1. Compute Hits the Economy (AI / Markets)
TSMC closed the second quarter with $40.2 billion in revenue and a 67.7% gross margin. It was the strongest quarter in the chipmaker's history, and the company expects revenue to rise again in the third quarter. 1
Meta expanded its Hyperion data-center plan in Louisiana from two gigawatts to five. The project is now expected to cost more than $50 billion. 2
The pressure is moving beyond the technology sector. AI infrastructure spending is likely to top $700 billion this year, lifting prices for memory, processors, electronics, and electricity. Some economists now expect the buildout to add about half a percentage point to core inflation by year-end. 3
Why it matters
The AI race is moving down the stack. Better models still matter, but chips and power now decide how many people can use them and what that access costs. TSMC's record quarter shows the demand. Meta's expansion shows the scale. Higher device and electricity prices show who else is paying.
Reality check
Strong chip sales prove demand, not a permanent shortage. Meta had already signaled that Hyperion could reach five gigawatts, and the full buildout will take years. The inflation effect is also an estimate. New capacity could eventually lower prices as quickly as today's demand is raising them.
2. China Makes Open AI a State Strategy (AI)
Moonshot AI released Kimi K3 on July 16. The model has 2.8 trillion total parameters and a one-million-token context window. In early blind tests, developers preferred it over the leading US models for front-end coding, while its broader text score tied GPT-5.6 Sol. 4
The model is available through an API, but the weights are not public yet. Moonshot says they will arrive on July 27.
One day later, China opened the World AI Conference in Shanghai. Xi Jinping used the event to promote open models and wider access to AI, while criticizing restrictions on technology sharing. 5
China also launched a new AI cooperation group with 28 other countries. The message was simple: the US can lead on the most powerful closed models, while China competes by making capable systems cheaper and easier to spread. 6
Why it matters
China is turning open models into industrial and foreign policy. Kimi K3 is the product, and the Shanghai conference supplied the political message around it. The competition is no longer only about who builds the best model. It is also about whose models become the default outside their home market.
Reality check
Kimi's full weights are not available yet, and the strongest results come from early tests. Open models still require expensive chips to train and run. Many regulated companies will also avoid Chinese systems because of data and legal concerns. The distribution strategy is real, but the outcome is not settled.
3. Robots Raise Money and Go to Work (Robotics)
Walden Robotics launched from stealth with $300 million at a $1.1 billion valuation. The company grew out of Toyota Research Institute, and says its robots have been doing production work at a Toyota plant since February. 7
China's LimX Dynamics raised another $200 million before a planned public listing. It has now raised $400 million in six months and says the money will support thousands of autonomous humanoid deployments. 8
China is already sending thousands of humanoids into logistics hubs, battery factories, and other industrial sites. Much of the work is still simple, but every deployment creates more real-world training data. 9
The capital and the deployment serve the same goal. Robots need factories to become useful, and factories need enough robots to produce the data that improves them.
Why it matters
Humanoid robotics is moving beyond demonstrations. The biggest advantage may not be one robot's hardware. It may be the feedback loop between funding, deployment, and data. China currently has more places to run that loop, while Walden shows US capital is trying to close the gap.
Reality check
The funding numbers are real, but the deployment claims mostly come from the companies involved. Simple sorting and factory tasks do not prove general intelligence or strong unit economics. A large robot fleet can collect useful data and still fail to become a durable business.
4. Wall Street Moves Stocks Onchain (Crypto / Markets)
DTCC converted securities held at its central depository into tokens and used them in real production trades on July 15. More than 30 firms took part ahead of a wider service launch planned for October. 10
The tokens were digital versions of assets still held inside the existing market system. They moved across private and public blockchain networks without removing the regulated custody underneath them.
Alpaca raised $135 million to expand the infrastructure behind tokenized US stocks. The company says it already clears or holds most tokenized US equities and more than $1.5 billion of the shares backing them. 11
Cantor Fitzgerald also began preparing for companies to place a small part of future stock offerings directly onchain. 12
Why it matters
Tokenization is moving from crypto exchanges into the plumbing of traditional markets. DTCC sits at the center of US securities settlement. Once that layer can move assets onchain, tokenized stocks stop looking like a side market and start looking like an upgrade to the existing one.
Reality check
The underlying shares still need a regulated custodian, and dividends, voting, and other corporate actions still run through old systems. The production trades were controlled, not a broad public launch. Tokenization may improve settlement without replacing the institutions that already control it.