Note (March 2026): I wrote this piece before Arm officially unveiled its own data center CPU. That does not make the original argument irrelevant, but it does change the context in an important way. I am keeping the article largely as it is because the framework still helps explain what to watch. What has changed is that some of the questions discussed here are no longer purely hypothetical. They can now be read
Executive Summary Following CES, NVIDIA’s Vera Rubin platform did not introduce a dramatic shift in specifications. Instead, it clarified a broader direction. In the era of AI inference, the core challenge is shifting away from pure compute performance toward how context is managed. What the Vera Rubin platform reveals is not merely a next generation GPU, but a moment in which the platform itself begins to assume responsibility for memory. As long context and multi
Key Events of the Week: What Happened At its annual re:Invent conference, AWS introduced a series of new services, including the Trainium 4 chip, an AI agent platform, and a cloud security framework. Together, they signal a shift from showcasing AI technology to governing AI infrastructure. At the same time, the U.S. government indicated plans to increase support for robotics, automation, and manufacturing reshoring, aiming to embed AI capabilities into the nation’s industrial and security
Executive Summary Broadcom has long been one of Google’s key partners in building the TPU architecture. As Google plans to make its TPU technology available to external customers, this partnership is entering a new phase. Broadcom’s role is shifting from that of a co-designer to that of a modular integrator, serving as a stable and indispensable technical pillar within Google’s institutional framework. This position provides stability, but it also limits Broadcom’s ability to influence the
Executive Summary The growth of AI was once imagined as limitless, but energy is becoming its most tangible boundary. As power shortages emerge as the new ceiling of computation, competition among tech giants is shifting from who owns the most GPUs to who can govern the rhythm of energy. This article examines how seven major technology companies are redefining their relationship with power: Microsoft institutionalizes energy, building a governable system through long-term contracts and nuclear
Executive Summary AI development now stands between the prosperity of language and the turning point of understanding. While today’s large language models demonstrate remarkable generative power, they also reveal a fundamental limitation: they excel at imitating language but have yet to truly understand the world. This growth, detached from reality, suggests that beneath the surface of prosperity, signs of a bubble are beginning to appear. From Microsoft / OpenAI, which focus on language generation, to
Executive Summary Each collaboration OpenAI undertakes is more than a business transaction. It has become a focal point for global capital and industrial belief. Although the company has yet to establish a stable business model, it has already reshaped the rhythm of the global technology supply chain. This article argues that OpenAI is transforming industrial reality through reflexivity. Corporations and investors believe it can define the future, and that very belief is actively shaping the
Executive Summary Over the past two decades, Silicon Valley’s formula for success has been built on the belief in staying asset-light. The scalability of software and the power of network effects became the most efficient levers for growth, driving companies to pursue speed, scale, and operational lightness. Generative AI, however, is disrupting this model. From Microsoft and Amazon to Google, Meta, Apple, and Tesla, the world’s largest tech companies are collectively returning to a world
Executive Summary Jensen Huang’s “even free chips” remark is not simply about price. It is a deliberate attempt to reset the rules of competition. By shifting the focus from chip cost to power-constrained economics, total cost of ownership, and revenue per watt, NVIDIA positions itself as the designer of AI factories rather than a commodity supplier. In this framing, ASICs are relegated to the role of secondary components, their best prospects limited to niches or
Executive Summary NVIDIA recently announced two major moves: investing in Intel to co-develop custom x86 CPUs with NVLink, and partnering with OpenAI to build AI infrastructure at the scale of a million GPUs. These actions may seem independent, but they reveal the same trend: the bottleneck in AI is shifting from the number of GPUs to the efficiency of CPU–GPU integration. In this transition, NVIDIA is reinforcing cross-platform standards through NVLink, Intel is focusing on