Apple filed a 41-page lawsuit accusing OpenAI of stealing hardware secrets, then confirmed Siri is dropping ChatGPT for Google's Gemini this autumn. SK Hynix's Nasdaq debut closed up 13% on its first day, instantly valuing the memory-chip maker at over $1 trillion. Goldman Sachs told clients to buy Chinese AI models over Western ones, citing a cost gap Western labs can't easily close. Here's what each move means for who you build with, buy from, and trust.

Apple filed a 41-page lawsuit against OpenAI on July 10 in Northern California federal court, accusing former Apple design executive Tang Tan and engineer Chang Liu of stealing confidential hardware plans, supplier data, and internal codenames after joining OpenAI's device team. The complaint alleges Tan told recruits to bring Apple hardware to interviews and Liu downloaded dozens of confidential files. OpenAI, which paid $6.4 billion for Jony Ive's hardware startup io Products in 2025, called the claims baseless; within days, Apple also confirmed the next Siri will run on Google's Gemini instead of ChatGPT starting this autumn.

Why It Matters: This isn't a generic IP spat, it's two of the most valuable companies on earth (Apple worth $4.6 trillion) unwinding a partnership through a lawsuit and a product swap in the same week. For any founder hiring from a competitor or a fast-moving AI lab, it's a live example of how aggressive recruiting can turn into a courtroom problem, and how quickly a "strategic AI partner" can become a rival.

What To Do:

Source: Fortune

SK Hynix, the world's dominant supplier of the high-bandwidth memory chips that power AI data centers, closed its first day of Nasdaq trading on July 10 up 13%, in a listing that raised roughly $28-29 billion, the largest-ever US IPO by a foreign company. The offering was more than 7 times oversubscribed, and the stock's close valued the company north of $1 trillion, ranking it among the 15 largest companies on the US market. SK Hynix's chairman told CNBC that "demand is enormous."

Why It Matters: Investors aren't just betting on AI models anymore, they're betting on the physical chip supply that AI runs on, and they're paying record prices to get exposure to it. If your business relies on cloud AI pricing staying flat or falling, a memory market this tight, and this hyped, is a signal worth watching, not ignoring.

What To Do:

Source: CNBC

Goldman Sachs published a note on July 12 recommending specific Chinese open-weight models, including DeepSeek V4, Kimi K2.6, GLM-5, and Qwen3.5, telling clients they now rank among the best available at a fraction of Western pricing. The call follows CNBC reporting that Chinese models already handle up to 46% of US enterprise AI traffic on platforms like OpenRouter, with GLM-5.2 alone seeing adoption jump roughly 80-fold in a single week as customers chase costs reported 60-90% lower than OpenAI and Anthropic.

Why It Matters: A top Wall Street bank is now telling clients, in writing, to route production workloads to Chinese-built models, which was a fringe idea a year ago. For any business watching its AI spend climb, this is the clearest signal yet that "which model" is now a real cost-line decision, not just a technical preference, and that the gap Western labs need to close is measured in multiples, not percentage points.

What To Do:

Source: CNBC

British studio Particle6, founded by former actor Eline van der Velden, announced its AI-generated performer Tilly Norwood will star in "Misaligned," a feature film described as a coming-of-age story about an AI being with access to other people's childhood memories but none of her own. Van der Velden says the film, which began development this month, will be "funny, chaotic and self-aware." SAG-AFTRA immediately restated its position that Norwood is not an actor and that "creativity is, and should remain, human-centered."

Why It's Interesting: This is the first time a fully AI-generated performer has moved from a marketing stunt to an actual lead film role, and the union pushback is happening in real time rather than after the fact.

Key Takeaway: No release date or budget has been disclosed yet, so the real test is whether "Misaligned" gets distribution and an audience once it's finished.

Source: CBS News

Engineers and surgeons at UC San Diego used a 5-foot, 60-pound humanoid robot named "Surgie" to perform two laparoscopic gallbladder removals on live animals, with results published in Nature. In the first procedure, a teleoperated Surgie worked alongside a human surgeon; in the second, two Surgie units operated as a team with no human in the surgical role at all, controlled remotely via a teleoperation console.

Why It's Interesting: It's the first time a humanoid robot, rather than a fixed surgical arm like the existing da Vinci systems, has performed live surgery under remote human control, which changes the shape of what "remote surgery" could look like.

Key Takeaway: Researchers see this as a path to bringing surgical expertise to remote or underserved regions long before it's used on human patients.

Source: UC San Diego

Within hours of Apple's lawsuit going public, Elon Musk resurfaced his "Scam Altman" jab and accused Sam Altman of going from "stealing an open source AI charity" to allegedly trying to steal Apple's phone technology. Altman fired back on X: "homeboy you're the one selling public market investors on short-term space datacenters," then added that the clearest sign GPT-5.6 Sol is a strong model is that "elon is obsessed with me again."

Why It's Interesting: The two have been sniping at each other for years, but a real federal lawsuit against a mutual rival gave both of them fresh ammunition, and neither held back.

Key Takeaway: Entertaining as it is, the feud is a distraction from the actual legal exposure OpenAI now faces, worth remembering before taking either side's framing at face value.

Source: Moneywise

Context: Apple's lawsuit and its Siri-to-Gemini switch both come down to the same root issue: what happens when you depend on someone else's roadmap, and what leaks when people move between companies. This prompt helps you find that same exposure in your own business before it becomes a legal or operational problem.

Prompt: I run a [type of business] that depends on [specific AI vendor, tool, or platform] for [specific function]. Walk me through: what confidential information my team could be exposing to that vendor or to people who move between us and a competitor, what breaks in my operations if that vendor changed its product, pricing, or access tomorrow, and three concrete steps to reduce that exposure this quarter.

Do Not: Do not treat the output as legal advice, especially anything involving NDAs, trade secrets, or contract terms, have an actual lawyer review those.

If/Then: If the model can't name a specific, concrete exposure, tighten the prompt with more detail about your actual vendor relationships and data flows before trusting the answer.

Example: A product studio founder named Marcus ran this after reading about the Apple-OpenAI split and realized his three-person team had been pasting unreleased pricing sheets into a shared AI chat with no enterprise data controls. He moved that workflow to a paid tier with contractual data protections the same week.

An internal Meta memo reviewed by Reuters shows the company's in-house "Iris" AI chip, designed with Broadcom and manufactured by TSMC, completed six weeks of testing with no major issues and will enter production in September. Iris will work alongside, not replace, Meta's Nvidia and AMD chips to help ranking, recommendation, and generative AI features across Facebook and Instagram, as Meta targets 14 gigawatts of AI compute capacity by 2027, doubling this year's 7 gigawatts, on top of a planned $145 billion AI infrastructure budget for 2026 alone.

Why It's Important: Meta is the latest major buyer to start building its own AI silicon instead of relying entirely on Nvidia and AMD, following similar moves already underway at Google, Amazon, and OpenAI.

Business Takeaway: As more of the biggest AI buyers build custom chips to cut costs, expect that pressure to eventually show up in lower inference pricing across the market, not immediately, but directionally.

Source: CNBC

The New York Times and several other publishers filed a motion on July 9 asking a federal judge to sanction OpenAI for allegedly withholding evidence about how ChatGPT was trained on their copyrighted articles, in an escalation of the ongoing copyright lawsuit over whether AI training on news content qualifies as fair use.

Why It's Important: This is a discovery fight, not a ruling on the underlying copyright question, but sanctions motions signal the publishers believe they've found something OpenAI doesn't want examined closely, and judges take withheld-evidence claims seriously.

Business Takeaway: If your product trains on or summarizes third-party content, this case is the one to watch, since its outcome on "fair use" will likely set the template every AI company after it has to follow.

Source: TechCrunch

Blackstone Credit & Insurance, alongside Apollo and KKR-managed insurance vehicles, agreed to invest $5.34 billion into five behind-the-meter power generation projects being built by Williams to feed data centers and other high-demand energy users, with Blackstone taking a 49% stake while Williams retains operating control. The deal includes $4.4 billion earmarked directly for construction.

Why It's Important: Private equity and insurance capital, not just tech companies, are now underwriting the power plants AI data centers need directly, because the normal utility grid can't connect new demand fast enough.

Business Takeaway: If your AI costs are exposed to energy prices in any way, this is a sign that power, not chips or model access, may become the tighter bottleneck over the next few years.

Source: Williams

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas