April 29 Roundup: OpenAI lands on AWS Bedrock, the Microsoft AGI clause dies, Musk takes the stand, and AI policy goes from principle to procurement

1. OpenAI’s frontier models and Codex go live on Amazon Bedrock — 24 hours after Microsoft exclusivity ends

The single biggest enterprise AI development of the cycle is also the most concrete: OpenAI’s latest models and its Codex coding agent are now available on Amazon Bedrock, in limited preview, alongside a new service for building and deploying agents on AWS. Reuters reported the launch from Amazon’s San Francisco event on April 28, with AWS CEO Matt Garman framing it as something customers had been waiting on for years.

“This is what our customers have been asking for a really long time. Their production applications run in AWS. Their data is AWS,” Garman said, per Reuters. Sam Altman, appearing by pre-recorded video while preparing to testify in his trial, added: “The opportunity ahead of us is enormous, and the most exciting part is that this is not something in the future, it’s starting right now.”

Reuters notes that this builds on Amazon’s previously disclosed $50B investment in OpenAI, OpenAI’s commitment to spend $100B on AWS over eight years, and a separate two-gigawatt deal for OpenAI to use Amazon’s in-house Trainium AI chips. The Verge’s Richard Lawler captured the strategic flavor bluntly, saying it “seems clear that ‘OpenAI’s focus is going to be on AWS,’ particularly with an eye toward the new Amazon Bedrock Managed Agents setup.”

For OpenAI, this matters for two reasons. First, Codex now has a frictionless path into the millions of enterprises that already standardize on AWS — and OpenAI says Codex already has more than 4 million weekly active users. Second, Amazon Bedrock’s position as a multi-model marketplace means OpenAI is now competing for attention against Anthropic on Anthropic’s most strategic distribution surface.

2. The Microsoft–OpenAI AGI clause is officially dead — and the partnership is now non-exclusive

The day before the AWS news, OpenAI and Microsoft formally renegotiated the contract that had defined the AI era’s most consequential alliance. CNBC reported that revenue share payments from OpenAI to Microsoft will continue at 20% through 2030 but “subject to a total cap,” independent of any future AGI declaration. Microsoft’s license to OpenAI IP runs through 2032 but is no longer exclusive. Microsoft will also stop paying a revenue share back to OpenAI for Azure-hosted usage.

The Verge’s Hayden Field summarized the symbolic shift: “there’s no independent panel, there’s no if-this-then-that language for if or when AGI is declared, and OpenAI may never have to actually announce if it reaches that milestone. Microsoft’s license to OpenAI’s models and products that it holds through 2032 is now non-exclusive.”

OpenAI’s own statement called the change a move “grounded in flexibility, certainty, and a focus on delivering the benefits of AI broadly,” adding that OpenAI can now serve “all of its products” to customers across any provider. CNBC notes that Microsoft has invested more than $13 billion in OpenAI since 2019, but the relationship had visibly strained as both companies moved onto each other’s turf — most recently, OpenAI’s revenue chief Denise Dresser told staff the partnership had “limited our ability to meet enterprises where they are.”

The death of the AGI clause is the part operators should not skim. For years it was the trip-wire that would have realigned IP rights and revenue mechanics if either side ever declared artificial general intelligence. Removing it tells you something important about how both companies actually think about the next 24 months: this is now a market-share fight, not a milestone race.

3. Musk v. Altman: a trial about charity law is really a trial about who controls the AI economy

Elon Musk took the witness stand in San Francisco on April 28, opening Musk v. Altman with the line that he’s being asked to defend charitable giving itself. The New York Times noted that the first day of testimony produced “two notably different tales of how OpenAI evolved from a nonprofit artificial intelligence lab into one of the most influential tech companies in the world.” Reuters added that Musk testified the company was his idea “before executives looted it.” The Verge’s Elizabeth Lopatto, watching the proceedings closely, described Musk’s opening as “funereal” and reported him telling the jury, “if the jury finds in favor of Sam Altman, Greg Brockman, OpenAI, and Microsoft, it will become precedent and give precedent to looting every charity in America.”

The Wall Street Journal reported, via Reuters, that OpenAI “has fallen short of its goals for new users and revenue in recent months, sparking concern among some company leaders over whether it can support its extensive data-center spending.” Oracle and CoreWeave shares dropped on the news as investors recalibrated assumptions about AI infrastructure pull-through.

Musk is set to return to the stand today. The case itself centers on breach of charitable trust and unjust enrichment — Judge Yvonne Gonzalez Rogers dismissed Musk’s fraud claims at his request last week — but the substantive question for the AI industry is whether the courtroom turns OpenAI’s 2019 capped-profit pivot, its 2025 recapitalization, and its 2026 IPO trajectory into a single legal narrative about whether the public-benefit story still holds.

4. Google’s up-to-$40B Anthropic deal turns compute into the real moat

Google’s commitment to invest up to $40B in Anthropic — $10B now at a $350B valuation, with another $30B tied to performance milestones — kept reshaping enterprise AI economics this week. TechCrunch’s Rebecca Bellan reported that the new deal expands an earlier Google–Broadcom–Anthropic partnership by adding 5 gigawatts of fresh capacity over the next five years, with room to scale further.

“The AI race is increasingly defined by access to the compute needed to train and deploy these systems,” Bellan wrote. Anthropic, she noted, has been in “a scramble of its own,” recently striking a CoreWeave capacity deal and securing an additional $5B from Amazon as part of an arrangement under which Anthropic is expected to spend up to $100B for around 5 gigawatts of compute capacity over time.

The Motley Fool flagged a quieter signal in the same week: Anthropic made a tender offer to long-tenured employees at the same $350B valuation, while outside investors were reportedly eager to back the company at $800B or more. Bloomberg has reported the company is also weighing an IPO as soon as October. Translation: the public market is now pricing Anthropic at numbers that would have seemed deranged six months ago — and it is doing so primarily because the company has secured the compute to keep growing.

This is a structural change in how AI companies are valued. The asset on the balance sheet that matters most is no longer the model weights. It is the contracted access to gigawatts of training and inference capacity, locked in for years, with strategic supplier relationships across Google TPUs, Amazon Trainium, and CoreWeave-class GPU fleets.

5. A bipartisan U.S. AI bill goes after deepfakes — and quietly reshapes whistleblower protections

The U.S. policy track moved from rhetoric to text. CNBC’s Emily Wilkins broke a new bipartisan AI bill from Rep. Ted Lieu (D-CA), top Democrat on the House AI Task Force, with backing from Republican task force chair Jay Obernolte. The bill stiffens penalties for distributing deepfakes and non-consensual AI-generated imagery, protects whistleblowers reporting AI safety risks, mandates U.S. participation in international AI standards bodies, and creates a federal prize competition for AI R&D.

“It is not designed to be controversial,” Lieu told CNBC. “It is based on bipartisan legislation that other members have introduced, as well as the recommendations of the bipartisan House AI Task Force. So we’re trying to do something this term right now with this bill.”

Lieu’s draft deliberately sidesteps the bigger preemption fight. Cooley’s legal analysts noted that most existing state AI laws extend existing consumer protection principles into AI, “making companies liable where AI-driven conduct would otherwise violate deceptive or unfair practices laws.” And Reuters reported today that EU countries and lawmakers failed overnight to reach a deal on watered-down AI rules, leaving Europe’s AI Act implementation in the same uncomfortable middle ground it has occupied for months.

The whistleblower protection clause is the part operators should pay attention to. It expands the surface area for AI lab employees and contractors to report concerns externally without retaliation. Combined with state-level deceptive practices liability and a federal framework that explicitly preserves consumer protection enforcement, this is how AI governance is actually moving: not as one giant Act, but as a steady accumulation of liability surfaces that turn responsible deployment into an enterprise compliance requirement.

6. Google’s split TPUs and the new shape of inference

Google Cloud’s Next ’26 announcement that its eighth-generation TPUs will split into a TPU 8t (training) and TPU 8i (inference) keeps reverberating because it tells you where margin pressure is going. TechCrunch’s Julie Bort reported Google’s claim of “up to 3x faster AI model training, 80% better performance per dollar, and the ability to get 1 million+ TPUs to work together in a single cluster.”

“Google’s chips are not a full frontal assault on Nvidia’s future, at least not yet,” Bort wrote. “Like the other giant cloud providers, including Microsoft and Amazon, Google is using these chips to supplement the Nvidia-based systems it offers in its infrastructure. It is not flat-out replacing Nvidia.”

The detail that matters operationally: Google is engineering its inference-tuned TPU 8i specifically for the workloads agentic AI generates — long-running chains, tool calls, retrieval, and multimodal responses. Reuters separately flagged that Big Tech investors are now sizing up $600B in annual AI capex as the new normal. That capex doesn’t pay back unless inference unit economics keep falling.

7. The thought leaders are converging: abundance, but only if you build the systems

Peter Diamandis spent the week amplifying a familiar but increasingly load-bearing thesis. Speaking with Raoul Pal and Salim Ismail, Diamandis argued that institutions are struggling to adapt fast enough, that AI will generate more content than the entire prior history of humanity by 2028, and that decentralization is the path to genuine abundance. On Joe Polish’s show, he and Steven Kotler unveiled their new book We Are as Gods, a survival guide for what they call the age of AI, longevity, genetics, and robotics. And Business Insider walked through Diamandis’s breakdown of Elon Musk’s argument that an AI-driven future will mean universal income and a society where “you don’t need to save for retirement.”

Jason Calacanis’s This Week in AI is keeping its operator lens on the same theme from a different angle: products win, but only the products that own data, distribution, and execution loops will compound. The trial in San Francisco is a reminder of how fragile origin stories are. The Bedrock launch is a reminder of how quickly distribution can change. The Anthropic compute deal is a reminder of how capital-intensive durable advantage really is.

Put together, these voices are converging on a single operational truth: in 2026, optimism about AI’s potential has to be paired with discipline about systems. Smart organizations are not just choosing models; they’re designing the orchestration, governance, capacity, and accountability around those models so that the optimism actually turns into durable outcomes.

Sources: Reuters — OpenAI’s latest AI models, Codex now available on Amazon Bedrock, The Verge — OpenAI’s new AWS deal, CNBC — OpenAI shakes up Microsoft partnership, capping revenue share, The Verge — Microsoft and OpenAI’s famed AGI agreement is dead, Reuters — Musk takes the stand in OpenAI trial, New York Times — Two very different tales of OpenAI’s early years, Reuters — OpenAI falls short of revenue and user targets, WSJ reports, TechCrunch — Google to invest up to $40B in Anthropic, TechCrunch — Google Cloud’s new TPUs, CNBC — Bipartisan AI bill targets deepfakes and whistleblowers, Peter Diamandis with Raoul Pal, and Jason Calacanis’s This Week in AI.