One company rewrote its Microsoft deal, signed a $38 billion cloud agreement with Amazon, prepped a super app, and fought Elon Musk in federal court.

OpenAI's week was that loaded.

Add in big tech earnings that finally buried the AI bubble narrative and a new Microsoft platform quietly watching every AI agent inside your company — and yeah.

You needed this one.

A high-stakes legal battle between Elon Musk and OpenAI kicked off last week in an Oakland federal courtroom, and week one was Musk's show.

Musk spent three days on the stand, arguing the company betrayed its founding nonprofit mission by converting to a profit-driven model. His core claim: you can't just hand a charity to shareholders. OpenAI's counter is that its nonprofit parent still exists and that Musk himself supported the for-profit shift before walking away.

Brockman is scheduled to testify this week. Altman takes the stand the week of May 11. Satya Nadella may also be called, with Musk alleging Microsoft improperly funded OpenAI's commercial transition.

Starting this week, the courtroom streams live audio on YouTube.

One early bombshell worth flagging: Musk essentially admitted xAI used distillation techniques on OpenAI models to help train Grok. He wasn't exactly direct. But reporters in the room said the implication was clear.

What it means: If Musk wins, it doesn't just hit OpenAI. It reshapes how for-profit AI conversions work across the industry. Week one was Musk's show. This week OpenAI gets to respond, and that's where it gets genuinely interesting.

Reports out last week confirmed the Department of Defense signed new agreements with NVIDIA, Microsoft, Amazon Web Services, and Reflection AI, clearing them to deploy AI on its most sensitive classified military networks.

These four join SpaceX, OpenAI, and Google, which had already reached similar deals. That's effectively the entire AI industry in one room.

Except one company.

The agreements cover Impact Level six and seven systems, the highest US security classifications reserved for national security data. The Pentagon says the move specifically targets AI vendor lock-in prevention and long-term flexibility.

Anthropic is still in court with the Pentagon after resisting demands for unrestricted access to its AI. Anthropic recently won an injunction blocking the Defense Department from labeling it a supply chain risk, so they're not out of the fight yet.

What it means: When every other major AI player has signed and you're the holdout, negotiating leverage shrinks fast. Anthropic may be right on the substance. But that position gets harder to hold when literally everyone else in the room has already said yes.

This past week, four of the world's largest tech companies reported quarterly earnings, and the numbers made the AI bubble crowd very quiet.

Alphabet posted $109 billion in revenue with 22% growth, marking its 11th consecutive quarter of double-digit gains. Revenue from AI-powered products soared nearly 800%, and demand for AI compute actually outstripped supply. Amazon's cloud division, AWS, climbed 28% year over year to more than $37 billion, setting a record quarter-over-quarter jump. Meta and Microsoft rounded out a very strong reporting cycle.

Welp. The dot-com bubble comparison was always weak. The late nineties ran on speculation. This runs on six of the seven largest companies in America generating real, growing, enormous AI-driven revenue.

What it means: One bad quarter from any of these giants and the bubble talk comes roaring back. The data right now says otherwise. Real revenue is a very different story than hype, and right now revenue is winning that argument by a lot.

Microsoft announced this past week the general availability of Agent 365, its new control platform for organizations to discover, govern, and secure AI agents running across their workplaces.

The platform covers agents built by Microsoft, its partners, and third-party developers — including agents running locally on devices, inside SaaS tools, and across multiple cloud providers. New discovery tools through Microsoft Defender and Intune specifically let organizations identify shadow AI, including local coding agents like OpenClaw or Claude Code, see where they run, and block or restrict them if needed.

Agent 365 also supports syncing agent inventories from AWS Bedrock and Google Cloud.

It's available through Microsoft's 365 E7 plan or as a standalone license at $15 per user per month.

What it means: Shadow AI is already running inside most large companies right now without oversight. Microsoft is betting enterprises want one familiar platform to manage all of it. Given how much of enterprise infrastructure already runs on Microsoft, that's not a small bet to fade.

OpenAI and Microsoft announced a revised partnership this past week, and this update actually changes the competitive landscape.

The deal removes Microsoft's exclusive license to OpenAI technology, allowing OpenAI to sell its products across any cloud provider. Microsoft stays the primary cloud partner and gets first access to new OpenAI launches — but only if Microsoft can support the required capabilities. Microsoft retains a nonexclusive license to OpenAI's models and IP through 2032, revenue share payments continue through 2030 now with a cap, and Microsoft's stake in OpenAI remains valued at more than $135 billion.

One notable removal: the AGI clause, which tied revenue-sharing to an official declaration of AGI. Gone.

What it means: This deal made everything else OpenAI did this week possible. Without killing exclusivity, the $38 billion AWS partnership announced the very next day couldn't have happened. The era of OpenAI operating exclusively inside Microsoft's walls is over.

According to OpenAI, the company rolled out a major redesign of its Codex desktop app this past week — and this time it's aimed squarely at non-engineers.

Nah, this isn't a developer tool update. The new onboarding flow prompts users to connect email, calendar, and Google Drive. New browser and computer use features let users annotate web pages and generate artifacts directly inside Codex. A new SSH option allows Codex to connect to a remote desktop and operate local applications from afar.

OpenAI is explicitly positioning Codex as a super app tier alongside ChatGPT and its Atlas browser, targeting knowledge workers broadly.

Here's the honest comparison. Anthropic's desktop has Claude Chat, Cowork, and Claude Code running as three separate modes that don't share memory or context. Codex maintains consistent memory across everything. Side-by-side computer use testing shows Codex running five to 15 times faster than Claude Code.

What it means: The desktop AI agent race is moving fast, and the gap between Codex and Anthropic's current offering on computer use is real and measurable. If your team hasn't revisited Codex in the last 60 days, it's worth another look.

Just one day after OpenAI and Microsoft removed their exclusivity agreement, OpenAI announced a $38 billion multiyear strategic partnership with Amazon Web Services.

The timing was not a coincidence.

AWS will provide OpenAI with Amazon EC2 UltraServers connecting hundreds of thousands of advanced NVIDIA GPUs, with the ability to scale to tens of millions of CPUs as needs grow. All planned capacity is targeted for deployment by end of 2026. The infrastructure supports both real-time ChatGPT responses and the training of next-generation models.

In the ongoing race between OpenAI and Anthropic, compute and uptime have been the decisive battleground. Anthropic's uptime has consistently fallen short of the three to four nines of reliability enterprise customers expect. OpenAI is meaningfully closer, and this deal extends that advantage.

What it means: More compute means faster model releases and better infrastructure at scale. Yuuuuup — Anthropic is still playing catch-up on uptime while OpenAI locks in massive infrastructure deals. That gap compounds over time, and it's one worth watching closely.