The Week in AI: April 27 – May 3, 2026 (Week 18)

News of the week: April 27 – May 3, 2026 (Week 18)

The week was dominated by a reworked relationship between OpenAI and Microsoft, which put OpenAI models on Amazon’s cloud the very next day. Big Tech companies reported quarterly results and raised their data center spending yet again, with combined outlays this year heading toward hundreds of billions of dollars. The US Department of Defense expanded its work with Google and other companies on classified networks, while Anthropic remains on the sidelines. Elon Musk’s trial against OpenAI opened in Oakland, and in his testimony Musk admitted that xAI partly trained Grok on OpenAI models. According to sources, Anthropic is weighing a funding round at a valuation of more than $900 billion.

Business and Partnerships

OpenAI and Microsoft rewrite their contract, ending Azure exclusivity

On Monday, April 27, OpenAI and Microsoft announced a revised partnership agreement. OpenAI may now offer its products on any cloud, including Amazon Web Services and Google Cloud (TechCrunch).

Microsoft remains the primary partner, and new OpenAI products are to launch first on Azure. Microsoft’s license to OpenAI’s intellectual property runs until 2032 but is no longer exclusive. The agreement settled the dispute over Amazon‘s $50 billion investment in OpenAI from February.

Two days later, on the earnings call, Microsoft CEO Satya Nadella said he is ready to take full advantage of the new deal (TechCrunch).

OpenAI models on AWS the very next day

As early as April 28, OpenAI announced that its models, the coding agent Codex, and the new Amazon Bedrock Managed Agents service are available through AWS (OpenAI, CNBC).

Bedrock is Amazon’s platform that lets companies use language models from various vendors. Managed Agents are meant to let them build their own AI agents, programs that carry out tasks on their own and remember previous interactions. Codex on Bedrock is currently in limited preview.

Quarterly results: cloud is growing, spending even faster

On Wednesday, April 29, the four biggest AI investors reported their results. The common thread was rapid growth in cloud services and another increase in data center investment, driven in part by rising memory prices.

Microsoft: revenue of $82.9 billion, with the Azure cloud growing 40%. The company expects capital expenditures of $190 billion for 2026 (CNBC). M365 Copilot has more than 20 million paid business seats (TechCrunch).
Alphabet: Google Cloud topped $20 billion in quarterly revenue for the first time, though the company said growth was held back by insufficient capacity (TechCrunch). Its contracted backlog rose to $462 billion.
Google also announced that it will start delivering its TPU AI chips to select customers directly in their own data centers (Google). Until now, they were available only as a service in its cloud.
Amazon: AWS grew 28% to $37.6 billion, its fastest pace in nearly four years. The company plans to spend around $200 billion this year, almost all of it on AI infrastructure (CNBC).
Meta: revenue rose 33% to $56.3 billion. It raised its spending forecast for this year to $125 to $145 billion, and its shares fell after the release (Fortune).
Apple said on April 30 that it was surprised by demand for Mac computers driven by AI use (TechCrunch).

Anthropic weighs a valuation of more than $900 billion

According to TechCrunch sources on April 29, Anthropic, the maker of the Claude models, is in talks about a new funding round of roughly $50 billion. The round would value the company at approximately $900 billion (TechCrunch).

A day later came a report that the round could close within two weeks (TechCrunch). So far these are talks, not a completed investment.

Acquisitions and Investments

China blocks the Manus takeover (April 27): China’s National Development and Reform Commission prohibited Meta‘s $2 billion acquisition of Manus, a startup developing AI agents. The parties must unwind the transaction (TechCrunch).
David Silver raises $1.1 billion (April 27): the former DeepMind researcher has founded the British lab Ineffable Intelligence, which aims to develop AI that learns without human data. The lab is valued at $5.1 billion (TechCrunch).
Legora at $5.6 billion (April 30): the legal AI startup reached a new valuation, heating up its rivalry with competitor Harvey (TechCrunch).
Meta buys into robotics (May 1): the company acquired Assured Robot Intelligence, a startup focused on humanoid robots. The team will join Meta Superintelligence Labs (TechCrunch).
A phone from OpenAI? (April 27): analyst Ming-Chi Kuo said that OpenAI is developing a smartphone in which AI agents would replace apps. The chips are to come from Qualcomm and MediaTek; the companies have not confirmed this (TechCrunch).

New Models and Products

Nvidia releases the open model Nemotron 3 Nano Omni

On April 28, Nvidia introduced Nemotron 3 Nano Omni, a model that handles text, images, audio, and video in a single unit (NVIDIA Blog). It is an open model: anyone can download its weights and run it themselves.

It uses a mixture of experts architecture: of its 30 billion parameters, only about 3 billion are active in each computation, which saves compute. According to Nvidia, it is designed mainly for AI agents, for example for reading documents or operating a computer.

Other Product News

Safer ChatGPT accounts (April 30): OpenAI launched an optional protection called Advanced Account Security and is releasing two physical security keys with Yubico (TechCrunch).
Gemini in cars (April 30): Google began rolling out the Gemini assistant to vehicles with Google built-in, for now in the US and in English. General Motors will bring it to roughly 4 million vehicles (TechCrunch).
AI search on YouTube (April 28): YouTube is testing a search feature that offers guided answers instead of a list of videos (TechCrunch).
Business AI at Meta (April 30): Meta‘s business AI tools now facilitate about 10 million conversations a week, up from 1 million at the start of the year (TechCrunch).
A wallet for agents (April 30): Stripe updated its Link digital wallet so that autonomous AI agents can use it to make purchases too (TechCrunch).
A developer tool (April 29): Anthropic released the open Claude API skill, which gives Claude up-to-date reference material for programming against its API (Claude Platform).

Military and Security

The Pentagon expands AI on classified networks

On April 28, it emerged that Google has expanded the Pentagon‘s access to its AI models to include work with classified information (TechCrunch). The department may use them for any lawful purpose.

The department’s AI chief, Cameron Stanley, confirmed the collaboration and added that relying on a single vendor “is never a good thing” (CNBC). Anthropic had earlier rejected similar terms, and the department put it on a blacklist.

The day before, hundreds of Google employees wrote a letter urging CEO Sundar Pichai to turn down classified military work (The Washington Post).

On May 1, the Department of Defense signed further contracts with Nvidia, Microsoft, AWS, and Reflection AI to deploy AI at the highest classification levels, IL6 and IL7 (TechCrunch). The military AI platform GenAI.mil has already been used by more than 1.3 million department employees.

OpenAI restricts access to its cyber model

On April 30, Sam Altman announced that in the coming days the GPT-5.5-Cyber model will be available only to vetted defenders of critical infrastructure (TechCrunch). It is a model specialized in cybersecurity that can find vulnerabilities in software.

In doing so, OpenAI is taking a similar approach to Anthropic with its Mythos model, which OpenAI had previously criticized Anthropic for restricting. According to the company, the Trusted Access for Cyber program is used by thousands of verified security professionals.

Courts and Regulation

The Musk v. OpenAI trial begins

On April 27 in Oakland, jury selection kicked off the lawsuit brought by Elon Musk against OpenAI, Sam Altman, and Greg Brockman. Musk claims the leadership betrayed the company’s original nonprofit mission and is seeking, among other things, Altman’s removal (NPR).

Musk testified from April 28 to 30 and accused OpenAI’s leaders of “looting the nonprofit” (NPR). During cross-examination he repeatedly clashed with OpenAI’s attorney (CNBC).

Asked whether his company xAI had trained Grok through distillation of OpenAI models, he answered “partly” (TechCrunch). Distillation means that a smaller model learns from the outputs of another, more capable model.

Senate committee approves the GUARD Act

On April 30, the Judiciary Committee of the US Senate unanimously approved the GUARD Act bill by a vote of 22 to 0 (Senator Hawley). The bill’s sponsors are Senators Josh Hawley and Richard Blumenthal.

The law would ban offering AI companions, chatbots that simulate a friendly or emotional relationship, to minors and would introduce age verification. Chatbots would also have to tell all users that they are not human. The bill now awaits a vote by the full Senate (Roll Call).

Oscars only for human actors and screenwriters

On May 1, the Academy of Motion Picture Arts and Sciences published new rules: only acting performances demonstrably given by humans with their own consent are eligible for an Oscar (TechCrunch). Screenplays must be written by a human, and the Academy may request details about the use of AI in a film.