OpenAI burns, Alibaba chases code & more (September 8, 2025)

Good morning!

OpenAI’s cash burn projections are ballooning, Alibaba is chasing scale with a trillion-parameter model, and Anthropic just wrote one of the largest checks in copyright history. Big tech is moving fast, but the cracks are showing.

Let’s dig in👇

💰 Big Money & Big Bets

🏢 Alibaba unveils Qwen3-Max-Preview

Alibaba launched Qwen3-Max-Preview, a trillion-parameter model with a 262K token context window. Benchmarks show it outperforming Claude Opus 4, Kimi K2, and DeepSeek-V3.1 in reasoning and coding. Available on Qwen Chat, Alibaba Cloud, and OpenRouter, pricing starts at $0.86 per million tokens.

📚 Anthropic pays $1.5B in record author settlement

Anthropic agreed to settle a copyright lawsuit with authors for $1.5B, paying $3,000 per work allegedly used for training Claude. The deal requires destroying original files, though future claims remain open. Analysts say this could reshape how AI companies source data.

💸 OpenAI projects $115B cash burn

OpenAI expects to burn $115B through 2029, with $8B in losses this year and $45B annually by 2028. Costs are driven by soaring AI infra needs, pushing OpenAI to build its own chips with Broadcom. The bet: owning compute is cheaper than renting it from hyperscalers.

🔬 Science & Strategy

💰 Databricks projects $4B in annual revenue on surging AI demand

Databricks announced it expects $4B in annual revenue as demand for AI software surges across industries. This milestone underscores Databricks’ rapid growth and its role as a key competitor to cloud giants. With AI adoption accelerating, Databricks is well-positioned to capture a significant slice of enterprise spending.

🇪🇺 ASML takes stake in Mistral

ASML is investing €1.3B into Mistral, becoming its top shareholder and boosting Europe’s most valuable AI startup to a €10B valuation. The deal also gives ASML a board seat. For Europe, this is about AI sovereignty against US and China dominance.

🍏 Apple hit with AI copyright class action

Apple faces a lawsuit claiming it used pirated books to train its “OpenELM” models. If successful, this could mirror Anthropic’s $1.5B payout and force Apple into costly settlements. The case spotlights the mounting risks of gray-area training data.

Tools of the Day

→MemSync – Access and recall content across apps seamlessly
→SciMaster – “Your scientist friend,” built to accelerate research workflows
→ Steve.AI – Fastest way to turn scripts into videos using AI-generated voiceovers

Prompt of the Day

Fitness Plan Generator
“You are a certified personal trainer. Create a progressive, beginner-friendly fitness plan tailored to my data: [age], [height], [weight], [fitness level], [goal], [equipment], [schedule]. Include workout breakdowns, nutrition guidance, and progression strategy.”

Quick Hits

→ Medicare will pilot AI for coverage decisions starting Jan 2026
→ Higgsfield AI launches Ads 2.0, replacing full creative teams with a single mini app
→ Anthropic expands its ban on use in Chinese-owned companies
→ Moonshot AI drops Kimi K2-Instruct-0905, a trillion-parameter MoE model

TLDR

Alibaba is going all in on trillion-parameter models, Anthropic’s $1.5B settlement is the biggest copyright payout in US history, and OpenAI is burning money at a rate that makes even Tesla look frugal. Meanwhile, Europe is flexing its muscles with ASML’s stake in Mistral, AI research shows hallucinations rising with model capability, and Apple is now at the receiving end of AI-copyright litigation.

Until tomorrow,
David