Apple and OpenAI partnership strain & Nvidia valuation and China trip - Tech News (May 15, 2026)

Starting with the uneasy state of big-tech AI alliances: Apple and OpenAI’s partnership is reportedly under strain. Sources say OpenAI is frustrated that ChatGPT’s role inside Apple software feels buried and hard to reach, falling short of expectations for driving paid subscriptions. Apple, meanwhile, has its own concerns around privacy posture, platform control, and OpenAI’s broader ambitions. If this escalates, it’s a reminder that “AI integration” deals aren’t just about features—they’re about distribution, branding, and who owns the user relationship.

That tension lands as new legal filings in the Musk v. Altman saga expose unusually candid messages about Microsoft’s posture toward OpenAI. The newly surfaced communications paint a picture of Microsoft trying to avoid being dependent on anyone else’s core advantage—whether that’s model IP or the hardware stack powering it. Beyond the courtroom drama, the takeaway is simple: the biggest AI partnerships are also contingency plans, and everyone is building an exit ramp.

On the markets and geopolitics side, Nvidia became the first publicly traded company to reach a $5.5 trillion valuation. The move reflects relentless investor confidence that AI infrastructure demand is still expanding—and that Nvidia remains the key tollbooth. Adding intrigue, CEO Jensen Huang is now expected to join President Donald Trump for meetings in China, which puts the company’s role in U.S.–China tech relations back in the spotlight at a sensitive time.

And speaking of that relationship: as Trump meets China’s Xi Jinping in Beijing, Taiwan is again described as the most delicate topic on the agenda. For the tech economy, Taiwan isn’t an abstract geopolitical talking point—it’s a central pillar of advanced semiconductor supply. Any perceived shift in U.S. support, or any increase in regional pressure, can ripple directly into planning for AI hardware, consumer electronics, and defense technology.

That context makes TSMC’s new forecast even more notable. The company now expects the global chip market to exceed one and a half trillion dollars by 2030, upgrading its previous outlook. TSMC is signaling that AI and high-performance computing are becoming the main growth engine for the entire industry, and it’s expanding manufacturing and advanced packaging capacity accordingly. The headline here isn’t just scale—it’s where the industry is placing its bets for the next decade.

Now to security, where the tone is getting more serious as AI tooling matures. A new industry argument making the rounds is that security-focused AI models could make vulnerability discovery and exploit development dramatically faster—turning what used to be specialized craft into something closer to an assembly line. Some of the language is heated, but the underlying concern is real: if exploitation becomes cheaper, defenders can’t rely on the old assumption that complex attacks are rare.

That concern is reinforced by a separate report from security researchers who say they found a method to bypass macOS protections while testing an early version of an AI-assisted security tool. The key point isn’t the specific chain, but what it signals: advanced AI-enabled testing may help uncover multi-step vulnerabilities faster, including the kind of bug combinations that historically took longer to piece together. For platform vendors, it raises the pressure to shorten patch cycles and tighten coordinated disclosure.

A broader view comes from Wiz’s latest cloud report, which argues AI is no longer a side project—it’s becoming core infrastructure. The report highlights a growing problem it calls “transitive AI,” where organizations end up depending on models and agents through third-party software, without clear ownership of the risk. If your AI capabilities arrive through a vendor, your exposure does too, and governance has to catch up.

Switching to space and connectivity: Amazon’s satellite broadband effort, now branded Leo, says it plans to increase launch tempo over the next year as it moves toward a wider commercial rollout. The company is also navigating regulatory deployment deadlines and the reality of launch capacity constraints. Even with faster progress, Amazon is still in catch-up mode versus Starlink, but the competitive pressure is intensifying—especially for enterprise and government connectivity where AWS integration could be a differentiator.

NASA, meanwhile, says it’s testing a new radiation-hardened space computing chip designed to bring far more onboard intelligence to future missions. The practical significance is autonomy: when a spacecraft is too far away for real-time control, extra computing power can mean better navigation, faster science decisions, and fewer mission interruptions. This is one of those advances that quietly raises the ceiling for what spacecraft can do on their own.

On the robotics front, Figure AI claims its humanoid robots can now run autonomously for a full factory-style shift without human intervention. The company is pointing to longer, uninterrupted task demos and improved whole-body coordination. As always, these are company-reported results, so it’s worth waiting for broader validation—but it’s another data point that humanoid robotics is moving from flashy prototypes toward sustained reliability.

In the developer world, there’s a growing sense that AI-assisted coding is changing what “lock-in” means. One prominent reflection notes that teams feel more comfortable choosing frameworks because they believe AI agents can help them rewrite or port later if the bet goes wrong. That mindset shift—treating big rewrites as painful but not impossible—could reshape how engineering leaders evaluate platform decisions.

Related to that, a small but telling trend is emerging around “/goal” style instructions for coding agents: define what done looks like, and let the tool run until it reaches that end state or hits limits. The interesting part isn’t the syntax—it’s the cultural shift toward treating AI agents as semi-autonomous workers that must be verified, not just prompted. In practice, this is about building trust through repeatable checks, like tests and clean builds, rather than taking an agent’s word for it.

And in local AI, Redis creator Salvatore Sanfilippo says his DwarfStar 4 project took off because local models have finally become good enough—and fast enough—for serious daily use on high-end personal machines. If that holds, it’s a meaningful inflection point: more AI workloads could move from the cloud to the desk, changing cost, privacy, and how quickly developers can iterate.

Let’s close with a quick round in health and science, where AI and bioengineering keep colliding in interesting ways. Researchers at the University of Pennsylvania introduced an AI-driven method for improving antimicrobial peptides through iterative optimization, and early lab results look promising. It’s not a near-term medicine, but it’s a credible approach to narrowing the search for new antibiotics as drug resistance keeps rising.

Harvard’s Wyss Institute also reported a platform for what it calls implantable living materials—essentially a tough hydrogel that can safely contain engineered bacteria inside the body for months. In mouse experiments, it helped reduce infections associated with orthopedic implants while keeping the bacteria confined. This tackles one of the biggest barriers for microbial medicines: control and containment.

In cancer research, another team demonstrated a drug-delivery “molecular grappling hook” concept that helps therapeutics stay anchored in tumors longer, improving effectiveness in mice and reducing side effects. They also showed the same platform could potentially support both imaging and therapy, which is a compelling direction for more targeted treatment strategies.

And finally, a more speculative idea: researchers in South Korea tested smart contact lenses that deliver mild electrical stimulation intended to influence brain circuits tied to mood. The early results in mice are intriguing, but the setup required abnormal vision conditions and faces big translation hurdles. For now, file it under ‘creative, early-stage neuroscience,’ not an impending depression treatment.

Before we go, back to that Bitcoin recovery story teased up top. A user says they regained access to five Bitcoin after forgetting an old wallet password for more than a decade. The twist is that an AI assistant didn’t “hack” anything—it helped search through messy old files, identify a forgotten backup, and spot a configuration mistake in the recovery process. It’s a neat example of what AI is genuinely good at today: organizing evidence, debugging workflows, and connecting dots that humans miss when the trail goes cold.