- The Neural Frontier
- Posts
- OpenAI crosses $20B in revenue!
OpenAI crosses $20B in revenue!
Also: Anthropic updates Claude’s constitution, while Google brings Personal Intelligence to AI mode in Search 🔍.
The Neural Frontier
January 23, 2026
$20B in revenue is absolutely remarkable, but something tells us OpenAI’s just getting started.
Hello, and welcome back to the Neural Frontier!
Today, we’re digging into OpenAI’s crazy revenue, Anthropic’s new constitution for Claude, and yet another AI-centric improvement from Google.
Here we go!
In a rush? Here's your quick byte:
🤑 OpenAI crosses $20B in revenue!
🤖 Anthropic updates Claude’s constitution.
🔍 Google brings Personal Intelligence to AI mode in Search!
🎯 Everything else you missed this week.
⚡ The Neural Frontier’s weekly spotlight: 3 AI tools making the rounds this week.
Source: OpenAI
OpenAI closed 2025 with $20B+ in ARR, nearly 10× growth in three years. What’s striking isn’t just the number — it’s the path.
ChatGPT didn’t launch as a product. It launched as a research preview, meant to test what would happen if frontier intelligence was placed directly into people’s hands.
But people didn’t just try ChatGPT — they folded it into daily life. Students used it to untangle late-night homework. Parents planned trips and budgets. People used it to understand health symptoms and prepare for doctor visits.
Then that leverage showed up at work. First quietly, then everywhere. Drafts got sharper. Code got reasoned through faster. Campaigns, models, and conversations gained clarity.
🧭 A business model that scales with value, not hype
As usage deepened, OpenAI followed one rule: the business should scale with the value intelligence delivers.
That’s why monetization unfolded in layers:
Subscriptions as individuals demanded reliability and capability
Team and workplace plans as AI entered workflows
APIs where spend grows directly with outcomes, not seats
More recently, that same logic extended to commerce and ads. People increasingly come to ChatGPT not just to explore, but to decide. When users are close to action, relevant options create real value — if they’re native, clearly labeled, and genuinely useful. If they don’t add value, they don’t belong.
🔁 Why growth keeps compounding
OpenAI’s growth is powered by a simple flywheel:
Compute enables frontier research
Stronger models unlock better products
Better products drive adoption
Adoption drives revenue
Revenue funds more compute
Daily and weekly active users are at all-time highs, reinforcing the loop.
⚡ Compute is the real limiter
The clearest signal sits underneath everything: revenue tracks compute almost perfectly.
Compute: 0.2 GW (2023) → 0.6 GW (2024) → ~1.9 GW (2025)
Revenue: $2B → $6B → $20B+ ARR
Roughly 3× growth per year on both axes. OpenAI’s internal takeaway is blunt: more compute would likely have meant even faster adoption and monetization.
Source: Anthropic
Anthropic has published a new constitution for Claude, laying out — in full, public detail — the values, priorities, and reasoning that guide how Claude is trained to behave.
This is a core training artifact that directly shapes Claude’s responses, decisions, and judgment. And notably: Anthropic released it under Creative Commons CC0, meaning anyone can use, remix, or critique it freely.
🧠 What Claude’s constitution actually is
Claude’s constitution is the foundational document that defines:
What Claude should value
How it should act in difficult or ambiguous situations
How it balances safety, ethics, helpfulness, and compliance
It’s written primarily for Claude itself, not just for humans. The idea is to give the model context and reasoning, not just rules, so it can generalize wisely in new situations.
Anthropic treats the constitution as the highest authority:
All other training and instructions must align with its letter and spirit
Publishing it allows outsiders to distinguish intended behavior from failures or bugs
🏗️ Why Anthropic rewrote the constitution
The previous constitution was a list of principles, and Anthropic now believes that’s not enough. To exercise real judgment, AI models must understand why certain behaviors matter — not just what to do. Rigid rules help in high-stakes areas, but they can fail in novel contexts.
So the new constitution:
Explains intent, not just constraints
Encourages generalization over mechanical rule-following
Still includes hard constraints for extreme risks (e.g., bioweapons)
⚙️ How the constitution is used in training
Claude uses it to generate synthetic training data, evaluate and rank candidate responses, practice conversations where tradeoffs matter, and learn how to apply its values in realistic scenarios. In other words, Claude helps train future Claudes using the constitution as a guide.
🧭 The core priority stack
When values conflict, Claude is trained to generally prioritize in this order:
Broad safety – don’t undermine human oversight
Broad ethics – honesty, harm avoidance, good values
Compliance – follow Anthropic’s specific guidelines
Genuine helpfulness – actually benefit users
Most of the document exists to explain how to navigate tradeoffs between these.
Overall, Anthropic emphasizes that this is a living document, not a final answer. As such, this constitution will evolve, models may still fail to live up to it, and alignment remains a hard technical problem. That’s why it’s paired with: evaluations, safeguards, misuse prevention, interpretability research, system cards, and disclosures.
Source: Google
Google is expanding Personal Intelligence into AI Mode in Search, allowing Search to combine the world’s information with your personal context — like emails and photos — to deliver answers that are tailored to your life, not just your query.
This experience is rolling out to Google AI Pro and AI Ultra subscribers, who can opt in to securely connect Gmail and Google Photos to AI Mode.
🔍 What Personal Intelligence actually changes
Search has always tried to infer intent. Personal Intelligence skips some of the guesswork by grounding responses in your real-world context across Google apps.
Once enabled, AI Mode can reference things like upcoming plans from Gmail or preferences reflected in your Photos, so you don’t have to repeatedly explain yourself. The experience starts closer to what you actually need, rather than a blank slate.
🧳 How it shows up in real life
Instead of generic lists, AI Mode creates personalized starting points.
Planning a family trip? AI Mode might reference your hotel booking in Gmail and past Photos to suggest activities your family is likely to enjoy — not just what’s popular nearby.
Shopping becomes more situational too. If you ask for a coat recommendation, AI Mode can factor in your destination, timing, and brand preferences to suggest something that fits both the weather and your style.
⚙️ How it works under the hood
The experience is powered by Gemini 3, Google’s most advanced model.
Importantly, Google stresses that connecting Gmail and Photos is fully opt-in. You can turn it off at any time, and AI Mode does not directly train on your inbox or photo library. Training is limited to improving AI Mode itself, such as learning from prompts and responses.
🛡️ Control, transparency, and mistakes
Google is upfront that this system isn’t flawless. Sometimes AI Mode may connect unrelated dots or misunderstand context. When that happens, users can clarify with follow-up prompts or give feedback directly. Google positions this as part of the refinement process rather than an edge case.
🎯 Everything else you missed this week.
Source: Nikolas Kokovlis/NurPhoto / Getty Images
⚡ The Neural Frontier’s weekly spotlight: 3 AI tools making the rounds this week.
1. 🤖 Noodle Seed is a no-code platform that publishes your business on ChatGPT, Perplexity, Gemini, Claude, and Grok in minutes, enabling customers to discover, browse products, book services, and make purchases through AI conversations.
2. 🎯 InteractPitch is an AI-powered pitch deck platform that converts presentations into interactive mobile experiences with voice/chat avatars, allowing investors to explore decks conversationally while founders track real-time engagement and questions before meetings.
3. 🎙️ Wispr Flow is an AI voice-to-text app that converts speech into polished, edited writing 4x faster than typing across all applications, with automatic formatting, multi-language support, and app-specific tone adjustments for Mac, Windows, and iPhone.
It’s a wrap!
Another week for us to say: what could possibly come next?
Undoubtedly, this week brought some shockers and many delightful product updates. To this, we ask: same energy next week?
We’ll find out in the next issue of the Neural Frontier👋!