Anthropic surpasses OpenAI despite military ban

Anthropic has officially crossed a thirty billion dollar annualized revenue run rate, officially surpassing OpenAI in revenue generation.

Yeah, the velocity of this financial growth is just um genuinely difficult to comprehend.

Yeah.

Especially considering this company was generating just a fraction of that amount in the very recent past. And to put this in perspective for you. Secondary markets are currently placing an implied valuation of one trillion dollars.

A trillion dollars.

Exactly. And they are achieving all of this while simultaneously fighting a ban from the United States military.

Which is just crazy. I mean, how did a research lab that didn't even exist a few years ago manage to overtake the most recognizable name in artificial intelligence, all while becoming the center of this huge geopolitical standoff?

We are basically looking at a complete realignment of the artificial intelligence industry here. The focus has heavily shifted toward their enterprise strategy. Unprecedented physical infrastructure build out required to maintain it.

Right, right.

And of course, that resulting friction with the Pentagon over who actually controls the deployment.

technology. So we have to look at the vertical climb of anthropics revenue to really understand the mechanics at play here. The annualized recurring revenue went from$1 billion to$9 billion to$14 billion and recently hit that$30 billion mark.

Wild.

It is. And you compare that to OpenAI's reported twenty five billion dollars and the contrast and trajectory becomes incredibly sharp.

Wait, back up. How exactly are these two companies counting their money?

Good question.

Cause thirty billion dollars is a massive number, but accounting practices in the cloud computing sector can vary wildly, right? We really need to be precise about what that number is.

Yeah, that is a crucial distinction. The accounting differences explain a significant part of how these revenues are calculated. Okay. Anthropic reports gross annualized recurring revenue, or ARR. That means they count the full cloud sale first. And then they pay the cloud provider their cut. But OpenAI reports net ARR, meaning they only count the revenue after the partner, like Microsoft, takes their cut. Oh, I see.

We can extend that point by looking at their fundamental business models actually, which sit at opposite ends of the spectrum. Yeah. OpenAI relies heavily on consumer subscription. They have hundreds of millions of weekly active users, but those individual users are paying a small monthly fee.

Like twenty bucks or whatever.

Exactly. And the conversion rate from a free user to a paid user is remarkably low, sitting around five.

Why hold just five percent?

Yeah.

Huge difference.

It really is. The metric that highlights the difference in these strategies is revenue per year. Anthropic pulls in roughly two hundred eleven dollars per month of the year.

Oh wow.

Yeah, compared to open AI's twenty five dollars per weekly.

You can really see the success of this enterprise focus in the specific deployment of products like uh Claude Code and Claude Co-Work.

Oh definitely.

Code reached a two and a half billion dollar run rate in a matter of months. Businesses are not just using this to draft emails, you know, they're using it to automate incredibly complex coding tasks within their internal environment. We are looking at a scenario where eight of the Fortune ten companies use Claude.

Eight of the top ten, that's wild.

Yeah, and four percent of all public GitHub commits globally are authored by Claude Coch.

Wait, four percent of all of them?

All of them globally. And to understand the gravity of that for you, GitHub is the repository where the vast majority of the world's open source software is stored and updated. Of all changes to that global code base are being written by an autonomous system, you are looking at a massive footprint in the foundational architecture of modern software.

Well, this completely limits the commercial ceiling for consumer chat box.

How so?

Consumer applications suffer from really high turn rates. If an individual gets bored or uh finds a cheaper alternative, they cancel their twenty dollar subscription instantly.

For sure. Very easy to just cancel.

Right. But what Anthropics model opens up is a highly lucrative, sticky market for autonomous enterprise workflow integrations.

Sticky is the key word.

Exactly. When a multinational corporation integrates an AI model into their proprietary code base and daily operations, ripping that infrastructure out and replacing it takes years.

Yeah, nobody wants to do that.

The revenue basically becomes locked in.

It's the difference between selling a single ticket to a million tourists versus securing an exclusive contract to supply the engine parts for a massive global airline.

That's a great way to put it.

The airline relies on you to stay in the sky. If your product works, the switching costs are simply too high for them to ever leave.

The secondary market for these shares reflects exactly that reality too.

Oh yeah. Let's talk about that.

If we look at the official Series G funding Anthropic raised thirty billion dollars at a three hundred and eighty billion dollar post money valuation.

Which is already a huge number.

Right. This round was backed by massive sovereign wealth and institutional entities like uh GIC, CO2, MGX, and DEShop.

But you have to contract that official valuation with what is actually happening on the secondary market. Yeah. On trading platforms like Forge Global and Jupiter, pre IPO instruments imply a one trillion dollar valuation.

A trillion.

Some specific bids are even reaching one point w five trillion dollars.

That is just hard to wrap your head around.

Even on a more conservative trading platform like Hive, the valuation sits at eight hundred and fifty one billion dollars.

The investor behavior surrounding these shares is completely franchise. There is two billion dollars in buy side demand chasing shares with almost no sellers willing to part with their equity.

Nobody wants to sell.

Exactly. You have investors literally offering a fourteen acre estate in exchange for share.

Entire estate. That's wild.

Institutional brokers are also reflecting this disparity in demand. Goldman Sachs is charging a fifteen to twenty percent carry fee for anthropic allocations, while simultaneously waiving fees entirely just to push open AI charges.

Okay. So let's just reset the pace for a second here. Yeah. When people who invest for a living are willing to trade their houses for a piece of paper, the market has stopped looking at current profits and is betting entirely on total global dominance.

I mean I actually completely disagree with applying that logic without a serious degree of caution.

Why?

A trillion dollar valuation for a company that makes thirty billion dollars in revenue is mathematically.

Well the multiples are high, sure.

Multiples are stretched far beyond historical precedent for any software or infrastructure company in You're basically assuming perfect execution for the next day.

But the buyers are not pricing in thirty billion dollars. True. They're pricing in the company, potentially hitting one hundred billion dollars revenue very soon. They are looking at the trajectory, the enterprise lock in we just discussed and the fact that the actual cost of computation is scaling aggressively alongside the revenue. They view the current revenue as merely the opening app.

Well, this changes the entire dynamic of capital allocation regardless. Yeah. It opens up a new reality for private markets, where secondary trading sets the true price of technology companies long before they ever go public.

Exactly.

This fundamentally limits the influence of traditional initial public offerings. All the price discovery and all the massive valuation leaps are happening behind closed doors.

Right. Accessible only to accredited investors and sovereign wealth funds.

Yeah, completely locking out the retail investor from participating in the largest wealth creation event.

To maintain that astronomical level of growth, the physical requirements are just We have to look at the massive infrastructure investments from hyperscalers.

Yeah.

Yeah. Google committed up to forty billion dollars, providing five gigawatts of TPU capacity. And Amazon committed twenty five billion dollars. Providing five gigawatts of transity.

These deals are highly reciprocal in nature, though.

Yeah, they go both ways.

Anthropic secured the funding, but they simultaneously committed to spending more than$100 billion over the next decade on Amazon Web Services. Right. The sources point to their collaboration on Project Rainier.

Project Rain.

Yeah, which is a massive AI compute cluster featuring half a million chips working in tandem.

Hold on, wait back up. We need to pause and define what five gigawatts actually means in the real world.

Good idea.

That metric gets thrown around constantly in these infrastructure reports, but it is incredibly difficult to conceptualize without a direct comparison.

Yeah, so a single gigawatt can power hundreds of thousands of.

Okay.

When we talk about securing five gigawatts from Google and another five gigawatts from Amazon, the energy required to train these models is no longer comparable to running a traditional server farm.

It's way beyond that.

It is the equivalent of acquiring the power grid of a midsized country. You are dealing with massive land acquisition, dedicated electrical substations, and cooling infrastructure on an industrial scale. Wow. You're essentially building a small city purely to house processors.

And this fundamentally limits the entire field of competition.

Absolutely.

It limits the AI race to only those entities capable of securing multi gigawatt power agreements, effectively locking out any new startups from competing at the frontier level.

Yeah, it's a huge barrier to entry.

If you have a brilliant algorithm, it does not matter if you cannot physically plug it into a power source large enough to train it. The era of a few engineers building a frontier model in a garage is completely over.

And to manage the integration of these massive models into corporate environments, Anthropic introduced the Model Context Protocol, or MCP. They open sourced this protocol, and it was later donated to the Linux Foundation. The explicitly stated goal was to create a standard way for AI agents to connect to external data sources. Bridge. Exactly. If you want an AI to read your customer relationship management system or pull data from your secure file systems, MCP provides the standardized pathway.

We have highly detailed survey data from Zuplo that illustrates exactly how fast this protocol is spreading.

The adoption is crazy.

It is. The model context protocol has reached ninety-seven million installs.

Yeah.

Seventy percent of users configure between two and seven MCP servers for their AI environment. And what is fascinating mechanically is that fifty nine percent are using streamable HTTP for transport, rather than the default STDIO.

Okay, wait, what's the difference there?

While STDIO is typically used for local machine to machine communication, while streamable HTTP allows for continuous data streaming over networks. The heavy use of streamable HTTP indicates highly complex remote server deployments rather than simple local testing.

But the survey also reveals significant hurdles accompanying this rapid adoption.

Oh yeah, the security issue.

Fifty percent of builders cite security and access control as their absolute biggest challenge.

That makes sense.

Even more concerning from a structural standpoint, twenty four percent of M C P servers have no authentication at all.

Zero.

Zero author. To manage this complexity and attempt to secure these endpoints, Thirty percent of developers are hosting these servers on API gateways. Right. An API gateway acts as a traffic cop, basically verifying requests before they reach the sensitive data, but the sheer volume of unauthenticated servers remains a glaring vulnerability.

Now Anthropic builds this as an open standard to help everyone in the industry communicate smoothly. They present it as a rising tide, lifting all books.

Well, that is the public messaging, but I view it as a highly strategic defensive maneuver.

How so?

So by establishing the protocol that everyone uses, Anthropic ensures its ecosystem becomes the permanent infrastructure layer of the internet. Think of it as creating the USB C for AI.

Oh, I see.

Even if a competitor builds a slightly better model next year, all the corporate data pipes, all the API gateways, and all the enterprise permissions are already perfectly formatted for Anthropic's protocol. The switching cost becomes prohibited.

Right, this opens up a universal standard for AI tool integration, allowing enterprise developers to connect incredibly secure databases to large language models seamlessly. However, it introduces severe access control risks. If the AI agent has the keys to your entire corporate database to perform its tasks, and the authentication layer is weak or nonexistent, the entire corporate network becomes vulnerable to external exploitation.

And that exact vulnerability leads directly into the development of mythos.

Mr.

Mythos is Anthropic's highly restricted cybersecurity model. Its capabilities extend far beyond standard conversational AI. The sources detail how it can complete a 32-step cyberattack simulation entirely without human intervention.

Wow, thirty two steps autonomously.

Yeah, that is an act that normally takes professional, highly trained human hackers days to execute as they manually test endpoints, look for misconfigurations, and attempt to escalate privileges. Right. Mythos automates that entire lateral movement process. Mozilla utilized Mythos in a secure environment and found two hundred and seventy one zero day vulnerabilities in the firefox.

Two hundred and seventy one.

And a zero day vulnerability is a software flaw unknown to the vendor, meaning there is zero time to fix it before it can be exploited. Finding two hundred and seventy one of them autonomously is unfortunate.

The power of this specific model led directly to the Project Glasswing Initiative. This program heavily restricted Mythos access to a very select group of organizations, primarily focusing on critical infrastructure partners like Apple and JP Morgan. Okay. The objective was to allow these entities to continuously scan their own code bases and patch their vulnerabilities long before state sponsored adversaries could exploit them.

Then the breach occurred.

Ah yes.

The breach Demonstrated that unauthorized users successfully gained access to the Mythos preview, and the method of access was surprisingly rudimentary.

It wasn't some complex hack.

No, not at all. A forum user working for a third party contractor combined their authorized basic credentials with data obtained from an unrelated data breach at the start up merchant.

Oh wow.

By cross-referencing this information, they manage to locate and access the restricted model.

Wait, hold on, back up. How was a model explicitly designed to secure the most critical financial and technological systems in the world compromised through a basic supply chain vendor flaw?

Exactly.

Let's just state this plainly. The most advanced AI hacker in the world was accessed because a human contractor left the digital door unlocked.

Yeah, that's exactly what happened.

It wasn't a sophisticated algorithmic jailbreak. Yeah. It was basic human error and poor access control within the extended vendor network.

This completely changes the paradigm of cybersecurity defense. It opens up the possibility of automated zero day detection, where vulnerabilities are found and patched instantly. shifting the advantage back to the defenders. Right. However, it severely limits the confidence that private companies can safely hoard frontier models. If a third party contractor can leak access through simple credential reuse, hostile nation states certainly possess the capability to do the same.

Which brings us to the government side of things. Before we get into this, we must state clearly for you, the listener, that we are impartially reporting the facts of the dispute between the United States government and the company. Strictly as outlined in our sources, and we are not endorsing either political viewpoint.

Yes, absolutely.

Right.

Right.

Anthropic secured a two hundred million dollar contract to deploy Claude on classified government networks. However, they maintained incredibly strict acceptable use policies

Which caused some friction.

Oh big time. These policies explicitly prohibited the use of their technology for fully autonomous weapon systems and for mass domestic surveillance operations. The Department of Defense rejected these specific limitations and demanded completely unrestricted access for all lawful purposes.

And after those contract negotiations failed to reach an agreement, the fallout was immediate and severe.

It really was.

Defense Secretary Pete Hegzef officially designated Anthropica supply chain. The president subsequently directed federal agencies to begin a comprehensive phase out of anthropic technology across the entirety of the federal government.

The legal mechanisms utilized to execute this phase out are highly specific actually.

actually. Yeah.

The government invoked a FSEASA order, which stands for the Federal Acquisition Supply Chain Security Act, alongside ten USC section thirty two fifty two. Right. A FASCSA order functions essentially as an emergency kill switch for federal procurement. Right. These mechanisms force government contractors across the country to immediately halt new deployments. inventory their existing use of Claude, and actively seek alternative vendors.

They filed lawsuits in both California and the D C Circuit. In the California case, Judge Rita F. Lynn granted a preliminary injunction against the government. Okay. She ruled that the government's action constituted classic illegal First Amendment retaliation. Wow. Yeah, noting that internal government records revealed the supply chain risk designation was based largely on the company interacting in a hostile manner through the press rather than any actual technological vulnerability.

That's fascinating.

But in the DC circuit, the court refused to lift the designation. They prioritized the military's immediate operational needs during an ongoing conflict, specifically citing the requirements of Operation Epic Fury.

I have to say I question the internal coherence of Anthropic's ethical stance in this situation. at autonomous targeting. When the technology is already integrated that deeply into the kill chain of military operations, drawing an arbitrary line at the final trigger pole seems highly contradictory.

I see your point, but a private enterprise has the right and honestly the fundamental obligation to enforce its own terms of service.

Even against the government.

They built the underlying infrastructure, and they should be able to dictate the absolute boundaries of its use, even when negotiating against a global superpower. If the creator's belief fully autonomous targeting crosses an unbreachable moral line, they must enforce that boundary, regardless of the financial cost.

Well, this completely rewrites the rules of engagement between Silicon Valley and the Pentagon.

It really does.

It opens up incredibly lucrative long term defense contracts for artificial intelligence companies willing to waive their ethical restrictions while simultaneously punishing those who attempt to dictate operational terms to the military apparatus.

We are seeing a rapidly developing resolution to the standock, however.

Oh the White House pivot.

Yeah. According to industry sources, the White House is currently drafting plans to completely bypass the supply chain risk design. Specifically to permit federal use of the Mythos AI model we discussed.

The reasoning behind this pivot is purely pragmatic.

Totally.

Government cybersecurity agencies and intelligence branches quickly recognized that they simply could not afford to be locked out of the most advanced defensive cyber tool available on the market.

Exactly.

When your adversaries are rapidly deploying similar autonomous capabilities, handicapping your own defensive infrastructure over a standard contract dispute is a completely untenable strategic position.

This severely limits the government's ability to successfully blacklist critical technology.

It really does.

When the state's own defensive capabilities rely entirely on that private infrastructure to function, the threat of a ban loses all leverage. You cannot effectively ban the company that holds the digital keys to your own national security.

The trajectory of Anthropic proves the real value of artificial intelligence lies directly in enterprise infrastructure and automated coding. Creating an entity so vital to the global economy that it forces governments to rewrite their own rules of procurement.

If a private company controls the automated intelligence that patches our critical infrastructure, writes our corporate software, and negotiates terms of engagement with the military, who really holds the ultimate authority in the modern world, the government or the algorithm?

If you're not subscribed yet, take a second and hit follow on whatever app you're using. It helps us keep making this. We appreciate you being here. Also check out our YouTube channel for more business and tech updates. There's a link in the description.