作者：BlockBeats
Original title: The Remarkable AGI Trades of Daniel Gross
Original author: @johncoogan
Compiled by: Peggy, BlockBeats
Editor's Note: In early 2024, AI was still in a phase of both hype and uncertainty. At that time, Daniel Gross posed 18 questions on a single page: Where will value go? Will energy become a bottleneck? Will software engineers be replaced? How will the competitive landscape between nations change?
Looking back two years later, these questions are more insightful than any specific prediction. The benefits of AI are indeed concentrated at the infrastructure level—Nvidia is the biggest winner; energy and electricity are rapidly becoming new strategic bottlenecks; API costs have plummeted, while computing power, capital, and geopolitical risks are constantly amplifying.
This article reviews the key questions Gross raised at the time and examines them one by one in light of the real-world evolution over the past two years. It is not only a retrospective of the investment logic in AI, but also a roadmap for observing how technological revolutions reshape market structures, supply chains, and the global power landscape.
The following is the original text:
In January 2024, Daniel Gross, then CEO of Safe Superintelligence and now head of Meta AI products, published an article titled "AGI Trades".
This one-page article listed a series of questions about the potential impact of AI advancements. Looking back more than two years later, these questions seem remarkably prescient, even though no definitive conclusions were reached at the time. Below, we review each of the 18 questions he posed.
Markets
In the post-AGI world, where will value go?
Currently, value is indeed concentrated at the infrastructure layer—chips, packaging, power, and other areas. Nvidia has captured over 100% of the profits from the AI boom, as many companies are still losing money. This is clearly reflected in market capitalization changes: Nvidia's market capitalization increased by $3.2 trillion, from $1.2 trillion to $4.4 trillion; in contrast, the growth of cloud platforms has been much more modest (Microsoft rose 4%, Amazon rose 30%).
In the private equity market, the valuation growth of OpenAI, Anthropic, and xAI has also been remarkable, but the combined total value growth of $1.4 trillion is still lower than the increase in market capitalization of Nvidia during the same period.
This is a very critical issue right from the start of 2024.
What will happen to Nvidia and Microsoft?
Nvidia's performance has been extremely strong. Its revenue grew from $60.9 billion in fiscal year 2024 to $215.9 billion in fiscal year 2026, nearly tripling.
Microsoft, however, is not so dominant. While Azure's growth has indeed accelerated to a 40% year-over-year increase, Microsoft's stock price will only rise by 4% from January 2024 to March 2026. The market has raised questions about its annual AI capital expenditure of over $80 billion—it remains unclear when this investment will translate into returns.
In this AI gold rush of "selling shovels and spades," Nvidia is clearly the biggest winner, while Microsoft's bets on infrastructure have not yet brought obvious returns to shareholders.
Is copper being mispriced?
It was indeed severely undervalued. In January 2024, copper was priced at $3.75 per pound, and two years later it reached an all-time high of $6.61 per pound.
AI has an enormous demand for copper. For example, the NVIDIA GB200 NVL72 server rack uses more than 5,000 copper wires. If all of them were straightened, the total length would exceed 2 miles, and a 100MW data center would require approximately 3,000 tons of copper.
Overall, data centers may consume 500,000 tons of copper annually. Some have therefore called copper "the new oil." Of course, many other things are also referred to as "the new oil," because AI infrastructure construction is extremely complex, with bottlenecks at almost every stage. Therefore, this claim should be viewed with caution.
Real Estate
If AI can write all software, will San Francisco become the new Detroit?
It depends on what "the new Detroit" refers to.
AI actually saved San Francisco, preventing it from becoming a city that declined like Detroit. San Francisco is still thriving today.
The office vacancy rate decreased from 36.9% to 33.5%.
OpenAI has 1 million square feet of office space.
Anthropic owns a 25-story office building.
Sierra signs 300,000 square feet of office space.
In the first half of 2025, 78% of US AI venture capital flowed to the Bay Area. Of course, there's another side to the story: San Francisco's overall employment rate remains below pre-pandemic levels, but housing prices remain strong. Therefore, it's far from being a "ghost town." The urban environment has also become cleaner.
How will AI affect wealth inequality?
It is too early to draw conclusions, as the data changes are not yet significant, but some studies are worth noting.
An IMF 2025 study suggests that AI may reduce wage inequality (due to the automation of high-paying jobs), but could exacerbate wealth inequality (with capital gains concentrated in the hands of tech company owners). OECD research found that wages for low-skilled jobs grew the fastest (assemblers +11.6%), while those for high-skilled jobs grew the slowest (CEOs +2.7%). However, this likely reflects minimum wage policies more than AI itself.
In the capital markets, concentration is also rising: the "Mag7" accounts for about 32% of the S&P 500's market capitalization and contributes about 42% of the total returns in 2025; at the same time, the huge funding rounds for AI startups (OpenAI $110 billion, Anthropic $30 billion) have also created enormous private wealth for a few founders and investors.
Energy & Data Centers
If AI becomes an energy competition, how should we invest?
This assessment is absolutely correct. AI has indeed become an energy game.
Those who capitalized on this trade made a fortune. For example:
• Vistra: +321%, the second-largest gain in the S&P 500 in 2024 (second only to Palantir)
Constellation Energy: Its stock price has tripled since the release of ChatGPT.
• NRG Energy: Expected to increase by approximately 95% in 2025.
Oklo: Up 700%+ in 12 months
Nuclear energy experienced a boom:
Microsoft signs $16 billion, 20-year PPA to restart Three Mile Island nuclear power plant.
Google and Kairos Power sign 500MW Small Modular Nuclear Reactor (SMR) agreement
Meta signed 6.6GW of power contracts with several nuclear energy companies.
Energy has become one of the most successful investment themes in the AI era.
In the entire data center supply chain, which links are the most difficult to scale up 10 times?
The bottleneck in the chip industry is CoWoS packaging technology (TSMC's Chip-on-Wafer-on-Substrate).
In the data center field, the biggest bottleneck may be power transformers.
Delivery cycle is close to 3 years.
A 30% supply gap is projected by 2025.
Costs have increased by 150% since 2020.
This 100-year-old technology has become a key limitation on the speed at which data centers can connect to the power grid.
Is coal undervalued?
To some extent, yes, but far less than copper. Coal prices actually fell by about 22% in 2025, before recovering somewhat by early 2026.
The coal company's performance was acceptable:
Peabody Energy: +34%
CONSOL Energy: +37%
Meanwhile, U.S. coal-fired power generation is projected to increase by 13% by September 2025.
This is particularly evident in states with rapid data center growth:
Ohio: +23%
Oklahoma: +58%
Nations
Who is the winner and who is the loser?
The winner is clearly the United States.
In 2024, private AI investment in the United States reached $109 billion (compared to only $9.3 billion in China), bringing the cumulative investment since 2013 to $470 billion, exceeding the total investment of all other countries combined. In 2024, the United States released 40 significant AI models, while China released 15.
The game isn't over yet, but for now, the United States is at the center of the AI competition.
What will happen if India's $250 billion GDP exports rely on the GPT-4 token?
The situation is beginning to emerge, but it is still in its early stages. Hiring in India's IT outsourcing industry has declined significantly. Between 2024 and 2025, large IT companies are expected to lay off approximately 58,000 employees, compared to 360,000 new hires between 2021 and 2023.
Will software engineers be replaced like typists in history?
Software engineers are not currently doing blue-collar jobs, but the professional structure is already showing signs of differentiation:
Demand for AI engineers increased by 143%.
• Large technology companies saw a 25% decrease in hiring for entry-level positions.
• Internship positions reduced by 30%
The options for the future may be either to upgrade to "managers of AI agents" or to move into fields such as manufacturing—after all, many factories also need people who understand software to automate production processes.
Will there be a large-scale employment plan similar to the "new policy"?
Not yet.
In July 2025, the Trump administration launched the "American AI Action Plan," which includes:
AI Education Executive Order
Skills training program
• $84 million in apprenticeship grants from the Department of Labor
However, US spending on workforce training accounts for only 0.1% of GDP, almost the lowest among OECD countries. There are currently no plans to reach the scale of the WPA (8.5 million jobs plan).
Is lifelong learning a worthwhile investment?
This is a very abstract and very personal question. But my answer is: it's worth it.
Inflation
If AI is indeed deflationary, how would we first see this signal?
The best metric is probably the price of AI APIs.
GPT-4 level inference cost:
End of 2022: $20 per million tokens
December 2025: $0.40
A 50-fold decrease in three years. This rate even surpasses the decline in PC computing power costs or internet bandwidth costs. This could very well become a leading indicator of service price deflation.
If the demand for knowledge products continues to grow while production costs decrease, how should we understand deflation?
While AI API prices have plummeted, AI company revenues are soaring. Lower prices → explosive growth in usage → increased total spending. Meanwhile, SaaS companies are also charging an "AI tax" of 20%–37% on renewals. Therefore, even with software production costs approaching zero, SaaS revenue is still growing.
This is similar to the computing industry during the Moore's Law era: individual products became cheaper and cheaper, but the overall market size continued to expand.
Geopolitics
Is interconnection really important?
Extremely important.
In large GPU clusters, 30%–50% of training time is spent on communication between GPUs, rather than computation.
For example, Google TPUv7 Ironwood uses a 3D torus topology to connect 9,216 chips and Nvidia NVL72 to connect 72 GPUs, so interconnected networks are crucial for scaling AI.
If a country has more energy, can it achieve AGI using outdated processes?
It seems unlikely at the moment.
All leading AI chips use 4nm or 3nm processes, including Nvidia Blackwell, Google TPUv7, and AWS Trainium3.
Huawei's Ascend 910C (SMIC 7nm) is competitive in inference, but requires more chips and more energy for training. Simply increasing energy consumption to bridge the technological gap will ultimately be limited by economic costs.
What is the most likely "Taiwan incident"?
The most likely scenario is a blockade of the Taiwan Strait.
Tensions are already escalating:
• 2024: China holds the "Joint Sword-2024B" exercise
• 2025: "Mission for Justice 2025" will utilize over 100 aircraft and 13 warships.
· 27 rockets were launched from Fujian, 10 of which landed in Taiwan's contiguous zone.
Meanwhile, in its 2026–2030 Five-Year Plan, China began to separate the terms “peaceful reunification” and “reunification”.
TSMC is also making early preparations: eight wafer fabs are under construction in Arizona, which may account for 30% of the advanced chip production capacity in the future.
However, the entire system remains in an extremely fragile balance.
No Comments