China’s Lunar New Year tech showcase, Plus Figma’s Anthropic partnership 2/17/26

The Lunar New Year period produced a concentrated burst of AI announcements from Chinese technology companies that, taken together, suggest a more comprehensive competitive posture than individual model releases imply. The discussion covers releases from Alibaba and ByteDance alongside advances in robotics and semiconductors, arguing that the market may be underestimating the breadth of China's AI buildout. The framing used is a "full stack" analysis—examining China's competitive position layer by layer. At the foundation, basic chips powering robotics (sensors, processors) are already produced domestically, meaning China is self-sufficient at this level. Moving up the stack, memory chips represent a middle tier where global shortages are creating unexpected openings: prices have reportedly risen 90% in a single quarter, and Apple is said to be in discussions with Chinese chipmakers YMTC and CXMT to address supply gaps. This signals that even American companies are beginning to treat Chinese memory suppliers as viable partners under market pressure. At the top of the stack—advanced chips required to train large AI models—Nvidia retains dominance, but the gap is narrowing. Huawei's Ascend chips are already being used to train domestic models inside China. The broader pattern, as the discussion frames it, is that China is finding ways to "do more with less," a dynamic accelerated by U.S. export controls that effectively forced domestic innovation. ByteDance's Sea Dance video generator drew enough attention to prompt legal threats from Disney and Paramount—a signal of commercial-grade output quality. Alibaba open-sourced its most powerful model, and anticipation is building around DeepSeek's next release. These are not isolated product launches but evidence of coordinated ecosystem development.

A notable policy debate surfaces around whether the Trump administration's reported willingness to allow sales of older-generation Nvidia chips to China reflects strategic miscalculation or deliberate calculation. The discussion attributes to David Sacks and others in Silicon Valley a specific philosophy: cutting China off entirely accelerates domestic chip development by forcing necessity-driven invention, which is precisely what has occurred since export controls were tightened. The counter-argument is that permitting access to older Nvidia chips keeps Chinese developers partially dependent on the U.S. semiconductor ecosystem and limits the boost given to domestic accelerators like Huawei's Ascend line. This is an open question in the discussion—no resolution is offered—but the framing suggests that the debate has shifted from whether to engage to how much access serves U.S. competitive interests.

Figma announced a partnership with Anthropic centered on a feature called Code to Canvas, which allows interfaces built in AI coding tools—specifically Anthropic's Claude Code—to be converted into editable designs within Figma. The strategic logic, as articulated by Figma's CEO Dylan Field, is that as AI lowers the barrier to building software, the differentiated value shifts to what happens after initial creation: team collaboration, iteration, and deciding what is worth shipping. Field's stated philosophy is explicitly anti-lock-in: users should be able to enter a workflow from any starting point—a sketch, a design canvas, or generated code—and move fluidly between them. This is a meaningful departure from the traditional SaaS model of building switching costs into the product experience. The announcement is notable given Figma's context. The stock has declined sharply since its IPO last summer, consistent with a broader software sector selloff. The partnership with Anthropic—the same company whose coding tools represent an existential competitive threat to design software—drove shares up approximately 4% ahead of the company's earnings release. The discussion frames this as a case study in proactive disruption management, contrasting it with companies that have been slower to integrate AI tools. Field's advice to employees and fellow founders is to focus on controllable inputs—building the right things for customers—and let outcomes follow. He notes that the frequency of private startup CEOs seeking his IPO counsel has declined, suggesting cooling enthusiasm for public markets among late-stage private companies. The broader investor framing offered is a shift away from passive software exposure toward active evaluation of which companies are integrating AI offensively versus defensively—characterized as moving away from "lazy software investing." --- **Key takeaways:** - China's AI competitiveness is best understood as a full-stack buildout—from self-sufficient basic chips through memory to frontier model training—rather than competition on any single dimension. - A 90% quarterly price increase in memory chips and Apple's reported talks with Chinese suppliers indicate that U.S.-China chip interdependence is more complex than export control narratives suggest. - The U.S. policy debate over older Nvidia chip sales to China reflects a genuine strategic tension: restriction accelerates domestic Chinese innovation, while limited access may preserve ecosystem dependency. - Figma's Code to Canvas partnership with Anthropic represents a deliberate interoperability strategy—embracing the AI tools that threaten its core business rather than competing against them. - Investors are increasingly differentiating between software companies absorbing AI disruption passively and those restructuring their value proposition around it; Figma's 4% gain on the Anthropic announcement reflects that distinction.