This situation has created an opportunity for Solana, a competing cryptocurrency that processes transactions quickly and inexpensively without requiring additional layers.

Solana's native token has demonstrated stronger performance, rising 300% over the past year compared to Ethereum's 75% gain.

As a result, Solana has established itself as the second-largest platform for decentralized finance applications.

What I Read This Week…

Elon and Vivek shared their plans to reduce the size and scope of the U.S. federal bureaucracy.

Their plan focuses on five main areas.

First, they aim to remove regulations that Congress never explicitly authorized, using recent Supreme Court decisions as legal backing.

Second, they plan to cut the number of federal workers through workforce reductions, bypassing civil service protections by using existing "reduction in force" authorities rather than targeting specific employees.

Third, they will stop federal spending that wasn't authorized by Congress, which they estimate exceeds $500 billion per year.

Fourth, they intend to improve cost efficiency in government procurement by conducting large-scale audits of old contracts.

Fifth, they plan to address waste at the Department of Defense, which has a budget of more than $800 billion and has failed its seventh consecutive audit.

They plan to make these changes using presidential powers granted under existing legislation rather than trying to pass new laws through Congress, with a goal to complete this overhaul by July 4, 2026. Bitcoin's price has nearly approached $100,000 multiple times this week.

What's driving this price increase?

First, political support from President Trump and his cabinet nominees has increased confidence in cryptocurrency markets. President Trump's pro-crypto stance has increased support for legislation to establish a strategic U.S. Bitcoin reserve and more favorable regulation of crypto markets, which would increase Bitcoin's mainstream adoption and legitimacy of the entire industry.

Moreover, institutional investors are increasingly buying Bitcoin through Bitcoin ETFs, which saw over $2 billion of net inflow over the last few days and represents a shift from previous years when Bitcoin was mainly bought by individual investors.

Furthermore, market psychology and increased attention on both traditional and social media are creating momentum as Bitcoin breaks through previous price records. With no historical price levels to act as resistance above these levels, and sentiment indicators showing extreme optimism, investors are increasingly targeting $100,000 as the next milestone.

Adding further pressure, the amount of new Bitcoin being created by miners is far lower than current buyer demand. This scarcity means buyers must compete for a limited supply, pushing prices higher. As prices rise, this attracts even more investors hoping to profit from further increases.

Deep Dive: Understanding Federal Agencies

When Elon, RFK Jr., and Vivek talk about an overgrown and ineffective federal bureaucracy, they are referring to the 400+ federal agencies that form the operational backbone of the U.S. federal government.

Why have federal agencies become a lightning rod for criticism?

First, federal agencies are the largest employer in the U.S., employing more civil servants than both Walmart and Amazon. They also issue thousands of regulations annually, which far exceeds the number of laws passed by Congress during its biennial term.

Given that these regulations touch nearly every aspect of American life today and the vast majority of federal employees within the agencies are unelected, there have been calls to reform the federal government's agency system. In order to do that, we need to look at the problem from first principles.

What I Read This Week…

Scientists at Harvard Medical School have developed a versatile, ChatGPT-like AI model called CHIEF capable of performing various diagnostic tasks across multiple forms of cancer.

What makes CHIEF different from current AI approaches to cancer diagnosis?

CHIEF stands out by performing a wide range of tasks across 19 cancer types, maintaining consistent performance regardless of cell sample collection or digitization methods. The model detects cancer cells, predicts tumor origins, forecasts patient survival, and identifies treatment-related genetic patterns, achieving 94% accuracy in cancer detection across 11 of the 19 cancer types.

CHIEF was pretrained on 44 terabytes of high-resolution pathology data and validated using 19,491 images from 32 independent sets across 24 international hospitals and cohorts.

Through more efficient and accurate cancer evaluations, CHIEF could enable clinicians to better diagnose and treat cancer patients. Four college students from the University of Toronto developed an anti-drone technology that outperformed systems from major defense companies like Boeing in a competition hosted by the Canadian military.

How did they do it?

The students built a device that emits high-frequency sound waves to disrupt drones in flight. It works by exploiting materials' resonant frequencies, causing drone components to vibrate and malfunction, similar to how a powerful sound can shatter a wine glass.

To create this device, the team first experimented with car speakers in a living room, testing various sound frequencies on drone parts. They then upgraded to more powerful speakers that could produce ultrasound waves beyond human hearing. Through repeated backyard tests, they refined their system to destabilize drones' navigation systems from 50 meters away, making them unstable, veer off course, or crash.

This final prototype cost the students about $17,000 to develop.

Deep Dive: Is India the Next Economic Giant?

Around 20 years ago, China experienced a period of unprecedented economic growth and technological advancement. I watched many of my friends invest early in China and achieve remarkable success.

Yet, I didn't make a single investment then. Why?

I didn't feel like I understood China. I couldn't speak the language, I didn't grasp the political landscape, and I felt like an outsider looking in.

When I look at India, I feel entirely different about it. India's democratic ideals align well with those of the United States. Many Indians speak English, which removes the language barrier, and there are favorable conditions for American entrepreneurs and investors to create successful businesses.

Around 2016, when India launched UPI (digital payment system), a few of my friends and I made some bets then. Even though UPI was a huge success, in hindsight, 2016 was too early to invest in India and my bets didn’t pay off. Even though we saw a wave of successful startups like Flipkart and Zomato emerge during that time, in many ways, India's startup ecosystem was still nascent. Today, India feels very different from how it was in 2016. Even though some of the same fundamental characteristics remain (e.g., the world's largest working-age population, many talented and educated individuals), the tectonic plates of India's economy have shifted significantly in the past eight years, and many of the long-standing obstacles that were holding back India's economic growth have been addressed.

All this to say, through my own experience of investing in India, I realized that focusing on narrow aspects (e.g., momentum of investment in Indian startups) is necessary but insufficient. It is understanding how the structural pieces of India’s economy fit together that offers the clearest understanding of where India’s economy is today and if it’s the next economic giant.

Many articles have been written about India’s economic potential, but every article I came across, I felt that it was missing this foundational and multi-dimensional approach.

This deep dive builds a foundational understanding of India, from the lens of its economic potential. We will discuss India’s history, governance, demographic profile, technology sector, financial sector, infrastructure, and geopolitical position, to help you understand where India is in its economic development.

We assume the reader has no prior knowledge of India. One of our goals for this deep dive is to come at the subject with the right entry point, so you can develop an enduring perspective that you can build upon over time. If you don’t know very much about India, one sign we’ve done our job is that if, after reading this deep dive, the subject of India should feel less impenetrable (like China was for me), so you have the right footing to read news about India, meaningfully participate in discussions about India, and form a nuanced worldview.

Later, we will do more deep dives on India that branch out and dive into specific topics, with this one as the base.

One thing I want to say is that my team and I spend a lot of time and energy to create an enjoyable reading experience. Our deep dives are designed to be read in one go, like a flip book. If we’ve done our job well, after 20-30 minutes of reading, you should be able to form a clear picture of where India stands in terms of its economic potential.

We will be publishing our deep dive on India's economy next week, covering everything from how India's education system works to how the country's infrastructure development varies across different regions.

Like our previous deep dives, we will build from the ground up, assuming no prior knowledge.

Here’s what we’ll cover:

What is the current state of India's economy? How does it rank globally in terms of GDP, growth rate, and other macroeconomic indicators? What are the key sectors driving India's economy? How is India's education system preparing its youth for the future job market? How has India's IT sector evolved and what is its contribution to the economy? How is the startup ecosystem in India contributing to economic growth and innovation?

What I Read This Week…

The “Magnificent 7” stocks have lost over $1.7 trillion in market capitalization in just two weeks, with Nvidia's stock price decreasing 7% in a single day.

What's going on?

Investors are reassessing the lofty valuations of the largest technology companies, driven by concerns over their large investments in AI and mixed Q2 earnings.

Macroeconomic indicators are pointing to a weakening economy. The latest jobs report revealed new unemployment claims at their highest level since last August, while manufacturing has contracted for the fourth consecutive month.

These signs of economic weakness are prompting investors to anticipate Federal Reserve interest rate cuts starting in September, with the 10-year Treasury yield dipping below 4% for the first time in six months. The United States is engaged in an increasingly complex battle to control the export of advanced AI chips to China, amid fears they could be used to enhance Chinese military capabilities.

Despite stringent export controls implemented since October 2022, a thriving underground market and creative corporate structures are allowing China to acquire them.

For example, in Shenzhen's bustling electronics markets, vendors openly offer Nvidia's most advanced chips, with some deals reportedly involving hundreds or even thousands of units.

One vendor claimed to have facilitated a $103 million shipment of over 2,000 advanced Nvidia chips from Hong Kong to mainland China.

We will be publishing our deep dive on biotech at the end of this month, covering everything from how gene editing works to how AI is accelerating drug discovery and development.

This month’s deep dive on biotech focuses specifically on human therapeutics.

We will do future deep dives on agricultural biotech (e.g. GMO crops) as well as industrial biotech (e.g. biofuels and biodegradable plastics).

Like our previous deep dives, we will build from the ground up, assuming no prior knowledge.

Here’s what we’ll cover:

What is biotech? How do biotech companies turn breakthrough research into new products and treatments? How do small molecules, biologics, gene therapies, and other biotechnologies differ in their approaches and applications? What are the stages of developing a new drug? Why are biotech companies generally higher-risk and more capital-intensive compared to other technology companies? Why are drug prices so high?

What I Read This Week…

The U.S. economy added 206,000 jobs in June 2024, exceeding expectations, but the unemployment rate rose to 4.1%, its highest level since November 2021.

What do these mixed signals mean for the Federal Reserve's interest rate policy?

The Federal Reserve's dual mandate requires it to promote maximum employment and price stability, so some see the unemployment rate increase as a factor that may lead to two potential rate cuts this year, with the first expected in September or November.

However, economists note that while the labor market is showing signs of moderation, it remains relatively resilient overall. The investment world is increasingly focused on geopolitical risks, with many financial firms hiring geopolitical advisors and former government officials to help navigate a more complex global landscape.

What's driving this trend?

Investors perceive a fundamental shift in the global landscape, where geopolitical factors are increasingly driving inflation, amplifying risks, and complicating investment decisions.

This marks a departure from recent decades, when globalization typically led to reduced prices and risk across markets.

Key concerns include US-China tensions, conflicts in Ukraine and the Middle East, and political instability in South American countries like Venezuela, Bolivia, and Ecuador.

What I Read This Week…

Nvidia's market capitalization recently passed the $3 trillion mark, surpassing Microsoft and Apple as the most valuable company in the United States. Nvidia now also accounts for 34.5% of the S&P 500's growth this year.

So what's the problem?

Nvidia’s growth is partly driven by speculation about the future potential of AI, as well as Nvidia’s central role in shaping that future.

However, if these optimistic projections do not materialize as investors anticipate and Nvidia's stock were to decline, it would lead to a significant drop in the S&P 500 as well. Anthropic released Claude 3.5 Sonnet this week, the first release in a forthcoming Claude 3.5 model family.

Notably, Claude 3.5 Sonnet excels in the Graduate-level Proficiency Question Answering (GPQA) assessment, which is designed to assess advanced academic understanding across various disciplines, testing complex reasoning and knowledge application typically associated with graduate-level education.

While the average PhD graduate scores 34% on this test and a specialized in-domain PhD scores around 65%, Claude 3.5 Sonnet has set a record with an impressive 67.2% score.

Deep Dive: Creator Economy - The Next Phase of Media

Every day, we make decisions about which products and services to consume. With countless options and categories to choose from, how do we end up selecting certain products over others?

More than $60T is spent annually by consumers on various products and services so this month’s deep dive focuses on the changing nature of how consumers are influenced.

In a world where we often lack the time or expertise to evaluate every option, the media initially drove consumption decisions by selling advertising about which products and services were best, most valuable, or most suitable.

To understand how media will drive consumption decisions in the future, we look to the past to investigate the underlying economic dynamics that have driven its evolution. Through our investigation, we uncovered four distinct phases of media: We are already seeing evidence of this today, as influencers like MrBeast and Kim Kardashian build massive consumer goods businesses on the back of their captive audiences.

But they aren’t the only ones. We end our deep dive with an analysis of Elon Musk, the world’s first Mega-Influencer, who has built several massive businesses by following the playbook we outline in the deck.

1/ Founders: If you have competing term sheets, what factors are important beyond price? Firm A has a lower valuation BUT a much better track record of realizing that value. Firm B has a history of paying up but fewer exits. Who do you choose? 🧵👇🏾 2/ Startup employees: Similarly, how do you weigh offers from two competing startups? Is there data hiding in plain sight that may help you decide whose equity will be worth more?

New data from @AkiliLabs clinical trial for teens 13-17yo with ADHD show that 2 out of 3 teens had meaningful clinical improvement in attentiveness (using an FDA-cleared test to measure attention deficit) and 1 in 4 teens tested into the normal range. All through a video game

https://twitter.com/AkiliLabs/status/1611036476636073985

These results are 3X better than Akili's clinical trial results for 8-12yos, which was their original FDA authorization

The two most important drivers of the next decade: the marginal cost of energy AND the marginal cost of compute will both go to zero.

What does this mean? Take one of the biggest technology leaps of 2023: ChatGPT. It takes 6,000kWh and $100K per day to run ChatGPT, per @tomgoldsteincs, not including the $1B MSFT spent to fund @OpenAI 

https://twitter.com/tomgoldsteincs/status/1600196981955100694?s=20&t=kXgZEifntqmVjULNbDlTSw

What I read this week... cnbc.com/2022/09/15/fig…

What I read this week... wsj.com/articles/kim-k…