AI Model War Is Over. Now What?

The Next AI S-Curve

The funny thing about a technological arms race is that eventually, everyone’s weapons start to look the same. According to David Luan, the head of Amazon’s AGI research lab, that moment has arrived for the large language models at the heart of the AI revolution. Luan, a key architect behind GPT-3, has a provocative theory that the race for the "best" chatbot is effectively over, with all major models rapidly converging on a similar level of capability.

This idea, that the core technology of LLMs is becoming a commodity, signals a fundamental shift in the AI battleground. If having the smartest model is no longer a durable advantage, the new fight is moving from passive chatbots that answer questions to proactive "agents" that can accomplish tasks. In an interview with The Verge's Decoder podcast, Luan laid out this new reality:

All LLMs will converge to the same model of the world. I think that’s actually happening in practice from seeing frontier labs deliver these models... The labs have all figured out how to reliably tape out increasingly better models.

The next frontier is AI "agents"—systems that don't just talk, but do. The problem, of course, is that today's agents are notoriously unreliable. A chatbot that hallucinates is an amusing quirk; an agent that hallucinates while interacting with your bank account is a catastrophe. How to solve this reliability problem is now the central, multi-trillion-dollar question in Silicon Valley, and the major labs are pursuing starkly different answers.

One way to think about it is that there are now competing philosophies for building trustworthy agents:

1. The Amazon Approach: The Flight Simulator. Luan argues that today's models fail because they learn by "behavioral cloning" (mimicking text) rather than understanding cause and effect. Amazon’s strategy is to build massive "gyms"—simulated software environments where agents practice real-world tasks like accounting or travel booking. The bet is that by learning from trial and error, not just text, agents will become rock-solid reliable. Amazon's unique advantage here is its vast internal operations—from AWS to its retail logistics—which provide an unparalleled private dataset of real-world workflows to build these training gyms.

2. The OpenAI Approach: The Bigger Brain. OpenAI’s strategy seems to be that raw intelligence is the ultimate solution. The frontier AI lab is focused on building a "factory that repeatedly churns out increasingly better models," like the recently released GPT-5. The underlying philosophy is that a sufficiently powerful reasoning model will eventually be able to plan, self-correct, and execute tasks reliably simply by being smarter. Instead of practicing in a simulator, the model learns to think its way out of trouble.

3. The Microsoft Approach: The Ecosystem. The tech giant is betting on distribution and integration. Microsoft's strategy is to embed agents deeply into the products that hundreds of millions of people and businesses already use—Windows, Office, Azure and Bing. By providing "Agents Toolkit," the company is encouraging a vast ecosystem of third-party developers to build on top of their platforms.

The whole thing is a fascinating corporate chess match. The AI race has evolved from a sprint to build the smartest chatbot into a marathon to invent the most reliable worker. The core question is no longer just what these models know, but how they should learn to act on that knowledge. The answer will likely define the next decade of technology.

Japan's Chip Déjà Vu

One way to respond to an existential competitive threat is to form a powerful alliance with your domestic rival. It is a logical, if sometimes uncomfortable, move for survival. And yet, in Japan's power semiconductor industry, that is not what is happening. A major government-backed partnership between Toshiba and Rohm, designed to create a national champion to fend off a rising China, has reportedly stalled.

The basic situation is that Japan's makers of power chips—the essential "muscles" for electric vehicles and power grids—are getting squeezed. They face a slowing EV market and a flood of cheap Chinese competition that has sent companies like Rohm into the red for the first time in over a decade. The obvious solution is consolidation.

But the real story here is not about a temporary market slump; it's about a structural mismatch that feels eerily familiar. In the 1990s, Japan's vertically integrated giants lost the logic chip market to the specialized foundry model of Taiwan's TSMC. Today, a similar drama is playing out in power chips. Chinese competitors are specializing, dominating specific parts of the supply chain like silicon carbide substrates. Meanwhile, Japan's five major players remain fragmented, insular, and vertically integrated, a model that is proving inefficient and uncompetitive.

So if the strategic playbook is so obvious, why isn't anyone running the play? The problem, it seems, is one of trust and tradition. The industry is rife with deep-seated rivalries that even a national crisis cannot seem to overcome. As one insider explained to Nikkei Asia:

"Outsiders largely drive talk of consolidation," said a long-term employee of one major Japanese chip maker. "Company survival hinges on the capacity to develop products that meet customer specifications. With each company holding a wide-ranging product lineup, coordination is anything but simple." The employee said that power chip makers are careful not to share their product specifications even with customers, for fear of their know-how being exposed. "Trust is important," he said.

The Japanese government can throw billions of yen in subsidies at the problem, but it cannot force decades-old rivals to trust each other with their trade secrets. The technological gap with China is estimated to be closing fast—perhaps just three years in the most advanced chips. And so, Japan risks losing its second major chip war for the same reason it lost the first: not because it lacks the technology, but because it can't get out of its own way.

The Scoreboard

• Space: How the U.S. space industry became dependent on SpaceX (CNBC)
• Semiconductor: The Trump-Intel Deal Is Official (Wired)
• Streaming: Spotify flags price rises as it introduces new services, FT reports (Reuters)
• AI: AI’s Big Leaps Are Slowing. That Could Be a Good Thing. (WSJ)