Anthropic Takes the Pentagon to Court While AI Giants Race to Control Physical Infrastructure

Wednesday, 25 March 2026. The AI world woke up to lawsuits, billion-dollar bets on physical infrastructure, and a Nvidia CEO casually declaring that artificial general intelligence is, well, now. Here’s everything that matters today.

Anthropic sues the Department of Defense, a first for any AI lab

Anthropic filed a lawsuit in a California district court this week against the US government, challenging its official designation as a supply-chain risk by the Department of Defense. This is the first time a major AI lab has taken legal action against a national security blacklisting.

The background: over the past few weeks, Anthropic found itself in an escalating dispute with the Pentagon over what it would and wouldn’t allow its AI to do. Anthropic drew red lines on mass domestic surveillance and fully autonomous weapons systems. The Trump administration responded by designating Anthropic a supply-chain risk, a label typically reserved for foreign cybersecurity threats. All government agencies were ordered to drop its technology within six months.

The lawsuit argues that the government retaliated against Anthropic specifically because the company publicly stated its views on AI safety and refused to cross certain ethical lines. Those are First Amendment-protected positions, Anthropic’s lawyers say. The suit also invokes the Fifth Amendment, arguing the government’s actions could “destroy the economic value” of one of the world’s fastest-growing private companies without due process.

The fallout has been real. The General Services Administration terminated its OneGov contract with Anthropic, cutting off Anthropic products from all three branches of the federal government. The Treasury and State departments have reportedly followed suit. Meanwhile, major enterprise clients like Microsoft have publicly stated they’ll continue working with Anthropic, but are walling off any work that touches Pentagon contracts.

The case raises a question the industry hasn’t had to face before: what happens to an AI company’s commercial viability if disagreeing with a presidential administration’s policy preferences gets it classified as a national security threat? Whether or not Anthropic wins in court, the chilling effect on other labs is already visible. No other lab is going to rush to make news about the weapons systems it won’t build.

HUMAIN drops $3 billion into xAI as Elon’s empire consolidates

Saudi Arabia’s HUMAIN, a Public Investment Fund company focused on full-stack AI, has committed $3 billion to xAI’s Series E round. The timing is deliberate: the investment came just ahead of xAI’s acquisition by SpaceX, one of the largest tech mergers on record.

Here’s how it works mechanically: HUMAIN’s xAI stake is being converted into SpaceX shares as part of the merger. So HUMAIN ends up with significant minority equity in a combined xAI-SpaceX entity, a company that would control leading-edge AI models (Grok), satellite infrastructure (Starlink), and rocket manufacturing under one roof. That’s a genuinely unusual concentration of capabilities.

This isn’t HUMAIN’s first move with xAI. Back in November 2025, the two companies announced a partnership to jointly develop over 500 megawatts of AI data center infrastructure and deploy Grok models in Saudi Arabia. The Series E deepens that relationship considerably, and it’s not hard to see where this is heading: HUMAIN wants a stake in the winning AI infrastructure play, and the xAI-SpaceX merger might be it.

Tareq Amin, HUMAIN’s CEO, described the investment as targeting companies where “long-term vision, technical excellence, and execution converge.” The physical infrastructure angle is what stands out: compute, data centers, and now satellites. AI investment in 2026 increasingly means betting on physical things, not just model weights.

Jensen Huang says we’ve already hit AGI, sort of

On a recent episode of the Lex Fridman podcast, Nvidia CEO Jensen Huang said something that predictably went everywhere: “I think we’ve achieved AGI.”

Context matters here. Fridman had defined AGI for the purposes of the question as an AI system that could “essentially do your job”, specifically, start, grow, and run a tech company worth over $1 billion. Huang said under that definition, the answer is yes, it’s now.

But Huang himself partially walked it back in the same breath. He noted that while AI agents are being used for all kinds of things, he even mentioned OpenClaw’s viral spread, most people use them for a few months before moving on. “The odds of 100,000 of those agents building Nvidia is zero percent,” he said.

This is the AGI conversation in 2026 in miniature. The term has been stretched and redefined by so many tech leaders that it’s become nearly meaningless as a benchmark. Huang’s “we’ve achieved it” lands differently once you realize his definition is tied to a specific capability test, not a general philosophical threshold. Still, when the CEO of the world’s most valuable semiconductor company says it on one of the biggest tech podcasts around, it moves markets and generates headlines, which is probably part of the point.

OpenAI eyes Helion Energy in advanced acquisition talks

OpenAI is in advanced talks to acquire Helion Energy, the fusion startup that Sam Altman has personally backed for years. Altman has already stepped down from Helion’s board to avoid the obvious conflict of interest as the discussions accelerate.

The strategic logic is straightforward: AI training and inference are power-hungry at a scale that the existing energy grid struggles to meet reliably. Data centers are competing with each other, and with other sectors, for electricity. A lab that owned its own fusion energy supply wouldn’t be subject to those constraints. It’s vertical integration taken to an almost absurd extreme.

Helion has been working toward commercial fusion for years and has a power purchase agreement with Microsoft from 2023. Whether the technology is production-ready within a timeframe that meaningfully helps OpenAI’s near-term compute needs is an open question, fusion has historically been 20 years away for a very long time. But the signal is clear: the leading AI labs now treat energy infrastructure as a core competitive asset.

This pairs naturally with HUMAIN’s xAI-SpaceX bet. AI’s biggest players are building physical foundations that go well beyond software. Fusion reactors, satellites, data centers in the Gulf. Whether or not any individual bet pays off, the direction of travel is obvious.

Beehiiv launches MCP integration, your newsletter now talks to Claude, ChatGPT, and Gemini

Newsletter platform Beehiiv has launched its Model Context Protocol integration, making it the latest mainstream SaaS product to plug directly into the major AI clients. MCP, the open standard originally developed by Anthropic, lets AI tools like Claude, ChatGPT, and Gemini connect live to external software rather than working from static snapshots of pasted data.

What that means practically: you can ask your AI of choice to pull subscriber analytics, identify high-intent readers, correlate churn spikes with specific email campaigns, or generate a sponsorship media kit, all in real time, from live data, without copying anything into a chat window.

V1 is read-only, currently in early access for paid Beehiiv users. V2 will add write access, enabling things like drafting and saving newsletter content, building automation sequences, or generating landing pages from subscriber data. V2 is where it gets genuinely interesting.

What’s interesting about the Beehiiv MCP launch, beyond the product itself, is what it signals about where SaaS is heading. AI interfaces are becoming the primary way people interact with their tools. The dashboard isn’t going away, but querying your business in plain English against live data is quickly becoming the baseline for any serious platform. Beehiiv is positioning itself as the “operating system” for newsletter businesses. The companies that get MCP integration right will have a real advantage. Whether that includes Beehiiv depends on how fast adoption actually moves.

The bigger picture: consolidation and confrontation

Today’s stories share a thread. xAI-SpaceX and OpenAI-Helion are AI labs betting enormous capital on controlling physical infrastructure: energy, compute, logistics. The Anthropic-Pentagon lawsuit is something different, an AI company pushing back against government power in a way no lab has done before. And the Beehiiv MCP shows, in a small way, how AI tooling is quietly becoming infrastructure, built into the daily operations of businesses across the internet.

Whether Jensen Huang is right about AGI probably matters less than this: the industry’s biggest decisions right now are about power and infrastructure, not benchmarks. The labs that own the power plants and the satellites will have an argument that transcends whatever’s topping the leaderboards.

Frequently asked questions

Why did Anthropic sue the Department of Defense?

Anthropic sued the DoD after the Trump administration officially designated it a supply-chain risk, a move Anthropic says was retaliation for publicly refusing to allow its AI to be used for mass domestic surveillance or fully autonomous weapons. The suit argues the government violated Anthropic’s First and Fifth Amendment rights.

What is HUMAIN and why is it investing in xAI?

HUMAIN is a Saudi AI company backed by the Public Investment Fund, focused on building full-stack AI capabilities. It invested $3 billion in xAI’s Series E round as part of a long-term strategic partnership that includes building AI data center infrastructure in Saudi Arabia. Its xAI stake is being converted into SpaceX shares as part of the xAI-SpaceX merger.

Did Jensen Huang really say AGI has been achieved?

Huang said “I think we’ve achieved AGI” on the Lex Fridman podcast, but his answer was in response to Fridman’s specific definition of AGI as an AI that can run a billion-dollar tech company. Huang subsequently noted significant limitations, suggesting his statement was more rhetorical than a firm technical declaration.

What is the Model Context Protocol (MCP) and why does Beehiiv’s integration matter?

MCP is an open standard, originally developed by Anthropic, that allows AI tools to connect directly to external software and access live data. Beehiiv’s MCP integration means users can query their newsletter business, subscriber analytics, revenue data, campaign performance, through Claude, ChatGPT, or Gemini in real time, without manually exporting or pasting data.

For more daily AI news and analysis, explore FridayAIClub.com.