The first wave of AI startups seems to be sustaining rather than disruptive innovation.

A sustaining innovation targets high-end customers with better performance. Despite the technological complexity of AI, established competitors usually win in sustaining technology battles. Sustaining strategy involves creating superior products with higher profit margins for top customers. Incumbents have the resources and distribution channels to win these battles. In technology world, startups can almost never come from behind with a better product. Instead, they should serve a different market. During the Internet era, Microsoft released Internet Explorer for free as a default browser on Windows and killed Netscape. Since the release of ChatGPT, Adobe, Salesforce, Microsoft, ServiceNow, Facebook, and Google all have released, acquired, or announced AI products. Indeed, investing in startups with potential M&A opportunities might be a wise strategy for those who believe in luck.

In the early days of the Internet, Andy Grove famously said, “All companies will be internet companies.” The same will be true about AI.

The existing infrastructure is bloated. As performance attributes change, many parts of the tech stack are ripe for reinvention. Language models can simplify interfaces and flatten the learning curves of tools, disrupting non-consumption and creating new markets. Intelligence functions can create new software categories, productizing and decreasing costs in the service sector, particularly healthcare, legal, education, professional, and creative services. There will also be a host of applications in art, education, finance, entertainment, and commerce with better user experience. The transition from logic machines to probabilistic machines and the emergence of new bottlenecks like memory will usher in a new architecture for computing, reviving analog processors for edge devices. The tech stack to build robots (interface, perception, and navigation) is finally available, and digital twins will be the foundation for training robots and digitizing factories. Finally, transformer models, simulators, and AI solvers for physics, biology, and chemistry will revolutionize research and discovery and unlock the world of atoms by unleashing synthetic biology.

Where iconic AI companies will be created?

1. Frontier AI technologies, including symbolic and causal reasoning, complex and hierarchical planning, sequences and subtasks decomposition, short-term memory networks for AI, dynamic knowledge graphs and ontologies, explainable AI systems, adaptive real-time AI models, probabilistic programming frameworks, AI control systems, ground truth platforms.
2. Neural networks and foundational models for science. Some major areas include chemistry (e.g., molecule discovery), physics (e.g., solving Maxwell's equations), biology (e.g., gene analysis, protein folding, and drug discovery), electronics (e.g., analog chip design), and math (e.g., solving NP-complete problems).
3. AI hardware such as compute express link, advanced acceleration cards, photonics accelerators, analog chips, and neuromorphic computing.
4. AI tech stack improving emerging performance attributes, easier integration, and lower cost. Areas desperate for better products are around decreasing the training and inference cost, improving database inefficiencies, simplifying training and fine-tuning operations, solving data privacy issues with data marketplaces and data attribution services, controlling, stabilizing, and analyzing model outputs, facilitating the integration of copilots and chat interfaces with existing SaaS products, and enabling dynamic user interfaces powered by AI.
5. Synthetic data platforms, data transformation services, and enrichment APIs for various modalities as ****standalone offerings or as the key component of a much larger workflow.
6. Products with a simple user interface that flatten the learning curve, empowering non-consumption and new markets while counter-positioning and cannibalizing existing business models.
7. Robotic platforms and integrated robots for personal and industrial applications such as retail stores, restaurants, warehouses, hospitals, and elders care with innovations in language control systems, robotic motor adaptation, and digital twins.
8. AI-native products that apply intelligence functions (abstraction, logic, problem-solving, reasoning, critical thinking, and creativity) in novel ways and create new user experiences beyond the skeuomorphic application of LLMs and chatbots. Some prominent examples include:
   1. Productivity tools, platforms, and agents for knowledge workers and brokers (e.g., lawyers, accountants, scientific researchers, healthcare practitioners, procurement specialists, engineering and business consultants) that automate tasks, provide faster access to information and offer a better user experience across different tools.
   2. Enterprise knowledge portals and help desks trained on documents and policies (e.g., HR, IT, administration) to provide 24/7 support, troubleshooting, and access to resources.
   3. Automated content (image, video, voice, etc.) production and repurposing, content analysis and search, fingerprinting and metadata tagging, visual Q&A, and personalized media.
9. Workflows that can be significantly improved by language models and generative AI, such as professional design and development software (e.g., electric design software), customer services, sales calls, robotic process automation, and business intelligence and data visualization tools.
10. Many successful consumer AI products (e.g., social) are based on unpredictable consumer behaviors. There will also be room for successful consumer businesses to solve significant consumer problems, particularly in the services sector:
    1. Knowledge products, agents, and services (legal, accounting, finance, engineering) as a packaged solution. The first iteration is automating outsourced business processes. For example, legal documents such as agreements, leases, motions, power of attorney generated by AI, and form fillers for divorce, immigration, and company registration use cases.
    2. Democratize access to high-quality education. Gamified private tutor for K-12 with personalized teaching material, exams, and feedback. And content platforms creating engaging educational content and new form factors for books.
    3. Personal financial advisor and agent to manage income, investment, and budget, provide superior customer service, and help individuals and families make informed financial decisions.