Silicon Valley's Newest Export

The Global AI Tax

For every dollar Japan earns from tourists marveling at its temples, it seems to be sending another one straight to Silicon Valley. In the first half of 2025, the country posted a record-breaking ¥3.61 trillion ($24.4 billion) travel surplus. But this impressive gain was almost entirely wiped out by a massive and growing "digital deficit" of ¥3.48 trillion yen ($23.6 billion), a direct result of its reliance on foreign tech giants for everything from cloud computing to social media ads.

This widening gap highlights a new and powerful economic force. As nations and corporations race to adopt artificial intelligence, they are becoming increasingly dependent on a handful of US companies that control the foundational layers of the technology. This dependency comes with a hefty price tag, creating what is effectively a global "AI Tax"—a recurring fee that must be paid for the privilege of participating in the 21st-century economy. Here's Nikkei Asia on just how severe the problem could become:

In April, Japan's Ministry of Economy, Trade and Industry (METI) released an estimate based on its model that predicts a digital deficit of 18 trillion yen in 2035. That is 2.6 times greater than the 6.8 trillion-yen deficit logged in 2024.

In a pessimistic scenario, in which foreign companies increase their share of the Japanese market, the digital deficit could balloon to about 28 trillion yen, more than the 25 trillion yen Japan spent on mineral fuel imports in 2024.

One way to think about it is that Japan could soon be spending more to import digital services than it currently spends to import all of its oil and gas. This "AI tax" is levied at multiple layers of the technology stack:

1. The Compute Tax: Running AI models requires immense processing power, which is overwhelmingly concentrated in the cloud platforms of Amazon, Microsoft, and Google. Japanese companies wanting to use AI must rent this capacity, sending billions of dollars directly to US hyperscalers.
2. The Model Tax: Even with compute, companies often need to license a foundational model from a US lab like OpenAI, Anthropic, or Google.
3. The Application Tax: As enterprise software from companies like Salesforce and Microsoft embeds AI into its core products, the cost of these essential business tools rises, adding another layer to the deficit.

Japan has already tried to use this imbalance as a political bargaining chip. In tariff negotiations with the Trump administration, officials pointed to the massive US surplus in digital trade as a counterpoint to American complaints about Japan's surplus in physical goods like cars. The effort failed. According to a Japanese government official, the reaction from Washington was "muted." This reveals a core tenet of modern geopolitics: while the trade of physical goods is subject to negotiation, the US appears to view its dominance in digital services and AI as a strategic, non-negotiable advantage.

The risk for Japan and other nations that are primarily consumers of AI is that they become permanent digital colonies. They would be locked into paying a perpetual "AI tax" to access a future they had no hand in building, with their economic fortunes increasingly tied to the pricing and access decisions made in Silicon Valley boardrooms.

Tesla's Dojo Is Dead

For a company defined by its relentless pursuit of vertical integration—from car seats to charging networks—killing off a core technology project is a notable event. This week, Tesla did just that, shuttering its ambitious, in-house supercomputer project, known as Dojo. The executive leading the effort, Pete Bannon, a veteran of Apple’s legendary chip team, is also leaving the company.

The move represents a stunning strategic retreat and raises serious questions about the company's ability to deliver on its promise of becoming a vertically integrated AI and robotics powerhouse.

The shutdown is the first major casualty in the brutal and increasingly expensive AI hardware arms race. For years, the conventional wisdom in Silicon Valley was that any company serious about AI needed to build its own custom chips and supercomputers to gain an edge. Tesla's Dojo was the centerpiece of this vision, a bespoke machine designed to process the unique firehose of video data from its millions of vehicles. Now, its apparent demise suggests a new reality is setting in: the cost and complexity of competing directly with hardware giants like Nvidia may be a battle that even Elon Musk can't win.

Dojo's purpose was to process the unique firehose of video data from millions of Tesla vehicles, a task that would theoretically allow it to perfect the AI for its robotaxi ambitions. The scale of the project was immense, as reported by CNBC:

On Tesla’s earnings call in July, Musk said the company expected its newest version of Dojo to be “operating at scale sometime next year, with scale being somewhere around 100,000 H-100 equivalents,” referring to a supercomputer built using Nvidia’s state of the art chips.

The shutdown of such a critical project highlights the brutal economics of the AI hardware war. Building a world-class supercomputer is not just about designing a chip; it's about mastering an entire ecosystem of software, networking, and power infrastructure that established players have spent decades perfecting. The "build vs. buy" calculation for AI compute has shifted dramatically.

Tesla's retreat is a powerful signal of a consolidating market. The early days of the AI boom saw many companies pour resources into developing their own custom AI chips. But the idea that every major tech company will have its own completely bespoke supercomputer is giving way to a more pragmatic reality where they are all, ultimately, customers of a few dominant platforms.

Without Dojo, Tesla loses a key part of its vertical integration story. Instead of processing its vehicle data on its own custom-built, cost-optimized machines, it will likely have to spend billions of dollars to buy or rent GPU clusters from Nvidia or other providers. This makes Tesla a customer, just like everyone else. The dream of a self-sufficient AI powerhouse has, for now, collided with the harsh reality of the market.

The Scoreboard

• Venture: SoftBank’s Vision Fund posts best performance in 4 years (CNBC)
• Semiconductor: Intel CEO responds to ‘misinformation’ and Trump threat in letter to employees (CNBC)
• Semiconductor: Chinese state media says Nvidia H20 chips not safe for China (Reuters)