ZanSpark Daily AI & Digital Marketing Briefing
April 3, 2026 | Covering Key Developments from April 2
AI Industry Hot News
【Doubao Model Daily Token Usage Exceeds 120 Trillion; Seedance 2.0 Enterprise Beta Launched】
At the AI Innovation Roadshow in Wuhan, Volcano Engine announced that as of March 2026, Doubao LLM’s daily token usage has exceeded 120 trillion, doubling within three months and growing 1,000x compared to its launch in May 2024. Growth is primarily driven by explosive AI video creation and accelerated AI agent adoption. 140 enterprises have now accumulated over one trillion tokens on Volcano Engine. Simultaneously, Seedance 2.0, the global SOTA video generation model, officially opened enterprise beta testing with industry-leading copyright and portrait security protections, supporting multimodal input and synchronized audio-video generation.
Key Takeaway: Doubao’s rapid growth confirms large-scale domestic AI agent deployment; Seedance 2.0 enterprise beta signals accelerating video AI commercialization with profound impact on advertising content and film production industries.
🔗 IT Home
【NVIDIA Blackwell Ultra Shatters All AI Inference Records in MLPerf v6.0】
MLCommons released MLPerf Inference v6.0 benchmark results, with NVIDIA Blackwell Ultra GPUs setting records across the board—the company’s 291st time leading this benchmark. A system of 288 GPUs achieved 2.49 million tokens per second throughput processing DeepSeek-R1 in offline mode. Through TensorRT-LLM software optimization, single GPU DeepSeek-R1 performance increased from 2,907 tokens/sec six months ago to 8,064 tokens/sec (2.77x improvement). Partner data shows this reduces per-token inference cost on existing hardware by over 60%. NVIDIA is the only platform to submit results across all new benchmarks (including text-to-video WAN-2.2).
Key Takeaway: The 2.77x performance gain from software optimization alone means enterprises can significantly reduce AI inference costs without hardware upgrades; lower inference costs will accelerate AI agent commercial deployment.
【Xinhua Globe Magazine: 2026 Defined as “AI Agent Explosion Year”】
Xinhua News Agency’s Globe magazine published an in-depth report on April 2, with experts identifying four key reasons 2026 marks AI agent scale deployment turning point: significantly improved long-task capabilities in new-generation models (Claude Opus 4.6 supports 14 hours 30 minutes continuous task execution); mature protocols like MCP and A2A enabling agents to directly access real-world systems; AI inference costs dropping over 95% within two years; and dense enterprise AgentOps governance framework development. The report warns: true agent-native ecosystems require 3-5 years to form, with marketing hype to be wary of.
Key Takeaway: Authoritative media confirms AI agents entering explosive turning point—valuable reference for enterprise digital transformation and marketing automation planning.
【Meta Builds 10 Natural Gas Power Plants for Hyperion AI Data Center, 7.5GW Capacity】
Meta announced plans to build 10 natural gas power plants in Louisiana for its Hyperion AI super data center, with combined output approximately 7.5GW—slightly higher than South Dakota’s entire power generation capacity. This would make Meta one of the largest electricity consumers in the US. This contradicts Meta’s previous commitments to renewable energy and nuclear power, sparking environmental controversy. Experts note methane leakage may harm the climate more than coal in the short term. The plan highlights extreme energy demands of AI infrastructure; nearly half of US data centers face delays or cancellations due to transformer shortages.
Key Takeaway: AI energy demand has exceeded traditional infrastructure capacity; power supply chain shortages will become a critical bottleneck constraining global AI compute expansion—energy and infrastructure investment opportunities merit attention.
【NMPA Issues “AI + Drug Regulation” Implementation Guidelines, AI Policy Accelerates】
On April 2, 2026, China’s National Medical Products Administration officially released “Implementation Guidelines for AI + Drug Regulation,” clarifying “digital-intelligent empowerment of key regulatory scenarios” and promoting AI scale deployment in drug approval, monitoring, and inspection. The same day, Ministry of Industry and Information Technology announced inclusive computing power services, aiming to significantly lower SME computing access thresholds by end of 2028 through inclusive computing power empowerment centers and AI application service providers. The Global AI Innovation Index report was also released, with Beijing becoming one of the world’s AI innovation “core hubs.”
Key Takeaway: Policy continues intensifying AI industry deployment; medical regulatory AI has strong demonstration effect; MIIT inclusive computing plan will effectively lower AI agent commercial baseline costs and promote SME AI adoption.
AI Innovation & Applications
【Alibaba Releases Qwen3.6-Plus, Domestic Programming Model Tops Rankings】
On April 2, 2026, Alibaba officially released the new-generation LLM Qwen3.6-Plus, equipped with a 1 million token ultra-long context window. In global programming benchmarks like HumanEval and MBPP, it improved over 25% compared to Qwen3.5, leading domestic models with less than 3% gap from the Claude series. The model supports direct frontend code generation from Figma designs (over 90% fidelity) and increased complex repository task completion rate from 47% to 82%. API access opened on Alibaba Cloud Bailian; enterprise beta shows engineer efficiency improved over 40%. Wan2.7-Image generation model was simultaneously released, solving the “AI standard face” homogeneity problem with precise color control and long-text rendering support.
Key Takeaway: Gap between domestic programming models and international top-tier continues narrowing; Qwen3.6-Plus有望成为中小软件团队替代 Claude 的重要选项; Wan2.7-Image offers practical breakthrough for AI visual content creation.
🔗 Alibaba Cloud AI Creation Navigation
【Google Launches Veo 3.1 Lite, Video Generation Cost Reduced Over 50%】
Google launched low-cost video generation model Veo 3.1 Lite, with generation cost reduced to $0.05 per second (720p), over 50% lower than the previous generation, supporting landscape and portrait formats. ByteDance’s Jimeng AI simultaneously released a CLI tool supporting one-command video generation, driving AI video generation toward agentization. Kling AI released AI Qingming short film “Paper Phone,” trending on social platforms for emotional nuance, demonstrating commercial potential of AI creative content generation.
Key Takeaway: Significant video generation cost reduction will accelerate AI video commercial applications in advertising marketing; CLI tools lower developer barriers, making batch AI video content generation reality and disrupting traditional video production industry.
AI Investment & Startups
【OpenAI Completes Largest Single Round in History: $122B, Valuation $852B】
OpenAI announced completion of its latest funding round, raising $122 billion at a current valuation of $852 billion, shattering tech company fundraising records. SoftBank, A16Z, NVIDIA, Amazon and others participated. According to TechCrunch and Crunchbase data, global startup funding in Q1 2026 reached $297 billion, 2.5x the previous quarter and exceeding any full year of global venture capital before 2019, with generative AI accounting for 48% of total funding. OpenAI’s single round accounts for over 41% of quarterly total funding, breaking its own record ($40 billion in March 2025).
Key Takeaway: Capital extremely concentrated in AI leaders; non-AI track funding severely squeezed; under high valuations, AI companies face stronger profitability pressure—H2 commercialization pace worth watching.
【Cognichip Raises $60M Series A, Pioneering “AI Designs AI Chips” Track】
AI chip design startup Cognichip announced completion of $60 million Series A led by Seligman Ventures, with former Intel CEO Lip-Bu Tan joining the board; total funding reached $93 million. The company pioneered the “ACI (Artificial Chip Intelligence)” track, using physics-informed deep learning models to automate semiconductor design processes, claiming over 75% chip development cost reduction and over 50% time savings. Meanwhile, domestic embodied intelligence company Xinghai Tu disclosed nearly 2 billion RMB Series B+ funding; domestic embodied intelligence funding exceeded 20 billion RMB in the past two months.
Key Takeaway: “AI designing AI chips” represents new direction for semiconductor design intelligence; domestic embodied intelligence leading companies seeing concentrated funding, industry accelerating differentiation with winner-take-all pattern emerging.
Digital Marketing Updates
【ChatGPT Ads Revenue Hits $1M in 6 Weeks, Self-Serve Platform Launching April】
According to Two Octobers’ digital marketing monthly report, OpenAI’s ChatGPT advertising business generated $1 million revenue from approximately 600 advertisers in 6 weeks (only shown to 20% of eligible users). User feedback shows high ad relevance with only 7% marked “low relevance.” OpenAI plans to launch self-serve ad purchasing in April 2026; early adopters may enjoy lower CPMs and less competition. Advertisers currently report issues including insufficient budget consumption and incomplete performance reporting.
Key Takeaway: ChatGPT ads entering early dividend period; self-serve launch impact on digital marketing landscape cannot be ignored; brands advised to test first and seize low-competition window.
【Google AI Overviews Appear in 58% More Queries YoY, Deeply Reshaping Search Marketing】
BrightEdge research shows Google AI Overviews appeared in 48% of search results in February 2026, a 58% year-over-year increase. AI overview overlap with organic rankings is stabilizing (healthcare highest at 24%, finance lowest at 11%). Google also expanded personal intelligence features to Gemini in Chrome (US users), integrating Gmail, Photos and other data for personalized search results, further reducing traditional SEO factor weights. Google also abolished multiple legacy ad policies since March 17, 2026, fully transitioning to AI-driven ad formats.
Key Takeaway: Google AI overview coverage rapidly increasing, directly impacting content marketing and e-commerce SEO relying on search traffic; brand exposure must shift from “ranking competition” to “entering AI answers” new paradigm (GEO/AEO strategy).
【Meta Updates Conversion Type Definitions; Microsoft Ads Simplifies Bidding Strategy】
Major digital marketing platform updates in April 2026: Meta reclassified “likes” and “clicks” from “click conversions” to new “engagement conversions” type; advertisers need to update reporting metrics with transition period recommendation to track both types. Microsoft Ads integrated Target CPA and Target ROAS into “Maximize Conversions” strategy, aligning with industry automated bidding trends and reducing operational complexity. Both changes signal ad platforms accelerating AI automated bidding management transition.
Key Takeaway: Advertisers must promptly update Meta reporting dimensions to avoid optimization metric misjudgment; Microsoft Ads bidding simplification signals toward full AI automation with further shrinking manual intervention space.
ZanSpark Generated: 2026-04-03 08:30
Disclaimer: Content based on public information for reference only, not investment advice.