# Anthropic Overtakes OpenAI in Revenue. Nobody Saw It Coming.

> Google and Amazon pour $65B into Anthropic in 7 days as its ARR overtakes OpenAI's. Cook hands Apple to Ternus. GPT-5.5 and DeepSeek V4 in 48 hours.

- Canonical: https://siliconvalleyconfidential.com/en/dossier/anthropic-overtakes-openai-in-revenue-nobody-saw-it-coming/
- Site: Silicon Valley Confidential (https://siliconvalleyconfidential.com) — weekly executive intelligence on Silicon Valley and global tech
- Author: Jose Luis Cases (https://es.linkedin.com/in/jose-luis-cases-lozano)
- Language: en
- Published: 2026-04-25 (original LinkedIn edition: https://www.linkedin.com/pulse/anthropic-supera-openai-en-ingresos-nadie-lo-vio-venir-cases-5ynee/)

---
MY TAKE
---------------

I've been in tech for fifteen years, and this is the first week I've truly felt the ground move.

It's not any single piece of news. It's the pattern they all form together. Google puts $40 billion into Anthropic. Amazon raises its stake to $25 billion. In total, $65 billion committed to a single company in seven days. That's more than the GDP of Luxembourg...

Meanwhile, Meta and Microsoft announce 20,000 layoffs on the same day — the same day — and the market goes up... again.

The market no longer rewards growth. It rewards substitution.

And Tim Cook is leaving Apple. Those of you who've been reading this newsletter for a while will remember that in the March 21 edition I wrote that Cook went on Good Morning America to deny the retirement rumors, and that this public denial, paradoxically, confirmed the conversation was real.

I said Ternus at 50 was the same age Cook was when he succeeded Jobs. Five weeks later, here we are.

Cook leaves on September 1. Ternus steps in. And what matters most to me isn't who sits in the chair, it's what the choice tells us: Apple believes the next battlefield is hardware integrated with AI, not services or software. They've put a mechanical engineer in charge of the most valuable company in the world. That's a bet, and given the great experience I've had — and still have — with the M1, M2, M3 chips, I like the bet.

Down to business.

According to industry data reported this month, Anthropic has reportedly overtaken OpenAI in revenue. A $30 billion annualized run rate against OpenAI's $24–25 billion. And it did so spending four times less on training. No consumer product. No ChatGPT. Just enterprise. Just Claude Code. Just agents. If confirmed, this isn't a commercial victory.

It's confirmation that the real AI market isn't the chatbot — it's the automation of knowledge work. And to me, that's just how it is.

And while the money moves, the models don't stop. OpenAI launched GPT-5.5 on Wednesday, internally codenamed "Spud," with a one-million-token context window and autonomous multi-step workflows, redefining ChatGPT as an agent "super app."

I'm using it, and it's more agentic — it doesn't need as much detail to give me a powerful solution. I don't know whether I like it or it scares me.

The next day, DeepSeek released V4: 1.6 trillion parameters, performance comparable to GPT-5.4 on code, and its first external funding round in motion at a $20 billion valuation with Tencent and Alibaba.

I haven't tried it yet. There aren't enough hours in the day.

And a mention for ChatGPT Images 2.0: the best image-generation model in existence right now, by a wide margin. Flawlessly rendered text in Japanese, Arabic, Korean, Cyrillic. Styles ranging from hyperrealistic photography to manga to pixel art — today's newsletter image was created this way, with a lot of text.

It's the first time an image generator produces results you don't need to retouch.

Two frontier models in 48 hours, one American and one Chinese, each backed by its geopolitical bloc. The technological bifurcation between the US and China is no longer a hypothesis — it seems to be the architecture of the market.

And speaking of things you have to try to believe:

Anthropic launched Claude Design this week. I've been using it these past few days and at first the experience was "meh," but once you put time into it, it is, without exaggeration, a religious experience.

It connects to your code repository, analyzes your project, and proposes designs that don't look AI-generated — no AI slop, none of those generic interfaces you've seen a thousand times. It makes distinctive proposals, with judgment, that understand the context of what you're building.

But what really kills Figma isn't the design — it's the handoff to Claude Code, which is unbeatable. You go from prototype to working code in minutes, inside the same ecosystem, without exporting assets or translating specifications. Figma needs an army of plugins and a developer on the other side interpreting the design. With Claude Design you go straight to Claude Code — that intermediate step disappears.

Figma's stock fell 7% the day of the announcement. I don't think Figma can survive, because the current paradigm no longer separates the developer from the designer. They're merging.

The question I ask myself as a CTO: when your infrastructure provider, your language model, and your development tool are all controlled by two hyperscalers that just invested $65 billion in the same company, and the models update every 48 hours... what's your plan B?

I don't have an answer. But I know that ignoring the question is no longer an option. I don't know whether you should spend hundreds of thousands of euros on GPU racks and run powerful open-source models... like Kimi, etc... or what to do.

Because you can't delegate your entire company to a subscription...

I've made you a game, so you can see the power of ChatGPT Images 2:

The image has the letters SVC somewhere — an easter egg... a secret signature in a painting... I'll read you in the comments if you find it.

Thanks for reading. And you know: share, comment... that's life. :)

I'll leave you with the minions.

### THE BOMBSHELL OF THE WEEK

### Google invests $40 billion in Anthropic. Amazon raises to $25 billion. $65 billion in one week.

The largest investment in an artificial intelligence company in history was announced on April 24. Google (Alphabet) committed up to $40 billion in Anthropic, structured as $10 billion committed immediately at a $350 billion valuation, with an additional $30 billion contingent on performance milestones. The deal includes 5 gigawatts of compute capacity on Google Cloud over five years.

Four days earlier, on April 20, Amazon had announced an expansion of its investment: $5 billion immediately, with the right to invest up to $20 billion more. In exchange, Anthropic committed to spending more than $100 billion on AWS over the next decade.

The timeline:

*   April 20: Amazon announces expansion to $25 billion. Anthropic's commitment: $100 billion on AWS over 10 years.
*   April 24: Google announces up to $40 billion. Valuation: $350 billion. 5 GW of compute on Google Cloud.
*   Total committed in 7 days: more than $65 billion.

What nobody is saying: Anthropic is now hostage to two hyperscalers that compete with each other. Amazon has locked Anthropic into AWS with $100 billion in spending commitments. Google has secured preferential access to Anthropic's frontier models from Google Cloud. Anthropic's independence as a company is already a legal fiction. It is, in practice, an infrastructure joint venture between two giants using the neutrality of the "Anthropic" brand as competitive cover.

The figure that changes the conversation: According to industry data published in early April (SaaStr, The AI Corner), Anthropic has reportedly overtaken OpenAI in revenue. Its annualized run rate (ARR) is said to have reached $30 billion, versus OpenAI's $24–25 billion. The reported trajectory: $1 billion in January 2025, $9 billion at the end of 2025, $14 billion in February 2026, $19 billion in March, $30 billion in April. Exponential growth, driven by Claude Code and enterprise adoption. More than 1,000 companies spend over a million dollars a year on Anthropic. These figures have not been officially confirmed by Anthropic, but they are consistent with the level of investment Google and Amazon announced this week.

OpenAI, for its part, internally projects $14 billion in losses for 2026, with a $17 billion burn rate, and doesn't expect to be cash-flow positive until 2029–2030.

The signal: The frontier AI market has stopped being a model race and has become an infrastructure war. The hyperscalers are buying exclusive access to models in exchange for compute. Whoever doesn't have a hyperscaler behind them doesn't compete. OpenAI is on its own.

### POWER MOVES

### Tim Cook leaves Apple. The hardware engineer arrives.

On April 20, Apple confirmed that Tim Cook will step down as CEO on September 1, 2026, after fifteen years. His successor: John Ternus, 51, SVP of Hardware Engineering. Cook will become executive chairman.

Everyone is covering this as an orderly succession. It isn't. Cook transformed Apple into a logistics and services company — $100 billion a year in services, supply-chain mastery, geopolitical navigation. Ternus builds iPhones. Apple is betting that the next competitive battlefield is hardware-AI integration, not software as a service.

The power move the coverage missed: Johny Srouji, the architect of Apple Silicon, becomes Chief Hardware Officer. That's the real consolidation. The man who designed the chips that made Apple Silicon possible now controls the entire hardware layer. Ternus runs the company; Srouji runs the machinery.

Mark Gurman identified six new product categories in development: AI-powered AirPods, smart glasses, a smart pendant, a smart display, a tabletop robot and a security camera. Ternus has twelve months to prove he can turn this into modelable revenue.

Sources: [Apple Newsroom](https://www.apple.com/newsroom/2026/04/tim-cook-to-become-apple-executive-chairman-john-ternus-to-become-apple-ceo/)
 | [TechCrunch](https://techcrunch.com/2026/04/20/tim-cook-stepping-down-as-apple-ceo-john-ternus-taking-over/)
 | [9to5Mac - six categories](https://9to5mac.com/2026/04/23/apple-has-six-major-new-product-categories-coming-says-mark-gurman/)

### Fermi America: CEO ousted, CFO gone 48 hours later, 84% value destruction

The most under-covered corporate governance implosion of the week. Fermi America — the nuclear-power-for-AI-data-centers company backed by former Energy Secretary Rick Perry — removed its CEO and co-founder Toby Neugebauer on April 17 (the SEC filing says "stepped down"; Bloomberg describes it as a "removal"). CFO Miles Everson resigned on April 19. Both exits were reported on April 20.

Neugebauer and his family own roughly 40% of the shares and immediately demanded the company be sold. The board, led by Marius Haas (a Dell/HP/Intel veteran from BayPine), refused.

Market cap trajectory: $20 billion at the October 2025 IPO, $3.2 billion on April 21, 2026. 84% destruction in six months. This is a company with no revenue, no confirmed anchor tenant, one customer that canceled a $150 million contract in December, and its largest shareholder and its board at open war.

The signal: The classic bubble cycle — category hype (AI infrastructure), premature IPO, failure of the revenue narrative, founder-board conflict. The first visible casualty of the nuclear-AI fever.

Sources: [Fortune - CEO removed](https://fortune.com/2026/04/20/financially-struggling-ai-power-startup-fermi-loses-ceo-reset/)
 | [Fortune - feud](https://fortune.com/2026/04/21/feud-between-ai-power-startup-fermi-and-its-fired-ceo-and-top-shareholder-heats-up/)
 | [TechCrunch](https://techcrunch.com/2026/04/20/fermi-ceo-and-cfo-depart-texas-nuclear-power-ai/)

* * *

### MONEY TALKS

### SpaceX secures an option to buy Cursor for $60 billion

Announced April 21. SpaceX secured the right to acquire Cursor (the AI coding tool) before the end of 2026 for $60 billion. If SpaceX doesn't exercise the option, it pays Cursor $10 billion for the collaboration. Microsoft had examined acquiring Cursor before SpaceX closed the deal.

The context: SpaceX merged with xAI (Musk's AI lab) in February 2026. Cursor would integrate with the Colossus supercomputer. Musk is building a vertical AI stack entirely outside the Microsoft-OpenAI ecosystem: rockets + supercomputing + AI coding + pre-IPO capital structure.

In parallel, Cursor is raising a $2 billion round at a $50 billion valuation, led by Thrive Capital and Andreessen Horowitz, with Nvidia as a strategic investor. Its annualized ARR: $2 billion, targeting $6 billion by the end of 2026. It's already in 67% of Fortune 500 companies.

The signal: AI coding tools have stopped being software. They're infrastructure. Any CTO who isn't actively monitoring Cursor's penetration in their engineering organization is flying blind.

Sources: [TechCrunch - SpaceX/Cursor](https://techcrunch.com/2026/04/21/spacex-is-working-with-cursor-and-has-an-option-to-buy-the-startup-for-60-billion/)
 | [CNBC - deal terms](https://www.cnbc.com/2026/04/21/spacex-says-it-can-buy-cursor-later-this-year-for-60-billion-or-pay-10-billion-for-our-work-together.html)
 | [TechCrunch - Cursor $50B round](https://techcrunch.com/2026/04/17/sources-cursor-in-talks-to-raise-2b-at-50b-valuation-as-enterprise-growth-surges/)

### DeepSeek seeks its first external round at a $20 billion valuation

Bloomberg and The Information reported that Tencent and Alibaba are in talks to invest in DeepSeek's first external funding round, at a valuation above $20 billion. Tencent proposed acquiring up to 20% of the equity; DeepSeek is resisting. Talks remain open.

In parallel, DeepSeek unveiled DeepSeek-V4: a mixture-of-experts model with 1.6 trillion total parameters, 49 billion active, and a one-million-token context window. They claim to have "closed the gap" with frontier models, with performance "comparable to GPT-5.4" on code benchmarks.

The signal: DeepSeek raising from Tencent and Alibaba anchors it definitively in the Chinese tech ecosystem. Any future collaboration with Western clouds gets complicated. The US-China AI bifurcation consolidates.

Sources: [Bloomberg - DeepSeek funding](https://www.bloomberg.com/news/articles/2026-04-22/deepseek-in-talks-to-raise-at-20-billion-value-the-information)
 | [TechCrunch - DeepSeek V4](https://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models/)

### Blue Energy: $380 million to build nuclear reactors in shipyards

Blue Energy raised $380 million (led by VXI Capital) to build modular nuclear reactors in shipyards for AI data centers. Target: a 1.5 GW plant in Texas. The round was a mix of equity and debt.

It's not an outlier. It's a point on a straight line from Microsoft's deals with nuclear plants, Amazon's acquisition of Talen Energy, and Google's contracts with Kairos Power. The energy demand of AI data centers cannot be met with renewables alone.

Sources: [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-21/blue-energy-raises-380-million-to-build-nukes-for-data-centers)
 | [TechCrunch](https://techcrunch.com/2026/04/21/blue-energy-raises-380m-to-build-grid-scale-nuclear-reactors-in-shipyards/)

* * *

### PRODUCT SECRETS

### GPT-5.5 ("Spud"): OpenAI redefines ChatGPT as a "super app"

OpenAI launched GPT-5.5 on April 24 for the API and all paid ChatGPT tiers. Internal codename: "Spud." It's the first fully retrained base model since GPT-4.5.

Technical details: a one-million-token context window. API pricing at $5/$30 per million tokens (input/output). It uses fewer tokens than GPT-5.4 to complete the same Codex tasks. Standout capability: autonomous multi-step workflows — the user describes a complex task and the model plans, executes, verifies and iterates without per-step intervention.

OpenAI also launched Workspace Agents in ChatGPT (a research preview for Business/Enterprise/Edu): teams can build and share agents for long-running workflows. And ChatGPT Images 2.0 with improved text rendering across multilingual scripts.

The signal: OpenAI responds to the Anthropic threat by expanding ChatGPT's surface into agent territory. But the question is whether the consumer model can compete with Anthropic's enterprise-first strategy.

Sources: [OpenAI](https://openai.com/index/introducing-gpt-5-5/)
 | [TechCrunch](https://techcrunch.com/2026/04/23/openai-chatgpt-gpt-5-5-ai-model-superapp/)
 | [Axios](https://www.axios.com/2026/04/23/openai-releases-spud-gpt-model)

### Google Cloud Next 2026: next-generation TPUs and the agent platform

Google Cloud Next (April 22–25, Las Vegas) was the most relevant enterprise tech event of the quarter.

8th-generation TPUs, for the first time with a dual architecture: TPU 8t (training) and TPU 8i (inference) as separate chips. The 8t scales to 9,600 units with 2 petabytes of shared HBM. The 8i connects 1,152 TPUs with 3x on-chip SRAM. Google claims an 80% price-performance improvement for agentic inference.

Gemini Enterprise Agent Platform: Google relaunches Vertex AI as a complete platform for building, scaling and governing agents. It includes a no-code Agent Designer and access to 200+ models including Claude and Llama.

The A2A (Agent-to-Agent) protocol reaches 150+ organizations. Stable version 1.2, hosted at the Linux Foundation. Five SDKs in production. It is quietly becoming the HTTP of the agent layer.

A $750 million fund to accelerate the agentic AI partner ecosystem.

The signal: Google is betting that the agent orchestration layer will be more valuable than the individual models. If A2A becomes the standard, Google controls the plumbing of the agentic era.

Sources: [Google Blog](https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/google-cloud-next-26-recap/)
 | [TechCrunch - TPUs](https://techcrunch.com/2026/04/22/google-cloud-next-new-tpu-ai-chips-compete-with-nvidia/)
 | [Google - A2A](https://cloud.google.com/blog/products/ai-machine-learning/agent2agent-protocol-is-getting-an-upgrade)

### Claude Design: Anthropic attacks Figma head-on

Launched April 17 (one day before our window, but in full rollout this week), Claude Design is a research preview from Anthropic Labs that generates prototypes, slides and visual work from natural language. Powered by Claude Opus 4.7. No additional cost — included in all paid plans. Claude Design's output connects directly to Claude Code for the design-to-code handoff. Figma's stock fell 7% on the day of the announcement.

This is Anthropic's entry into the pipeline previously dominated by Figma + V0 + Cursor. The integrated Design → Code workflow is a flanking move against Figma and against standalone vibe-coding tools.

Parallel controversy: Anthropic briefly pulled Claude Code from the Pro plan, got massive developer backlash, and reversed the decision. A signal of how central Claude Code has become to Anthropic's revenue narrative.

Sources: [TechCrunch](https://techcrunch.com/2026/04/17/anthropic-launches-claude-design-a-new-product-for-creating-quick-visuals/)
 | [VentureBeat](https://venturebeat.com/technology/anthropic-just-launched-claude-design-an-ai-tool-that-turns-prompts-into-prototypes-and-challenges-figma)

* * *

### REAL NUMBERS

### Anthropic vs OpenAI — the great reversal

Industry data reported in early April 2026. Not officially confirmed by the companies.

Anthropic

*   Estimated ARR (April 2026): ~$30B\*
*   ARR growth (6 months): ~3.3x (from $9B)
*   Training spend: 4x less than OpenAI\*
*   Projected 2026 losses: not available
*   Annual cash burn: not available
*   Enterprise customers >$1M/year: 1,000+
*   Expected profitability: before IPO (Oct. 2026)

OpenAI

*   Estimated ARR (April 2026): ~$24–25B
*   ARR growth (6 months): ~1.5x
*   Training spend: the reference baseline
*   Projected 2026 losses: $14B
*   Annual cash burn: $17B
*   Enterprise customers >$1M/year: not disclosed
*   Expected profitability: 2029–2030

Sources: SaaStr and The AI Corner (~April 7, 2026). Not confirmed by Anthropic.

Capex against people:

The big four — Alphabet, Meta, Microsoft and Amazon — will spend a combined $700 billion on AI capex in 2026. Simultaneously, they're cutting the people those budgets used to fund. Meta raises its capex to $115–135 billion while laying off 8,000. Microsoft offers buyouts to ~9,000 senior employees. For context, Oracle announced in March the cutting of up to 30,000 jobs to free up the $8–10 billion it needs to service $58 billion in new debt for AI data centers.

IPOs on the radar:

*   Cerebras (this week): public S-1 filed April 18. Ticker CBRS on Nasdaq. Target: $35 billion valuation. 2025 revenue: $510 million (+76% YoY). Critical risk: 86% of its revenue comes from two UAE entities (G42 and MBZUAI).
*   SpaceX (context): confidential S-1 filed in early April. Target: raise $75 billion at a $1.75 trillion valuation. Roadshow expected the week of June 8. The public S-1 will reveal the accounting of the xAI merger — that disclosure will be a market event in itself.

Market concentration:

The 10 most valuable US startups now represent 51.8% of the total value of all unicorns (PitchBook), up from 18.5% in 2022. In Q1 2026, $300 billion was deployed globally in venture capital, of which 80% ($242 billion) went to AI. OpenAI, Anthropic, xAI and Waymo absorbed 65% of all global VC in Q1.

Sources: [Layoffs.fyi](http://Layoffs.fyi)
 | [CNBC - labor crisis](https://www.cnbc.com/2026/04/24/20k-job-cuts-at-meta-microsoft-raise-concern-of-ai-labor-crisis-.html)
 | [Tom's Hardware](https://www.tomshardware.com/tech-industry/tech-industry-lays-off-nearly-80-000-employees-in-the-first-quarter-of-2026-almost-50-percent-of-affected-positions-cut-due-to-ai)
 | [PitchBook](https://pitchbook.com/news/articles/top-10-startups-now-control-record-51-8-of-total-unicorn-value)
 | [Crunchbase](https://news.crunchbase.com/venture/record-breaking-funding-ai-global-q1-2026/)

* * *

### THE DRAMA

### Meta records its employees' keystrokes to train AI. The same week it lays off 8,000.

What hasn't received enough coverage alongside the layoff announcement: Meta revealed the Model Capability Initiative, a program to capture keystrokes, mouse clicks and periodic screenshots on all its employees' work computers, with the declared goal of training computer-use AI agents.

Reuters broke it. The Register reported internal protests. On Teamblind, an Apple employee called it "Dystopian AF" and Shopify employees called for collective resistance.

Meta's stated logic: "building agents that help people complete everyday tasks using computers" requires "real examples of how people actually use them." The real implication: employees are training the systems that will replace them. And they were told the same week they learned 8,000 of their colleagues were being shown the door.

The European legal exposure is significant. In Italy, keylogging for productivity purposes is explicitly illegal. In Germany, it's only permitted under suspicion of a serious crime. Meta has more than 100,000 employees across multiple jurisdictions. Expect regulatory friction within 30 days.

Sources: [Fortune](https://fortune.com/2026/04/21/meta-will-start-tracking-employees-screens-and-keystrokes-to-train-ai/)
 | [The Register](https://www.theregister.com/2026/04/22/meta_employee_surveillance_software/)
 | [Gizmodo](https://gizmodo.com/meta-plans-to-turn-its-employees-clicks-and-keystrokes-into-ai-training-data-2000749176)

### China closes the door on American capital in AI. The "Singapore washing" era is over.

Bloomberg reported on April 24 that China's National Development and Reform Commission (NDRC) has instructed multiple high-profile AI startups to reject US investment without government approval. Affected companies: Moonshot AI (considering an IPO), StepFun, and ByteDance (barred from selling secondary shares to US investors without authorization).

The trigger: Meta's $2 billion acquisition of Manus. Manus was a Chinese company founded in Shanghai that had redomiciled in Singapore to access US capital. Beijing saw the precedent and shut it down.

Practical implication: the "Singapore washing" strategy — where AI startups with Chinese founders incorporated in Singapore to access American capital — has been eliminated. Any VC with exposure to Chinese-origin AI companies operating out of Singapore needs an immediate portfolio review.

Sources: [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-24/china-to-curb-us-investment-in-tech-companies-after-meta-deal)
 | [China Money Network](https://www.chinamoneynetwork.com/2026/04/25/china-takes-measures-against-us-investment-in-tech-firms-post-metas-acquisition-of-manus)

* * *

### THE WEEK AHEAD

April 29 — The most important earnings day of the year. Microsoft (Q3 FY26, ~$81.4B revenue expected, Azure +38%), Meta (~$55.5B, +31% YoY), Amazon (~$177.2B) and Alphabet all report the same afternoon. The central question for all four: is AI capex translating into proportional revenue? Azure AI's growth will set the tone for the entire sector.

April 30 — Apple Q2 FY26. Guidance of $107–110B in revenue (+13–16% YoY). First results under the Ternus succession narrative. Watch for signals on the AI-device product pipeline and services attachment rates.

May 7 — Microsoft releases the details of its buyout program. Eligible employees receive the formal information. Expect internal leaks on sentiment and acceptance rates, which will reveal the true severity of the structural reduction intent.

Coming days — Meta Muse Spark. Meta Superintelligence Labs (led by Alexandr Wang) launched Muse Spark, the first proprietary model to replace the Llama branding for frontier models. Developer access and public rollout imminent. It's Meta's strategic pivot away from pure open source — the most significant shift in the AI ecosystem's dynamics since Meta launched Llama.

June 2–3 — Microsoft Build 2026 (San Francisco). An AI-centric conference with no filler. Focus: GitHub Copilot, Microsoft Foundry, and multi-model agent workflows. Keynotes streamed free.