The AI IPO window just cracked open, old AI drama is about to get a verdict, and the biggest release event of the year drops tomorrow morning.

This wasn't supposed to be a big week. And then Cerebras went public and shocked every analyst who thought they had a ceiling in mind, landing at a $95 billion market cap on day one and quietly rewriting the expectations for every AI company planning to go public this year. 

Meanwhile, the Musk vs. OpenAI trial wrapped testimony and sent a jury into deliberations with a decision that could restructure one of the most valuable companies on earth. Oh, and Google I/O kicks off tomorrow with more in the pipeline than any prior year, carrying leaked product after leaked product into a keynote that could genuinely reshape how you think about your AI stack before the week is over.

Welp.

This is the calm before the hot AI summer, and based on everything that happened this week, the summer is going to be scorching.

New AI companies going public. Old AI drama finally getting a resolution. Big releases literally hours away.

If your team slept through this one, they've got some catching up to do.

Here's your AI news that matters.

OpenAI rolled out remote control access to Codex inside the ChatGPT mobile app, and this one is bigger than it sounds.

Users can now monitor and control Codex agents running on other computers from iOS or Android, while the Codex app keeps running on the host machine.

It works on standard ChatGPT plans, though paid plans get better limits. Windows support is not available yet.

The app mirrors connected machines in real time: screenshots, terminal output, file changes, test results, and approval requests. OpenAI says more than four million people use Codex weekly.

The point is not basic screen sharing. It is persistent project context, active threads, plugins, and agent memory, so long-running work can keep moving while you commute, travel, or leave the desk.

What it means: The take is blunt: Codex is starting to look like a general knowledge worker’s always-on agent.

It still has bugs, but the comparison to Anthropic’s similar mobile control attempt was not close. For teams, approvals become the new workflow bottleneck.

Cerebras Systems gave AI investors the signal they were waiting for.

The AI chipmaker went public Thursday, and shares jumped 68% on the first day of trading. It closed with a $95 billion market cap, just short of the rare $100 billion first-day club.

Cerebras builds wafer-scale AI chips and high-speed inference systems for training and running large AI models faster than traditional GPU-based infrastructure.

Why the heat? The episode tied momentum to a multiyear OpenAI contract worth more than $20 billion, announced in January, plus an AWS partnership launched in March.

Demand was so high that the company adjusted its premarket position after being oversubscribed. SpaceX, OpenAI, and Anthropic now get a much warmer read on potential public-market demand.

What it means: This was not just a chip IPO. It was a pricing signal for the whole AI stack.

If Cerebras can nearly touch $100 billion on day one, the next private AI giants walk into negotiations with more leverage. Yuuuuup.

Google announced Googlebooks at the Android Show, a new category of laptops combining Android and ChromeOS with Gemini Intelligence in the middle of the experience.

The first devices are expected this fall, with Acer, ASUS, Dell, HP, and Lenovo attached on hardware.

The new part is screen-level context. Magic Pointer lets users point at something on the screen, then ask Gemini to understand it and take action.

The examples were practical: point at a date in an email to create a calendar event, or select two images to visualize them together.

Create Your Widget lets users generate desktop widgets from the web and personal Google apps like Gmail and Calendar. Pricing is not set, but the devices are expected to compete near Apple’s $600 MacBook Neo.

What it means: Google is poking Apple right where Apple Intelligence hurt most: useful personal AI that actually shows up.

If Gemini understands the screen, Gmail, Calendar, and Chrome, Googlebooks become less about hardware specs and more about cutting steps from everyday work.

Anthropic managed to turn a usage update into a developer trust mess.

Starting June 15, paid Claude users will no longer have programmatic usage counted against normal subscription limits. Instead, they get a separate monthly Agent SDK credit ranging from $20 to $200, depending on plan.

That hits workflows using claude -p, the Claude Agent SDK, GitHub Actions, and third-party Claude Code tools. Those workflows move into a smaller API-metered pool instead of the broader subsidized subscription usage many developers had been leaning on.

Anthropic framed the change as a new benefit. Developers did not buy it.

An Anthropic employee posted on X that users would not pay extra under the policy, then the post received a community note adding context. The backlash centered on bait-and-switch energy and reduced included usage.

What it means: The ugly part is not only the bill. It is the spin.

Claude won developer love by feeling generous and builder-friendly. Reclassifying work by interface, not intent, makes teams wonder which workflow gets taxed next. That is how trust leaks.

The United States and China held rare formal AI talks in Beijing, focused on guardrails for the most advanced models.

Treasury Secretary Scott Bessent led the U.S. side, which puts the meeting squarely in economic and national security territory.

The central concern was misuse by non-state actors, including hackers and terrorist groups. Both sides discussed best practices to reduce catastrophic risk from frontier models.

President Trump pushed light-touch, high-level guardrails designed to reduce extreme risk without slowing innovation. Bessent said the U.S. was willing to engage because it believes it still holds a clear AI lead.

But the chip fight did not move. The summit produced no agreement on easing U.S. export controls on advanced NVIDIA AI chips, even with Jensen Huang attending as part of the U.S. business delegation.

What it means: The real read is more skeptical than celebratory. Guardrails are nice.

The unresolved issue is model distillation, where U.S. labs say Chinese firms copy American model behavior. If that stays untouched, the talks look like optics with better lighting.

OpenAI and Apple’s partnership may be getting legally messy.

According to Bloomberg’s Mark Gurman and a New York Times report, OpenAI is considering legal action against Apple over ChatGPT inside Apple Intelligence, which was announced in 2024.

OpenAI reportedly believes Apple failed to meet key terms by making ChatGPT too hard to find and use. The whole point was supposed to be distribution. Apple’s massive user base would send people toward ChatGPT, and some of those users would become paid subscribers.

That reportedly did not happen at the level OpenAI expected.

Tensions rose after Apple confirmed in January that it had also struck a deal with Google to use Gemini for parts of Apple Intelligence, including the long-awaited Siri overhaul. OpenAI is now weighing a formal breach-of-contract notice.

What it means: This is distribution without discoverability, and that is nearly worthless.

Apple wanted outside AI help without making Siri look weak. OpenAI wanted new paid users. Wellllllll... Not really. If users cannot find the handoff, the partnership is mostly theater.

Google I/O starts tomorrow, and the leak list is not small.

Multiple credible leaks suggest Google is testing Gemini 3.2, including a Flash Live variant showing up in Google Cloud Console, which points to near-term production readiness instead of a far-off demo.

The bigger rumor is Gemini Spark from TestingCatalog: a persistent 24/7 agent that could manage inboxes, logged-in websites, connected apps, and multistep online tasks.

Google has confirmed Android XR glasses will be previewed at I/O, with expected partnerships involving Samsung and XREAL. Googlebooks are also expected to get deeper platform and developer details.

Chrome is another major surface. Google says Gemini in Chrome will handle browsing, summarizing, and image editing in the U.S. starting in late June for eligible users. Reports also point to Gemini Omni and a cheaper Gemini 3.2 Flash.

What it means: Google’s play is not one product. It is a stack: model, browser, laptop, glasses, and agent layer.

If the rumored cheaper Flash model lands near frontier quality, Anthropic has a pricing headache. If access is limited, nah, we’re good.