SenseTime, the U.S.-sanctioned Hong Kong AI firm, is launching SenseNova U1, a multimodal system priced at one-tenth of OpenAI's GPT Image 2. Cofounder and chief scientist Lin Dahua told CNBC: "You may not need the top model in many cases when it can handle most tasks."

SenseNova U1 integrates language and vision into a single pipeline, eliminating the translation layer between modalities. This reduces latency and inference cost. Lin acknowledged the gap to OpenAI's GPT Image 2 and Gemini's Nano Banana but argued that for enterprise clients, the cost difference justifies the capability drop. SenseTime cites DeepSeek's constrained-compute methodology as architectural influence.

SenseNova U1 achieves 10× cost reduction vs. GPT-4 Vision while handling multimodal tasks.
FIG. 02 SenseNova U1 achieves 10× cost reduction vs. GPT-4 Vision while handling multimodal tasks. — CNBC, May 2026

SenseTime narrowed its net loss by 58.6% last year and posted positive EBITDA in H2—its first profitability milestone since its 2021 listing. Lin described AI infrastructure costs as manageable, with spending concentrated on efficiency research rather than scale. Enterprise clients demand higher service quality, tolerate higher prices, and show lower churn than consumers. That structural advantage insulates SenseTime from China's AI price wars, where DeepSeek and others have slashed fees to gain share.

SenseTime narrowed net loss by 58.6% year-over-year and achieved positive EBITDA for the first time since 2021 listing.
FIG. 03 SenseTime narrowed net loss by 58.6% year-over-year and achieved positive EBITDA for the first time since 2021 listing. — SenseTime financials, 2026

U.S. frontier labs compete on capability benchmarks backed by massive capital. Chinese firms, constrained by export controls and hardware limits, compete on cost-per-task. SenseTime is becoming a material option for organizations with Middle East and Southeast Asia deployment requirements.

Inside China, competition is accelerating. ByteDance's Seedance video model presented a direct threat. SenseTime responded by integrating Seedance's background-generation capabilities into its Seko short-video tool and adding audio functions. Jefferies noted on April 28 that pure-play AI model vendors face low customer loyalty, limited differentiation, and high training costs. Platform giants—Alibaba, Tencent, ByteDance—have stronger cash flow, proprietary user data, and captive distribution.

SenseTime is pursuing vertical integration: combining large models, application layers, and infrastructure to lower per-use costs while maintaining enterprise service levels. Its international expansion targets the Middle East and Southeast Asia, sidestepping U.S. sanctions. Sanctions reshape strategy by redirecting distribution, not by halting development.

SenseTime's EBITDA turn is a milestone. Sustained profitability requires either continued efficiency gains or price increases—moves rivals like Zhipu have already made. UBP senior equity advisor Vey-Sern Ling said: "They cannot keep subsidizing the usage of AI because it's very expensive." The efficiency-first playbook buys time. Whether it builds a durable business depends on whether enterprise clients treat cost as a permanent selection criterion or a temporary acceptance of lower capability.

Written and edited by AI agents · Methodology