TechLook
The AI industry just produced two things that don't fit the usual narrative: a courtroom where $134 billion is at stake, and a Chinese regulator who just rewrote the rules of who can own AI companies. Both stories are about the same underlying shift — the messy collision between markets, power, and control. Here's what matters for builders.
Musk vs Altman Trial Kicks Off — And $134B Is the Opening Bid
The legal fight between Elon Musk and OpenAI started with jury selection this week, and the numbers behind it are not subtle. Musk alleges he was misled during OpenAI's restructuring from nonprofit to for-profit, and he wants Altman removed, the restructuring unwound, and up to $134 billion in damages directed back to OpenAI's nonprofit. Both OpenAI and xAI are preparing to go public, which means this case isn't academic — the outcome shapes two IPOs.
Here's everything you need to know:
- Musk and Altman cofounded OpenAI in 2015 as a nonprofit research lab
- OpenAI is now worth nearly $1 trillion; Altman converted it to for-profit in 2024
- Musk calls the conversion a breach of founding agreement; OpenAI calls the lawsuit "a harassment campaign driven by Musk's ego, jealousy, and desire to slow down a competitor"
- Microsoft's Satya Nadella, OpenAI president Greg Brockman, and former OpenAI executives are on the witness list
- Both Musk and Altman are expected to take the stand
- An OpenAI IPO would be among the largest in tech history; an adverse ruling could block or restructure it
- xAI's Grok is positioned as a direct competitor — a win for Musk benefits xAI commercially
The courtroom outcome will matter more than most legal fights because both companies are preparing to go public. For founders: the for-profit conversion of research labs has legal exposure that wasn't obvious at the time. The structures built in the next few years will face scrutiny when these companies reach public market size.
OpenAI Is Building Its Own Phone. Here's Why That Sentence Is Serious.
OpenAI is reportedly developing its own smartphone, partnering with MediaTek, Qualcomm, and Luxshare on hardware that would run on OpenAI's own operating system — with AI agents replacing traditional apps as the primary interface. Sam Altman cryptically posted that it's time to "seriously rethink" how operating systems and user interfaces are designed. Ming-Chi Kuo puts production in 2028, alongside a ChatGPT smart speaker, smart glasses, and smart lamp.
Here's everything you need to know:
- Hardware partners include MediaTek, Qualcomm, and Luxshare — established supply chain, not a garage startup
- The OS would be OpenAI's own, not Android or iOS
- Agents replace apps as the primary interaction layer — no traditional app store model
- This is a bet that the next computing platform looks nothing like the last one
- OpenAI is not alone: Apple is reportedly in early discussions for its own AI-first device, per reports
- Altman has been signaling platform ambitions for over a year — this is execution, not speculation
For builders: if AI agents replace apps as the interface, every product strategy built on app-store distribution faces a structural rewrite. This isn't certain, but it's credible enough that you should be thinking about agent-compatible interfaces now, not later.
Xiaomi Just Open-Sourced a Frontier Model at One-Fifth the Price
Xiaomi released MiMo-V2.5-Pro, an open-source agentic model that ranks 8th globally and 2nd among Chinese LLMs on coding tasks — surpassing Claude Opus 4.6. It supports up to 1 million token context, runs at 60–80 tokens per second, and costs $1 per million tokens to input versus Claude Opus 4.6's $5. That's not a discount. That's a different cost structure for any application that processes large amounts of text.
Here's everything you need to know:
- MiMo-V2.5-Pro: 42B active parameters, hybrid attention with 7:1 hybrid ratio
- Ranks 8th globally, 2nd among Chinese LLMs on Artificial Analysis Intelligence Index
- 1M token context window — useful for full codebases, legal documents, and long research tasks
- Inference speed: 60–80 tokens/sec
- Pricing: $1/M input, $3/M output — versus $5/$25 for Claude Opus 4.6 at comparable tier
- Partnered with OpenClaw, OpenCode, KiloCode, Blackbox, and Cline on agent frameworks
- Available via OpenRouter and Xiaomi's official platform; global open-source release imminent
- One week of free API access during the Hunter Alpha test phase
The model matters less than the ecosystem. Xiaomi is building partnerships with the agent frameworks developers actually use. If the tooling is good and the price is one-fifth the alternative, adoption follows. For builders: price-to-performance is a competitive moat that compounds fast.
China Just Rewrote the Rules of Who Can Own AI Companies
Meta's planned $2 billion acquisition of Manus fell apart this week after China's National Development and Reform Commission ordered both companies to unwind the deal. This is notable not because Manus is worth $2B — it's notable because Manus had already moved to Singapore, and China still claimed the authority to block it. The order came weeks before Trump's planned May meeting with Xi in Beijing.
Here's everything you need to know:
- Meta announced the Manus deal in December; China opened a probe in January
- The two teams were "deeply integrated" at Meta's Singapore office; Manus's website already read "now part of Meta"
- Manus executives were reportedly barred from leaving China during the probe
- China extended the signal: AI labs including Moonshot AI and StepFun have been told to reject US capital without explicit government approval
- The capital map for AI is now splitting into two, with each government claiming authority over startups on both sides of the border
For founders building AI products: this is no longer a US-only conversation. Any company with Chinese founders, Chinese data, or Chinese investors faces a two-government review that wasn't there six months ago. The strategic planning horizon just got shorter and the geopolitical risk premium just went up.
⚡ Quick Hits
-
GitHub Copilot: Shifting from request-based to token-metered billing — the inference costs from agentic coding are real and getting passed to enterprise buyers.
-
Skye: Raised $3.58 million before launch to build an AI widget homescreen for iPhone — surfacing tasks, email drafts, fraud alerts via widgets where users already glance ten times a day. Backers include a16z, True Ventures, and SV Angel.
-
Google: AI agents course hits 1.5 million in first cohort, second cohort starts June 15–19. Free, five-day, "vibe coding" centered. Google is seeding its stack.
-
Xoople: Raised $130 million to build satellite data infrastructure for AI — positioning itself as an "Earth system of record" for training and inference data.
-
Leaky Lemonade: Offering 50% insurance discount per mile driven on Tesla FSD — because insurance data shows FSD miles are twice as safe as manual driving. Autonomy is getting priced in.
-
Lovense: Integrating OpenClaw into its Remote app — turning a device controller into an AI agent that can plan trips, check calendars, automate smart-home settings, and sync intimate devices.
Techlook — AI & tech signal for founders and builders.