The battle for leadership in the AI industry has entered a new phase. As Google tops nearly all benchmarks with Gemini 3, OpenAI CEO Sam Altman warned in an internal memo that Google’s surge could bring “temporary economic headwinds” to the company. Amid this crisis awareness, OpenAI has launched a new project called “Shallotpeat” to mount a comeback in the AI race.
Sam Altman’s Internal Memo: A Sign of Crisis Awareness
According to an internal OpenAI memo obtained by The Information, Sam Altman frankly acknowledges the threat from competitors.
Key Contents of the Memo
Altman’s Message:
- Economic Headwinds: Recent AI progress by Google and Anthropic could create “temporary economic headwinds”
- Shrinking Technical Lead: OpenAI’s technological lead over competitors (like Google and Anthropic) is shrinking
- Optimistic Outlook: However, emphasizes that OpenAI is “catching up fast”
- Future Confidence: Expects to emerge as the leader in the AI race
Timing of the Memo
Importantly, this memo was written before the Gemini 3 release.
Timeline:
- Memo Creation: Altman issues internal memo before Gemini 3 release
- Gemini 3 Release: November 18, 2025, Google announces Gemini 3
- Benchmark Results: Gemini 3 tops nearly all benchmarks
Altman’s concerns became reality with the Gemini 3 release.
Unusual Nature of the Internal Memo
It’s unusual for a CEO to acknowledge “economic headwinds” from competitor success internally.
What This Indicates:
- Serious Crisis Awareness: OpenAI truly takes Google’s threat seriously
- Transparency: Attitude of frankly communicating risks to employees
- Motivation: Sharing crisis awareness to inspire employees
- Investor Consideration: Anticipating memo leaks and showing there are countermeasures
Google’s Surge with Gemini 3
Altman’s concerns proved accurate with the Gemini 3 release.
Overwhelming Advantage in Benchmarks
Google achieved top scores on nearly all major benchmarks with Gemini 3.
Key Benchmark Results:
- LMArena: Industry-leading score
- Coding Tasks: Superior performance in many areas
- Math Tasks: High accuracy
- Reasoning Tasks: Complex problem-solving ability
- Multimodal: Integrated processing of text, images, and video
Impact on OpenAI and Anthropic
According to The New York Times, the Gemini 3 announcement caused anxiety at OpenAI and Anthropic.
Industry Insider Concerns:
- Autonomous Coding: Gemini 3 could potentially surpass OpenAI and Anthropic in this area
- Image Generation: Enhanced multimodal capabilities could threaten competitors
- Business Impact: If Google’s model proves superior, it could negatively impact OpenAI and Anthropic’s businesses
Google’s Structural Advantages
Google has structural advantages that OpenAI and Anthropic lack.
Google’s Strengths:
- Massive User Base: Billions of Google Search, Gmail, Google Docs users
- Immediate Deployment: Integrating new model into Search from day one
- Data Advantage: Access to vast user data and search data
- Integrated Ecosystem: Complete integration of search, cloud, productivity tools
- Financial Power: Alphabet/Google’s enormous financial resources
OpenAI’s Challenges with Pre-training
The Information’s report revealed technical challenges OpenAI faces.
Pre-training Scaling Problems
OpenAI was reportedly struggling to make progress in pre-training.
Problem Details:
- Scaling Limitations: Optimizations stop working when scaling up models
- GPT-5 Struggles: This problem was prominent during GPT-5 development
- Unexpected Behavior: Optimizations that worked at small scale failed at large scale
- Development Delays: This problem delayed GPT-5 development beyond schedule
Shift to Reasoning Models
Difficulties with pre-training led OpenAI to focus on reasoning models.
Strategy Change:
- o1 Series: Development of models specialized for reasoning
- Chain-of-Thought: Explicitly modeling step-by-step thought processes
- Quality Focus: Focusing on improving reasoning quality rather than simple scaling
Pre-training is “Not Dead”
Altman’s memo and reports indicate OpenAI is recognizing the importance of pre-training.
Importance of Pre-training:
- Foundation Building: Learning knowledge and patterns that form the basis of all capabilities
- Generality: Acquiring general-purpose capabilities for broad tasks
- Scaling Laws: Predictable performance improvements through proper pre-training
- Source of Competitiveness: Excellent pre-training is essential for cutting-edge models
Solving this problem is OpenAI’s top priority.
Shallotpeat: OpenAI’s Comeback Project
OpenAI has launched a new project called “Shallotpeat.”
Overview of the Shallotpeat Project
Project Details:
- Codename: Shallotpeat
- Purpose: Counter Google and regain leadership in the AI race
- Focus: Improving pre-training and developing new models
- Timeline: Details unclear but progressing rapidly
Project Goals
What Shallotpeat Aims to Achieve:
- Solve Pre-training Problems: Overcome optimization issues during scaling
- Counter Google: Develop models with performance surpassing Gemini 3
- Restore Technical Leadership: Regain technological advantage in the AI industry
- Stabilize Business: Overcome “economic headwinds” from competitor success
Meaning of the Project Name
The name “Shallotpeat” is interesting.
Possible Interpretations:
- Shallot: A small but strongly flavored ingredient
- Peat: Foundational soil or fuel
- Combined Meaning: Perhaps intending that small improvements accumulate to create a strong foundation
However, since OpenAI’s internal project codenames are typically not publicly explained, this remains speculation.
Structural Changes in the AI Industry
The competition between OpenAI and Google indicates structural changes in the AI industry.
Shift in Technical Leadership
OpenAI has long been the technical leader in the AI industry, but that position is wavering.
OpenAI’s Past Advantages:
- GPT-3 (2020): Demonstrated the potential of large language models
- ChatGPT (November 2022): Mainstreamed AI chatbots
- GPT-4 (March 2023): Multimodal and advanced reasoning capabilities
- GPT-5 (2025): Further performance improvements
Current Situation:
- Google Gemini 3: Top in nearly all benchmarks
- Anthropic Claude: Surpasses OpenAI in some tasks
- xAI Grok: Temporarily topped benchmarks
Diversification of Development Approaches
AI companies are adopting different development approaches.
Main Approaches:
- Large-scale Pre-training: Google’s strength, successful with Gemini 3
- Reasoning Models: OpenAI’s recent focus, o1 series
- Safety-focused: Anthropic’s approach, Constitutional AI
- Real-time Data: xAI Grok’s strength, access to X data
These approaches each have different strengths and weaknesses, showing the diversity of the AI industry.
Intensifying Benchmark Competition
Benchmark competition among AI companies is progressing at unprecedented speed.
November 2025 Benchmark Changes:
- Early November: OpenAI GPT-5.1 among the top
- November 17: xAI Grok 4.1 tops LMArena
- November 18: Google Gemini 3 reclaims first place
This speed of competition demonstrates rapid AI technology advancement, but simultaneously creates significant pressure for companies.
OpenAI’s Response Strategies
OpenAI is taking multifaceted responses to Google’s surge.
Technical Responses
Shallotpeat Project:
- Solve pre-training problems and develop new models
- Aim for performance that can compete with Google
Continued Development of Reasoning Models:
- Improvements to o1 series and new versions
- Advanced reasoning techniques like Chain-of-Thought
Business Responses
Ecosystem Expansion:
- Hardware partnership with Foxconn (announced November 21)
- $3 billion data center investment
- Addition of ChatGPT group chat features
Partnership Strengthening:
- $38 billion contract with AWS
- Strategic partnerships with companies like Emirates
- Collaborative relationships with educational institutions
Communication Strategy
Improved Transparency:
- Frank situation explanation through Sam Altman’s internal memo
- Sharing crisis awareness and inspiring employees
Maintaining Presence Through Multiple Announcements:
- Multiple major announcements on November 21
- Countering Gemini 3 news
Google’s Strategic Moves
Google has gained the upper hand with the Gemini 3 release.
Advantages in Development Methodology
Gemini 3’s success demonstrates that Google’s development methodology is working.
Google’s Success Factors:
- Continued Large-scale Pre-training: Achieved excellent results in pre-training
- DeepMind Expertise: World-class AI research team
- Abundant Resources: Alphabet’s enormous financial power and computational resources
- Data Advantage: Vast data from Google Search, YouTube, etc.
- Integrated Ecosystem: Integration with search, cloud, productivity tools
Immediate Large-scale Deployment
Google integrated Gemini 3 into Google Search from day one.
Deployment Impact:
- Reach to Hundreds of Millions: Available to hundreds of millions of users from day one
- Real-world Feedback: Data collection from large-scale real usage
- Brand Strengthening: Enhanced brand value through integration of Google Search and AI
Continuous Improvement
Google is continuously improving Gemini 3.
Expected Improvements:
- Corrections based on user feedback
- Addition of new features
- Public release of Gemini 3 Deep Think
Impact on the Entire AI Industry
Competition between OpenAI and Google affects the entire AI industry.
Acceleration of Development Speed
Competition has dramatically accelerated AI technology development speed.
Impacts:
- Accelerated Innovation: New ideas are implemented rapidly
- Faster Benchmark Updates: Top positions change every few weeks
- Increased Pressure: Companies must constantly produce latest results
Impact on Smaller Companies
Intense competition among major companies affects smaller AI firms.
Challenges:
- Resource Gap: Growing gap in computational resources, talent, funding with major players
- Difficulty Differentiating: Cannot compete with major players in general-purpose models
- Acquisition Potential: May become acquisition targets for major companies
Opportunities:
- Specialization in Niche Markets: Models specialized for specific industries or applications
- Pursuit of Expertise: Advantages in specialized fields major players don’t address
- Cost Efficiency: Providing more efficient models or services
Benefits for Users
This competition is ultimately beneficial for users.
User Benefits:
- Performance Improvements: Competition leads to development of better AI models
- New Feature Additions: Companies add new features for differentiation
- Price Reductions: Competition makes services more affordable
- Increased Choice: Can choose from multiple powerful AI platforms
- Accelerated Innovation: New use cases and applications emerge
Future Development Predictions
Competition between OpenAI and Google is expected to intensify further.
Short-term Predictions (Next 3-6 Months)
Expected Developments:
- Shallotpeat Results: OpenAI announces new model or significant improvements
- Google’s Response: Improved version of Gemini 3 or new features added
- Anthropic’s Moves: Claude model updates to compete with OpenAI and Google
- Continued Benchmark Competition: Companies successively release new models
Medium to Long-term Outlook (6 Months - 2 Years)
Industry Changes:
- Ecosystem Competition: Competition at integrated ecosystem level, not single models
- Balancing Specialization and Generalization: Combination of general-purpose and specialized models
- Hardware-Software Integration: Each company develops proprietary hardware
- Strengthened Regulation: Development of regulations accompanying AI technology advancement
- Emergence of New Players: Startups and existing tech companies enter AI competition
OpenAI’s Challenges and Opportunities
Challenges OpenAI Faces:
- Solving pre-training scaling problems
- Countering Google’s integrated ecosystem
- Restoring technical leadership
- Stabilizing business
OpenAI’s Opportunities:
- Comeback through Shallotpeat project
- Differentiation through hardware entry
- Leveraging strong partnerships
- Maintaining advantage in reasoning models
Renewed Recognition of Pre-training Importance
This competition has renewed recognition of pre-training’s importance.
”Pre-training is Not Dead”
While some industry insiders believed “pre-training is no longer important,” Gemini 3’s success proved this wrong.
Why Pre-training is Important:
- Foundational Knowledge: Acquiring broad knowledge and patterns
- Generality: Basic capabilities for various tasks
- Scalability: Predictable performance improvements through proper pre-training
- Downstream Task Performance: Pre-training quality affects all downstream task performance
Balancing Reasoning and Pre-training
Cutting-edge AI models require both excellent pre-training and reasoning capabilities.
Ideal Approach:
- Strong Foundation: Large-scale, high-quality pre-training
- Advanced Reasoning: Reasoning techniques like Chain-of-Thought
- Efficient Architecture: Efficient use of computational resources
- Continuous Improvement: Improvements based on user feedback
Summary
The AI competition between OpenAI and Google indicates a new phase for the industry.
Key Points:
- Leadership Shift: Google tops nearly all benchmarks with Gemini 3
- OpenAI’s Crisis Awareness: Sam Altman’s internal memo warns of “economic headwinds”
- Technical Challenges: OpenAI faces pre-training scaling problems
- Shallotpeat Project: OpenAI launches new project as comeback strategy
- Importance of Pre-training: Proven that “pre-training is not dead”
Sam Altman’s frank acknowledgment of competitor threats internally demonstrates that OpenAI has a serious sense of crisis. Simultaneously, launching the Shallotpeat project shows that OpenAI is not merely taking a defensive stance but actively planning a comeback.
Google’s success proved that large-scale pre-training remains essential for building cutting-edge AI models. It became clear that OpenAI’s focus on reasoning models stemmed from scaling problems with pre-training. The Shallotpeat project aims to solve this problem and restore OpenAI’s position as the leader in the AI race.
While this competition is intense and sometimes harsh for companies, it’s ultimately good for users. As multiple powerful companies compete with each other, AI technology advances rapidly, and better products are delivered faster and at more affordable prices.
Over the coming months and years, it will be very interesting to see what results OpenAI’s Shallotpeat project produces, how Google improves Gemini 3, and how other players like Anthropic and xAI respond. The AI industry can be said to be at a historic turning point.