Anthropic's Skyrocketing Revenue, A Contract Compromise?, Nvidia Earnings

# Anthropic's Skyrocketing Revenue, A Contract Compromise?, Nvidia Earnings

Anthropic's enterprise business is reaching escape velocity, which increases the importance of finding a compromise with the government. Then, agents dramatically increase demand for Nvidia chips, even if they threaten software. 

Good morning,

Apple released [new MacBook Pro models](<https://www.apple.com/newsroom/2026/03/apple-introduces-macbook-pro-with-all-new-m5-pro-and-m5-max/>); the biggest surprise to me is that the RAM pricing stayed the same. [I owe John Gruber a steak dinner](<https://daringfireball.net/linked/2026/03/03/apple-introduces-macbook-pro-models-with-m5-pro-and-m5-max-chips>).

On to the Update:

### Anthropic’s Skyrocketing Revenue

From [Bloomberg](<https://www.bloomberg.com/news/articles/2026-03-03/anthropic-nears-20-billion-revenue-run-rate-amid-pentagon-feud?srnd=homepage-americas>):

> Anthropic PBC is on track to generate annual revenue of almost $20 billion, a projection based on current performance, more than doubling its run rate from late last year — a sign of the AI company’s rapid growth in the lead-up to its recent clash with the Pentagon. The artificial intelligence company recently surpassed $19 billion in run-rate revenue, up from $9 billion at the end of 2025 and roughly $14 billion a few weeks ago, said the people, who spoke on condition of anonymity as the information is not public. The growth in run rate was driven by strong adoption of Anthropic’s AI models and products including its coding tool, Claude Code, the people said.
> 
> Anthropic, now valued at $380 billion, has seen strong momentum this year. Multiple products from the company have gained viral attention for helping to automate more complex tasks, including Claude Code. However, a clash with the Pentagon over AI safeguards now casts doubt over Anthropic’s business.

To put this number in context, OpenAI said [in January](<https://www.reuters.com/business/openai-cfo-says-annualized-revenue-crosses-20-billion-2025-2026-01-19/>) that it passed $20 billion in annualized revenue in 2025; it seems plausible that Anthropic is actually going to surpass OpenAI, at least if this data from Ramp’s lead economist is any indication. First, Anthropic is, in a head-to-head comparison with OpenAI, dominating API spend:

[](<https://x.com/arakharazian/status/2028963743908909065>)

That API spend is primarily due to Anthropic’s lead in coding; what is notable is how Anthropic has also taken the lead in enterprise subscriptions as well:

[](<https://x.com/arakharazian/status/2028963738942927217>)

There are some caveats to this data: it only includes OpenAI and Anthropic, not Azure/Copilot or Google Cloud/Gemini. Ramp is also probably tilted more towards startups and smaller companies than larger enterprises. That noted, it certainly matches the vibes: Anthropic established that coding foothold and has been on an absolute tear over the last six months in particular, in terms of not just its model capabilities, but also Claude Code and the overall sense that it is in the lead, and enterprise CTOs are a smaller and more easily reachable set of customers than consumers at large.

Meanwhile, an AI subscription business model has always been more compelling for the enterprise than for the consumer space, for the same reason that productivity apps always inevitably end up as enterprise products, not consumer ones: enterprises actually value productivity and are willing to pay for it; consumers don’t, and won’t, which is why the business model that makes sense for the consumer space — at least at scale — is advertising.

### A Contract Compromise?

This rapid growth, meanwhile, actually raises the stakes for Anthropic’s standoff with the U.S. government. One of the implications of [what I wrote about yesterday](<https://stratechery.com/2026/technological-scale-and-government-control-paramount-outbids-netflix-for-warner-bros/>) about technology products addressing markets much larger than the government is that technology products don’t _need_ the government; this means that the government can’t really exact that much damage by simply declining to buy a product. That, by extension, means that if the government is determined to control the product in question, it has to use much more coercive means, which raises the specter of much worse outcomes for everyone.

Fortunately, I took Dario Amodei’s comments at yesterday’s Morgan Stanley TMT conference as a positive signal that Anthropic sees the value in deescalating this fight; I couldn’t find a video, but was passed this summary:

> “Anthropic and DoW have much more in common than we have differences”. He said Anthropic has “leaned forward” in working with the intelligence + national security community in deploying models in classified networks and they “really believe in defending America”, and have “never questioned specific military operations or see ourselves having an operational role”. The only thing they’re still talking about is issues of (1) fully autonomous weapons and (2) domestic mass surveillance, and ensuring those things don’t happen. Anthropic continues to talk with the DoW to “de-escalate” the situation and come to some agreement that works for both parties. “We will try our very best to do that”.

To that end, I thought [this analysis from Under Secretary of State Jeremy Lewin](<https://x.com/undersecretaryf/status/2029051832241094791?s=42>) about the difference between the rejected Anthropic contract and the OpenAI contract pointed to a way forward:

> In the final calculus, here is how I see the differences between the two contracts:
> 
>   * Anthropic wanted to define “mass surveillance’ in very broad and non-legal terms. Beyond setting precedents about subjective terms, the breadth and vagueness presents a real problem: it’s hard for the government to know what’s allowed and what’s permitted. In the face of this uncertainty, Anthropic wanted to have authority over interpretive questions. This is because they distrusted the govt regarding use of commercially available info etc. Problem is, it placed use of the system in an indefinite state of limbo, where a question about some uncertainty might lead to the system being turned off. It’s hard to integrate systems deeply into military workflows if there’s a risk of a huge blow up, where the contractor is in control, regarding use in active and critical operations. Representations made by Anthropic exacerbated this problem, suggesting that they wanted a very broad and intolerable level of operational control (and usage information to facilitate this control).
>   * Conversely, OpenAI defined the surveillance restrictions in legalistic and specific terms. These terms are admittedly not as broad as some conceptions of “mass surveillance.” But they’re also more enforceable because there’s clarity rewarding terms and limitations. DoW was okay with the specific restrictions because they were better able to understand what was excluded, and what was not. That certainty permitted greater operational integration. Likewise, because the exclusions were grounded in defined legal terms and principles, interpretive discretion need not be vested in OpenAI. This allowed DoW greater confidence the system would not be cut off unpredictably during critical operations. This too allowed for greater operational reliance and integration.
> 

Sam Altman and OpenAI have been [in pure damage control mode](<https://www.wsj.com/tech/ai/openai-ceo-altman-defends-pentagon-work-to-staff-calls-backlash-really-painful-76d769ec>) after swooping in to sign a contract to replace Anthropic just hours after talks fell apart. There are reports of consumer cancellations — which is much more of a risk for OpenAI given their much larger consumer business (and Claude is now number one in the App Store), but I’m sure the larger concern is employee retention. The ultimate irony, however, may be that OpenAI actually ends up writing the sort of contract that gives Amodei a point to de-escalate to.

Specifically, to the extent that Lewin’s analysis is correct is the extent to which there is a reasonable compromise. If in fact the Pentagon is willing to accept well-defined limits on AI model use, that’s something Anthropic should be happy with; it’s eminently reasonable that the Pentagon can not accept something more amorphous. I very much hope that Anthropic seizes the opportunity that OpenAI has presented them, for both their sake and for America’s.

### Nvidia Earnings

From the [Wall Street Journal](<https://www.wsj.com/business/earnings/nvidia-earnings-q4-2026-nvda-stock-73bd6dc5>):

> Nvidia reported a 94% increase in profit and record sales for the fourth quarter, helping ease concerns over a possible artificial-intelligence bubble that rippled through markets in recent months. The chip maker reported net income of $43 billion, up from $22.1 billion in the year-earlier quarter, on sales of $68.1 billion, up 73% from $39.3 billion a year earlier, easily beating consensus estimates. Analysts polled by FactSet had predicted net income of $37.5 billion and revenue of $66.1 billion for the quarter. Data center hardware—the chips and networking equipment that Nvidia sells to AI and cloud-computing companies—accounted for 91.4% of the quarter’s sales, or $62.3 billion, and the segment’s revenue grew slightly faster than the company’s overall sales.

Wall Street wasn’t impressed; from [Bloomberg](<https://www.bloomberg.com/news/articles/2026-02-25/nvidia-s-rosy-revenue-forecast-shows-the-ai-boom-remains-strong>):

> Nvidia Corp., the dominant maker of artificial intelligence chips, suffered its worst stock decline in 10 months after the company’s latest forecast failed to dispel fears of an AI bubble. The shares fell 5.5% to $184.89 in New York, marking the biggest one-day drop since April 16. The decline followed a first-quarter sales outlook that — on its face — looked impressive. Nvidia easily beat the average analyst estimate and delivered a 73% surge in fourth-quarter revenue. The reaction was a stark reminder of the skepticism now surrounding Nvidia. After explosive sales growth turned the chipmaker into the world’s most valuable company, investors are seeking stronger assurances that booming AI spending can be maintained.

Most of the market negativity around AI has been centered on software companies and how they are all threatened by AI; what is worth pointing out, however, is that this viewpoint is actually diametrically opposed to being skeptical about Nvidia’s continued growth.

There are three key inflection points that AI has already gone through that directly apply to Nvidia’s addressable market:

  * The first inflection point was the emergence of LLMs — call this the ChatGPT moment. In this first paradigm tokens were generated by GPUs and presented as the answer to a question.
  * The second inflection point was the emergence of reasoning models — call this the o1 moment. In this paradigm there are a very large amount of tokens that are generated to figure out the answer _before_ the answer is actually generated; this was an exponential increase in the addressable market for tokens.
  * The third inflection point was the emergence of functional agents — call this the Opus 4.5 moment. In this paradigm those reasoning models are not triggered by humans asking a question, but by an agent solving a problem. This increases the market in two directions: first, humans can run multiple agents, and secondly, agents can leverage reasoning models multiple times to accomplish a task. This isn’t just an exponential increase in the addressable market for tokens, it’s actually squared.

This is why CEO Jensen Huang felt confident that hyperscalers would continue to grow CapEx in 2027 and beyond, even as they plan to spend up to (or beyond) their free cash flow this year; from the question-and-answer section of [the earnings call](<https://seekingalpha.com/article/4874926-nvidia-corporation-nvda-q4-2026-earnings-call-transcript>):

> I am confident in their cash flow growing. And the reason for that is very simple. We have now seen the inflection of agentic AI and the usefulness of agents across the world and enterprises everywhere. You’re seeing incredible compute demand because of it. In this new world of AI, compute is revenues. Without compute, there’s no way to generate tokens. Without tokens, there’s no way to grow revenues. So in this new world of AI, compute equals revenues. And I am certain that at this point with the productive use of Codex and Claude Code and the excitement around Claude Cowork and just the incredible enthusiasm about OpenClaw and the enterprise versions of them. All of the enterprise ISVs who are now working on agentic systems on top of their tools platforms. I am certain at this point that we are at the inflection point, we’ve reached the inflection point and we’re generating profitable tokens that are productive for customers and profitable for the cloud service providers. And so the simple logic of it, the simple way to think about it, is computing has changed. What used to be software running on computers, modest amount of computers, call it, $300 billion or $400 billion worth of CapEx each year has now gone into AI. And AI in order to have — in order to generate tokens, you need compute capacity. And that translates directly to growth and that translates directly to revenues.

Huang added in a later answer:

> It’s really important to realize that inference equals revenues now for our customers because agents are generating so many tokens, and the results are so effective. When the agents are coding, it’s off generating thousands, tens of thousands, hundreds of thousands because they’re running for minutes to hours. And so these systems, these agentic systems are spawning-off different agents, working as a team. The number of tokens that are being generated is really, really gone exponential. And so we need to inference at a much higher speed. And when you’re inferencing at a much higher speed and each one of those tokens are dollarized, it directly translates into revenues. And so inference performance equals revenues for our customers.

What ties this all together is that to the extent that software companies are threatened by AI is the extent that _agents_ — not humans — are doing all of the software development. We are moving well beyond vibe coding: the real disruption scenario is AI providing (and maintaining!) a solution on-demand. Yes, this will cost money; the cost, however, will be much less than a human. Or, more optimistically, the amount of software that can be created when humans are not the limiting factor will be massively more. In either case — the first is bearish for software companies, the latter bullish — we’re going to need more tokens, which means we’re going to need more — and more efficient — chips from Nvidia.

* * *

This Update will be available as a podcast later today. To receive it in your podcast player, visit Stratechery.

The Stratechery Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.

Thanks for being a subscriber, and have a great day!