Anthropic denies price cuts, StartLux ranks with 27B model, OpenAI invests $1 billion in security.
In a Bloomberg interview, Anthropic executives explicitly denied following suit with price cuts to grab volume, insisting on pricing based on capability. The subtext of this statement is: the gross margin structure for frontier models has not yet reached the point of a price war, and customers' willingness to pay for high-value scenarios remains intact. For secondary markets, this means the narrative of "cutting prices to gain volume" will not erode top labs' revenue models in the short term; pricing power for enterprise APIs still rests with those leading in capability. It is worth comparing this with OpenAI's discussion regarding Astra model pricing on the same day—one leader talks about capability premiums, the other about universal accessibility—the landscape of tiered pricing is solidifying.
Chen Danian, founder of NetEase and Dianping, returns to the front lines. His new company, StartLux, launched its first model with only 27B parameters, yet it ranked second overall in the China Academy of Information and Communications Technology (CAICT) specialized MCP tests, trailing only DeepSeek's flagship which had been kept under wraps for nearly a year. The first tier of domestic large models has always been an arms race of tens or hundreds of billions of parameters. This time, a small-parameter team achieved a leapfrog ranking through tool-calling capabilities, indicating that after evaluation metrics shifted from "knowledge density" to "Agent usability," the competitive landscape is indeed being reshuffled. The valuation logic for "cost-effective model" teams in primary markets may need rewriting: models with low inference costs generate higher unit compute revenue in the MCP/Agent era; the cost curve where 27B matches trillion-parameter flagships is itself a moat.
Two peripheral news items surrounding OpenAI's new model Astra deserve separate analysis. First, Bloomberg confirmed that Astra's launch included enhanced cybersecurity guardrails; simultaneously, OpenAI announced investing $1 billion in critical infrastructure protection and partnered with Cloudflare to offer vulnerability discovery capabilities for frontline defenders—security investment is shifting from PR cost to fixed capital item for frontier models. Second, Wired disclosed that OpenAI actively terminated major partnerships with top coding tool companies over the weekend, citing direct business conflicts related to Elon Musk; downstream ecosystems' dependency on single model suppliers was priced in for the first time in the form of "supply cutoff."
Cloud security firm Zscaler beat Wall Street expectations for both Q4 revenue and guidance. Management attributed growth to security spending driven by enterprise AI deployment—as employees begin bulk calling large models, zero-trust gateways become the new mandatory path. After-hours gains voted in favor of the judgment that "the next stop for AI beneficiaries is security and data layers." In the SaaS sector, companies that can frame AI as "realized revenue" rather than "future narrative" in earnings reports are seeing their valuation centers shift upward.
Memory chip company Longsys finalizes its H-share issuance price. In a window resonating with memory price hikes and AI demand, this price sets an anchor for the next batch of A-share semiconductor companies issuing H-shares: whether southbound funds give a premium to memory assets with dual "cycle + AI" narratives, or continue discounting them as traditional cyclical stocks, will be revealed by subscription multiples.
Bessent named US companies heavily investing in AI, criticizing their failure to do community communication well amid public concerns that data centers drive up local living costs. This means the "anti-data center" wave officially enters the Treasury Secretary's rhetoric; site selection, grid, and water price disputes will translate into substantive approval risks, and hidden costs of compute infrastructure begin to be repriced. CNBC also reported: US AI data center supply chain dependence on key Chinese components (power electronics, upstream optical modules) is surfacing; combining these two issues yields—political costs and supply chain costs of AI capex rising simultaneously. Planet Labs announced expanding satellite monitoring capabilities to data center inspections; controversy itself became business.
📰 EastVoice — today's AI headlines, decoded
Five minutes a day, the global AI shift →