Good morning, it’s Thursday. China’s AI ambitions are hitting a wall as newly built data centers sit mostly idle, H&M moves forward with AI-generated model clones, and Microsoft rolls out powerful new research tools inside Copilot.

Plus, in today’s Forward Future Original, we break down how to master AI image creation using DALL-E, Midjourney, and Ideogram—covering strengths, styles, and smart prompt tactics.

Read on!

📊 MARKET PULSE

NVIDIA Faces $17 Billion Revenue Threat

NVIDIA is confronting a significant financial challenge as China’s recent regulatory measures threaten approximately $17 billion in annual sales. China accounts for about 13% of NVIDIA’s total revenue, and the new rules could substantially impact this segment. → Continue reading here.

🗞️ YOUR DAILY ROLLUP

Top Stories of the Day

👥 H&M to Use AI Model Clones in Ads
H&M will debut digital “twins” of real models in upcoming ad campaigns, created using AI with full consent and compensation agreements. The initiative allows for customizable, photorealistic fashion imagery without traditional photoshoots or physical presence. While efficient, the move raises concerns over the future of creative jobs in fashion. H&M says models retain control over their digital likenesses, a key distinction in an ethically gray area.

🚀 Databricks Boosts AI Without Clean Data Using TAO
Databricks unveiled a new technique—Test-time Adaptive Optimization (TAO)—that boosts AI model performance without clean, labeled data. The method combines reinforcement learning with synthetic data, selecting the best outputs to fine-tune models. Tested on a financial benchmark, it helped a small Llama model outperform OpenAI’s GPT-4o. This approach may accelerate custom AI deployment for clients lacking ideal training datasets.

🛍️ Amazon’s AI ‘Interests’ Tracks Your Shopping Passions
Amazon’s new AI feature, Interests, continuously scans its store to find products that match users’ hobbies, preferences, and budgets—all from a single custom prompt. The tool uses large language models to understand natural language requests and deliver relevant updates, restocks, and deals. In limited U.S. rollout, Amazon’s Interests uses generative AI to act as a 24/7 product scout, matching items to user preferences and making shopping more intuitive.

🧐 Microsoft Adds Deep Research AI Tools to Copilot
Microsoft is rolling out Researcher and Analyst, two new AI-powered deep research tools inside Microsoft 365 Copilot. Built on OpenAI’s latest reasoning models, the tools combine web search with internal work data to perform tasks like strategic planning and complex data analysis. Analyst can even run Python to crunch numbers. Launching in April via Microsoft’s new Frontier program, these tools aim to make Copilot a serious contender for enterprise research.

🇨🇳 CHINA

China’s AI Data Center Boom Collapses Under Its Own Weight

The Recap: China rapidly built hundreds of AI data centers to ride the generative AI wave, but many now sit unused as demand lags and project viability crumbles.
A mix of government hype, speculative investment, and poor planning has led to massive underutilization—up to 80% of new computing resources remain idle.
MIT Technology Review’s Caiwei Chen details how the rise of DeepSeek’s efficient open-source model upended the business model for China’s AI infrastructure.

Highlights:

• Over 500 new AI data center projects were announced in China during 2023–24, but only 150 became operational—and up to 80% of their capacity remains unused.

• NVIDIA H100 servers configured with eight GPUs now rents for 75,000 yuan per month, down from highs of around 180,000.

• DeepSeek’s R1 model, matching ChatGPT o1’s performance at a fraction of the cost, shifted the market focus from training to efficient deployment.

• Many data centers were built in remote regions with cheaper electricity but now struggle due to high latency and lack of local talent or clients.

• Some projects were reportedly used to exploit government subsidies—e.g., reselling green electricity or securing state-backed loans via land deals.

• Local governments backed these ventures to stimulate post-pandemic economies, often prioritizing short-term visibility over long-term feasibility.

• Despite the failures, China’s central government continues to promote AI infrastructure, and companies like Alibaba and ByteDance are investing billions in alignment.

Forward Future Takeaways:
China’s AI infrastructure bust underscores a central tension in tech-driven policy: infrastructure alone doesn’t guarantee innovation. As the AI landscape pivots from brute-force training to inference and deployment, agility and strategy matter more than sheer scale. Policymakers and companies alike must ask—are we building for hype, or for real, evolving needs? → Read the full article here.

👾 FORWARD FUTURE ORIGINAL

How to Create Stunning Visuals with DALL-E, Ideogram, and Midjourney

Artificial intelligence has made remarkable progress in recent years, especially in the field of generative media. While text-based AI models such as GPT-4 are revolutionizing the way we create content and interact with information, we are experiencing a parallel revolution in AI-powered image generation. Tools such as DALL-E, Ideogram and Midjourney have blurred the line between human and machine creativity, allowing people with no artistic background to create stunning visual content.

These technologies are based on complex neural networks that have been trained with millions of images to convert text input (prompts) into visual representations. The quality and versatility of the images produced has improved at a rapid pace - from simple, often flawed visualizations to photorealistic representations and artistically sophisticated works. → Continue reading here.

🤐 CENSORSHIP

Leaked Database Reveals China’s Use of AI to Supercharge Online Censorship

The Recap:
TechCrunch obtained a leaked dataset revealing that China is using a large language model (LLM) to automate and expand state censorship. The system flags politically sensitive content—ranging from anti-corruption complaints to satire about Taiwan—with far greater precision and scope than traditional methods. Researcher Xiao Qiang and rights advocate Michael Caster confirm the data points to state-level efforts to strengthen information control through AI.

Highlights:

• The leaked dataset contains 133,000 examples used to train an LLM to detect politically sensitive content in Chinese-language media.

• Content flagged includes complaints about poverty, police corruption, political satire, and references to Taiwan or military affairs.

• The dataset was discovered on an unsecured Baidu-hosted Elasticsearch database by security researcher NetAskari.

• A prominent example includes the idiom “When the tree falls, the monkeys scatter,” flagged for its metaphorical critique of power transitions.

• The system’s architecture mimics prompt-based LLMs like ChatGPT, with code referencing “prompt tokens.”

• The dataset references its purpose as “public opinion work,” a term linked to the Chinese government’s propaganda and censorship apparatus.

• The Chinese Embassy denied wrongdoing, stating China “attaches great importance to developing ethical AI.”

• OpenAI previously reported that Chinese actors used LLMs to monitor dissident activity and produce smear content against activists like Cai Xia.

Forward Future Takeaways:
This story underscores how authoritarian regimes are rapidly incorporating advanced AI into their censorship and surveillance arsenals. By using LLMs, the Chinese state can detect not just overt dissent but also subtle, metaphorical critiques at scale—making resistance harder to express and easier to silence. As generative AI becomes more capable, its dual-use potential raises urgent ethical questions about how this technology should be governed and by whom. → Read the full article here.

Layering Constraints for Better Prompts

Want your AI outputs to be sharper, shorter, or more on-brand? Add layered constraints—specific instructions stacked together—to get tighter, more controlled responses. → Continue learning.

🛰️ NEWS

What Else is Happening

🎬 OpenAI Shares Ghibli Mashups: Sam Altman used ChatGPT to generate anime-style images in the spirit of Studio Ghibli—stirring buzz and backlash.

🤝 OpenAI Backs Rival's Data Protocol: ChatGPT will support Anthropic’s MCP, letting AI tap into real-time data from apps and tools.

📈 Anthropic Teams With Databricks: Claude AI models are coming to Databricks, letting 1,000+ firms build smarter agents on their own data.

📋 Character.AI Adds Teen Usage Reports: Teens can now send parents weekly recaps of which bots they talk to — but not what they’re saying.

🕹️ NVIDIA Debuts On-Device Gaming AI: G-Assist runs locally on RTX GPUs to tweak settings, overclock, and answer gaming questions — no cloud required.

⚱️ Xavier Niel Warns on AI Lag: Europe risks becoming a “museum continent” if it misses its AI moment, says tech mogul Xavier Niel — speed is critical.

💰 Tim Cook Visits China’s AI Hotspot: The Apple CEO stopped in Hangzhou, home to rising AI star DeepSeek, and donated $4.1M to Zhejiang University.

⚖️ Anthropic Wins Lyrics Lawsuit: A judge denied music publishers’ bid to halt Anthropic’s AI training, citing vague harm claims and unsettled fair use law.

📽️ VIDEO

Google Gemini 2.5 Pro is Insane...

Gemini 2.5 Pro stuns with one-shot coding: from solving Rubik’s Cubes to building games, simulations, and complex web apps—all with zero edits. Fast, free, and insanely capable. Get the full scoop in Matt’s latest video! 👇

🧰 TOOLBOX

Real-Time Automation, 3D Emotes for Gamers, and Smarter Video Captions

🤖 OWL by CAMEL-AI: Boosts multi-agent task automation with real-time search, browser control, multimodal input, and document parsing.

🎮 Kinetix AI: Instantly turn videos into 3D emotes for games with AI—boost player expression using Unity, Unreal, or custom engine tools.

🔤 ByteCap AI: Instantly create accurate, customizable video captions with AI—enhance engagement using fonts, colors, emojis, and effects.

🗒️ FEEDBACK

Help Us Get Better

🤠 THE DAILY BYTE

140 Million Miles Away and Still Posing for the Camera

That’s a Wrap!

❤️ Love Forward Future? Spread the word & earn rewards! Share your unique referral link with friends and colleagues to unlock exclusive Forward Future perks! 👉 Get your link here.

📢 Want to advertise with Forward Future? Reach 450K+ AI enthusiasts, tech leaders, and decision-makers. Let’s talk—just reply to this email.

📥 Got a hot tip or burning question? Drop us a note! The best reader insights, questions, and scoops may be featured in future editions. Submit here.

🛰️ Want more Forward Future? Follow us on X for quick updates, subscribe to our YouTube for deep dives, or add us to your RSS Feed for seamless reading.

Thanks for reading today’s newsletter—see you next time!

The Forward Future Team

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀