Thursday. An eagle-eyed reader noted yesterday that we swapped a *million *for a *billion *in our coverage of the proposed Stripe-PayPal transaction. Apologies for the fat fingers. Today, we’re looking at TSMC’s mega earnings, a possible rebirth for American open-source AI, and why Twitter is screaming at the state of New York. To work! — Alex
📈 Trending Up #
Ethics over income…world models…the future of lawsuits? …Microsoft’s cybersecurity footprint…consumer spending, albeitless than expected…new models from China…Wonder……South Korean interest rates
**TSMC: **Shares of TSMC are off in early trading today, after the company reported second-quarter earnings. Did the Taiwanese chip giant underperform? No, compared to analyst expectations, the company’s revenue and net income came out ahead. TSMC also beat its internal guidance, as you can see below:
TSMC’s earnings per share rose 77% compared to the year-ago period and 23.4% compared to Q1 2026, against 36% revenue growth. That’s incredible operating leverage. Even better, TSMC expects revenue between “US$44.6 billion and US$45.8 billion” in the current quarter. A casual +10% result in a single three-month period. From a massive base.
Why are shares of TSMC falling in the wake of the news? The company’s planned capex is heading higher (now $60 billion to $64 billion this year), which may have spooked investors. Pledging an additional $100 billion for its Arizona footprint likely added to investor worries. That said, TMSC is not investing wildly. During its earnings call, the company said that rising prices could harm consumer demand for certain products, but that:
AI related demand continues to be extremely robust. […] Our customers and customers’ customer, who are mainly the cloud service provider, continue to provide us with their very strong signal and positive outlook. Thus, our conviction in the multi-year AI megatrend remains very high.
TSMC went on to state that rising CPU demand (thanks to agentic AI) is great news for its business, as “no matter what CPU approach is taken, whether it’s a x86, Arm-based, or RISC-V architecture, they are almost all TSMC’s customers.” That’s a flex. Regardless, as we hunt for fresh compute demand signals, indications that the AI capex boom may be fading, we find no evidence here.
**Thinking Machines Lab: **When Thinking Machines Lab announced Tinker last October, the drop didn’t reframe the AI conversation. Dubbed “a flexible API for fine-tuning language models,” we should have paid more attention. Not only was Tinker’s focus on helping other companies tune their own models early and correct, but the product has grown into a reportedly nine-figure business. (Do the meme!)
Well, what does a neolab with a fine-tuning business need? A model of its own. Correct. Thinking Machines Lab has one now:
Our model, called Inkling, is a Mixture-of-Experts transformer with 975B total parameters, 41B active. It supports a context window of up to 1M tokens. It was pretrained on 45 trillion tokens of text, images, audio and video. […]
Inkling reasons natively over text, images, and audio, and balances cost with performance through efficient and controllable thinking effort. We trained it to be a broad, balanced foundation model: strong across many domains, flexible enough to adapt. Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning
Clown on them, Mira. While TMLab talks down Inkling’s benchmark performance, it competes rather well with Nvidia’s Nemotron 3 Ultra model, which is perhaps the de facto American open AI model. Now we have two! More are welcome, as more competition is more good.
- Thinking Machines Lab now has a proven tuning product and a solid model to pair with it. I don’t fully understand the economics of offering tuning services, but if the market warms to Inkling, I do not see why the company can’t double or triple its top line, and thus economically derisk itself. Quickly.
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
AI usage limits…AI usage limits…grocery spending…the value of gold…shock, surprise…SpaceX shares, after an IPO pop…different human-agent password access…Chinese GDP growth…
Closed-source AI? Building off the Thinking Machines Lab news, Fireworks AI announced a $1.5 billion round at a $17.5 billion valuation, powered in part by its passing “$1 billion in annualized revenue run rate.” But more importantly, Fireworks helps customers tune models with their own data. You know, to chase better results and lower costs for their AI dollar. Expect related reporting from competing companies in short order.
SpaceXAI’s rep: Though progress is being made. Recently, developers discovered that Grok Build (Grok CLI) could upload a user’s codebase to private cloud storage controlled by SpaceXAI. Blowback was quick and torrential. In response to all the shouting, the company announced it would delete all stored data and worked to explain its data retention policies in non-sinister terms. The market was not completely soothed.
So, to protect its Grok 4.5 model — which people quite like — from withering away under a self-inflicted PR crisis, SpaceXAI open-sourced Grok Build. You can check out the code here and read developer-guru Simon Willison’s rundown of what’s inside here.
It’s worth noting that while Grok 4.5 is a strong model (especially in cost-performance terms), it’s not igniting the public AI charts. On OpenRouter, Grok 4.5 is less popular than OpenAI’s GPT-5.6 Sol, and when you add up total inference demand for the entire GPT-5.6 family, it’s not even close. SpaceXAI market share on the model router remains dramatically off-peak.
- We’re still looking at SpaceXAI pre-Cursor. That deal is ongoing and expected to close this quarter. Lots of developers like Cursor. Perhaps its eventual grafting onto SpaceXAI will resolve the issues that the larger company just ginned up for itself.
Everyone is mad at New York #
News that New York will enact the “first statewide moratorium on new hyperscale data centers” has the usual suspects up in arms. POTUS thinks the move is a terrible idea. Tech types are shouting about the state committing economic suicide. Industry is worried. The Journal’s Editorial Board is incensed.
Subscribe to Cautious Optimism to unlock the rest.
Become a paying subscriber of Cautious Optimism to get access to this post and other subscriber-only content.