Analysis
The Fights Multiplying Around AI's Biggest Names
Three unrelated disputes broke into public view within 48 hours this week, and together they trace the outline of a fight regulators and courts are only beginning to have with the AI industry. OpenAI filed a 31-page motion on Wednesday asking a federal judge to dismiss Apple's trade-secrets lawsuit, arguing Apple's own iCloud policies -- not any OpenAI wrongdoing -- created the credential gap a former engineer used to keep accessing company servers after he left, according to Bloomberg. The same week, Google entered advanced talks to pay roughly $1.5 billion to license technology from and hire staff at Mechanize, a 35-person AI coding startup, a structure that gets Google the team and the technology without the merger review a straight acquisition would trigger. And PC makers HP, Asus and Acer confirmed they've started using memory chips from China's CXMT in notebooks sold outside the US, a quiet concession to an AI-driven DRAM shortage that has pushed component prices up several hundred percent in 18 months, according to Semafor.
None of these three stories is really about the thing it looks like on the surface. The Apple-OpenAI case looks like a trade-secrets dispute; Pulse has covered the underlying suit since Apple filed it July 10, alleging former employees carried confidential hardware work to OpenAI's device team under the $6.4 billion acquisition of Jony Ive's iO. But the real fight is about who gets to define the next hardware category coming out of Cupertino and San Francisco, and courts are now the venue for a talent dispute that used to get settled with a non-compete and a handshake.
“None of these three stories is really about the thing it looks like on the surface.”
Why the Structures Are Changing, Not Just the Fights
The Mechanize deal is the clearer tell. Rather than acquiring the company outright -- the way OpenAI bought iO, or the way Google itself has bought AI teams in the past -- Google is paying for a non-exclusive license plus a hiring arrangement, following the same playbook it used with Windsurf and Character AI. That structure keeps the deal below the size and shape that typically draws a full FTC merger review, while still getting Google the model-evaluation staff it needs after Gemini's coding tools kept losing developers to Anthropic's Claude Code and OpenAI's Codex. It is an acquisition in every way that matters to the acquired company's roadmap, and not quite an acquisition in the one way that matters to antitrust regulators.
CXMT's arrival inside HP, Asus and Acer laptops is the same instinct playing out in supply chains instead of cap tables. AI data centers are expected to consume roughly 70% of global memory chip production this year, and PC makers facing DRAM prices up several hundred percent in some categories are choosing a workaround -- limited-volume Chinese chips in non-US models -- rather than eating the cost or waiting for Samsung, SK Hynix and Micron capacity that's already spoken for by hyperscalers. It's a narrow, contained decision today. It is also the first crack in a Western PC supply chain that has avoided Chinese memory almost entirely until now.
The Common Thread
Put the three together and the pattern is legal and structural engineering around AI's actual constraints -- talent, compute, and now memory -- rather than any single company doing something wrong. Apple is suing because it has no other lever to stop senior engineers walking out the door into a well-funded competitor's hardware group. Google is licensing instead of acquiring because a straight acquisition invites scrutiny a licensing deal doesn't. HP and Asus are buying validated-but-unproven Chinese memory because the alternative is a product that either doesn't ship or costs meaningfully more. Each company's response is a workaround, not a solution -- the underlying scarcity in senior AI talent, unrestricted M&A capacity, and memory supply hasn't gone away, it's just been routed around for now.
The counterweight worth stating plainly: none of this is proof AI is running out of room to grow. Apple's suit could just as easily be read as a company protecting real IP against real theft, not merely reacting to a talent drain; OpenAI's motion to dismiss argues exactly that framing is wrong, and a judge, not either company's press office, will eventually decide who's right. Mechanize's $1.5 billion price tag for a 35-person team six months removed from a $9.1 million seed round is also a number that only makes sense if AI coding tools keep compounding in value the way Anthropic's and OpenAI's own revenue growth suggests they might -- if that growth stalls, deals structured this aggressively around scarce talent will look expensive rather than prescient.
What GPs and Founders Should Actually Track
For portfolio companies competing for AI engineering talent, the diligence question is less "can we out-bid Google" and more "what does our retention plan look like against a market where a 35-person team can command $1.5 billion in acquihire terms." For anyone with AI hardware exposure, the CXMT story is worth tracking less for its 2026 impact -- HP and Asus both say volumes are small -- and more as the first data point on whether US brands are willing to normalize Chinese memory at scale if the shortage doesn't ease before 2027, which SK Hynix has separately warned could stretch past 2030. And for anyone with AI-related litigation exposure of their own, October 1 -- the date a federal judge in San Jose hears Apple's preliminary injunction request against OpenAI -- is now a real calendar date to watch, not an abstract one.
