Google is speeding up its own chip rollout, cutting the development cycle from two years per generation to two chips a year and eventually even more, as the company tries to stay ahead in the artificial intelligence race, Google Senior Vice President Amin Vahdat said on Wednesday at Semicon Taiwan 2026 in Taipei.
Who Gets to Move Faster
Vahdat delivered the keynote speech on Sept. 2, 2026, and laid out a plan that makes the scale of corporate power plain enough. Google wants to accelerate the pace of customized chip development, not because the public asked for it, but because the company wants to keep its edge in a race built on control over infrastructure, compute, and supply. The company also plans to double down on R&D in Taiwan by expanding local space 60%.
That expansion matters. It means more resources, more engineering capacity, and more concentration of decision-making in the hands of one giant firm. The people who actually build and maintain the systems don’t get a vote. The apparatus does.
Bottlenecks at the Bottom
Rani Borkar, president of Azure Hardware Systems and Infrastructure at Microsoft, said building massive new memory chip capacity will not resolve the supply chain bottlenecks holding back the AI infrastructure build out. She delivered her keynote speech at SEMICON Taiwan 2026 in Taipei on Sept. 2.
Borkar called for innovation to relieve limited components. That’s the language of the boardroom meeting the limits of its own machine. The problem, by her account, isn’t just production volume. It’s the choke points that keep the whole AI buildout dependent on scarce parts and tightly managed supply chains. When the bottlenecks tighten, the pressure lands on workers, suppliers, and everyone downstream from the corporate scramble for more power.
The source doesn’t offer any grassroots response here. No mutual aid network. No horizontal organizing. Just the familiar spectacle of giant firms trying to engineer their way around the limits they’ve created, while the costs of that race get pushed outward.
Long-Term Needs, Long-Term Control
Marubeni President Masayuki Omoto told Nikkei in a recent interview that the company is emulating U.S. investment conglomerate Berkshire Hathaway’s strategy of watching carefully and striking only at the best opportunities, like a disciplined batter in baseball. He said trading houses have the ability to lay the groundwork for long-term global changes.
Omoto said Marubeni should focus on long-term needs, not chase chip trends. That’s a cleaner phrase for the same old hierarchy: wait, watch, move when the profit is right, and call it strategy. The company also targets an 80% cut in AI data center power use with a Spanish tie-up.
That power-use target sits right beside the rest of the story. These firms are not just building chips and data centers. They’re building the machinery that decides who gets access, who gets squeezed, and who pays for the energy hunger of the AI race. The language is all efficiency and innovation. The reality is concentration.
Google’s move to accelerate chip development, Microsoft’s warning that more memory chips won’t fix the bottlenecks, and Marubeni’s push for long-term positioning all point to the same thing: a corporate scramble to control the infrastructure beneath AI, with Taiwan serving as one of the key sites where that control gets expanded, financed, and normalized.
The keynote speeches on Sept. 2 came from the top of the stack. The people at the bottom will live with the consequences.