Syz the moment

DeepSeek V4 Flash costs $0.09 per million input tokens. GPT-5.5 and Claude Opus 4.8 cost around $5. That's ~55× cheaper. Source: Lukas Ekwueme

China's central bank acquired +15 tonnes of gold in June, the largest monthly purchase in at least 2.5 years. This also marks the 20th consecutive monthly addition. Year-to-date, the country has increased its gold reserves by a total of +40 tonnes. This lifts China's total gold reserves to a record 2,346 tonnes, or 9% of its total FX reserves, near an all-time high. China is extremely bullish on gold. Source: Global Markets Investor

China's Producer Price Index (PPI) rose 4.1% YoY in June, the highest reading since July 2022 and the 4th consecutive monthly increase. Consumer inflation slowed to 1.0% from 1.2% in May, highlighting weak domestic demand even as manufacturers face rising input costs. Higher energy prices during the Iran conflict were the main reason producer prices increased. Source: Bull Theory, FT

Today Deep Seek and other AI open source Chinese AI has overtaken the US and now control about 52% of the market. Why? Because Chinese AI is much cheaper. Source: QE Infinity

China holds over $3.4 trillion in foreign exchange reserves, nearly 3x more than Japan. Seven of the world’s 10 largest reserve holders are in Asia, reflecting decades of export-led growth. Despite being the world’s largest economy, the U.S. ranks only 13th because the dollar is the world’s primary reserve currency. Source: James Wong, Visual Capitalist

Not because demand is slowing. Because costs are becoming impossible to ignore. Here are the numbers: 📈 Chinese AI models now process ~18.5 trillion tokens per week on OpenRouter. 🇺🇸 US models? Around 6 trillion. That's a 3x gap. Why? • Lower energy costs. • More efficient models. • Aggressive pricing that's reshaping the competitive landscape. Meanwhile, something interesting is happening inside large enterprises. Companies including Amazon, Walmart, Cisco, Uber, and Meta are reportedly introducing internal limits on AI usage as spending exceeds expectations. One striking example: A software company saw its AI bill jump 7x overnight after moving from a flat-rate subscription to usage-based pricing. Suddenly, the true cost of AI became visible. And this is only the beginning. 📊 Goldman Sachs estimates AI agents could increase token consumption by 24x by 2030. That creates a fundamental challenge: AI demand may keep exploding... while AI budgets become increasingly constrained. The next competitive advantage won't just be building the smartest models. It will be building the most cost-efficient ones. The AI story is evolving: ➡️ From "Who has the biggest model?" ➡️ To "Who delivers the lowest cost per useful output?" The winners of the next AI wave may not be those with the most compute... ...but those who make intelligence affordable at scale. Do you think AI spending is finally reaching a reality check, or is this just a temporary pause before the next investment wave? Source: FT, Global Markets Investors

Chinese domestic demand accounts for most of that growth. Source: Azeem Azhar

DeepSeek’s massive price cuts have made its AI token costs up to 50x cheaper than OpenAI and Anthropic, reshaping enterprise AI economics. Since hashtag#AI costs scale with token usage, companies running coding agents or reasoning-heavy models can spend millions—or even billions—annually. More advanced models consume huge hidden “reasoning” tokens, dramatically increasing compute costs. This is pushing firms toward cheaper, optimized models and tools, with companies like Microsoft and Uber already feeling budget pressure. The key competitive advantage in AI may shift from having the smartest model to delivering “good enough” AI at the lowest scalable cost. Source: Bull Theory