Ray Ozzie - The Future of Intelligent Machines - [Invest Like the Best, EP.391]

Thematic Sections

1. The Core of Intelligent Machines is "Cloud-Edge" Convergence, Not AI Alone

Ray Ozzie argues that current discussions on "intelligence" focus excessively on AI/LLMs, while true intelligent machines are a deep integration of cloud computing and edge devices. He cites Blues Wireless's RadNote radiation monitoring device as an example: this device is solar-powered, dynamically adjusts sampling frequency based on energy availability, and automatically notifies nearby devices to increase sampling rates upon detecting abnormal radiation—a typical paradigm of "swarm intelligence" in machine-to-machine collaboration.

• Historical Context: After the 2011 Fukushima nuclear accident, Ozzie participated in the SafeCast project and discovered that traditional radiation monitoring devices relied on the power grid and wired networks, failing immediately after the disaster. This prompted him to think about building autonomous, low-power, adaptive sensor networks for the physical world.
• Mechanism Breakdown: Blues's NoteCard hardware module integrates a SIM card, secure element, flash memory, and protocol stack, allowing developers to focus solely on business logic without dealing with underlying issues like cellular network authentication and data encryption. The NoteHub cloud service handles data reception and distribution.
• Extrapolation: In the future, all physical devices will become "data pumps," with their value determined not solely by hardware but by service-oriented capabilities (e.g., predictive maintenance, customer experience optimization). Falsification Condition: If device connectivity costs cannot be reduced to levels comparable to Wi-Fi modules, adoption will be slower than expected.

2. IoT's "Disappointment Cycle" Stems from Hardware Complexity, Not Lack of Demand

Ozzie notes that around 2015, the IoT concept was hyped but failed to take off, primarily because the barrier to hardware development is far higher than for pure software. Enterprise IoT projects have extremely high failure rates, as Wi-Fi is unsuitable for commercial products (requiring configuration, poor security), while cellular networks face complexities like device certification and operator data plans.

• Data Chain: During his time at Microsoft, Ozzie's research found that many enterprise IoT prototype projects failed due to "Wi-Fi being unsuitable for commercial products"; switching to cellular then stalled due to "difficulties in device certification" and "mismatched data plans."
• Competitive Landscape: Large enterprises (e.g., automakers, John Deere) can afford to develop proprietary IoT systems, but small and medium-sized enterprises (SMEs), which account for 99.9% of global manufacturing, lack the necessary talent and budget. Blues aims to enable these companies to connect physical devices "as easily as using SaaS."
• Extrapolation: Commercial IoT (e.g., industrial equipment, cold chain logistics) will take off first, while consumer IoT (e.g., smart homes) may face a new round of disappointment due to "connecting for the sake of connection." Signal: When device connectivity costs fall below 5% of the hardware BOM, adoption will accelerate.

3. Azure's Birth Stemmed from Foresight on the "Service-Oriented" Paradigm

Ozzie recalls that after joining Microsoft in 2005, he observed that the company's business units still clung to a "box mentality": Office as a PC software package, Windows Server as a departmental server, Xbox as a home gaming console. Meanwhile, Google had already entered the "services era." His memo, "Internet Services Disruption," analyzed how each business line would evolve in a service-oriented world—ultimately giving rise to Azure, Office 365, and Xbox Live.

• Historical Analogy: Ozzie compared Microsoft's situation to the transition from the mainframe era to the PC era—IBM missed the PC wave by clinging to mainframes. He warned Microsoft not to repeat that mistake.
• Mechanism Breakdown: The core methodology of the memo was "jump to the future and look backward": assume all technologies are commoditized in 10 years, then reverse-engineer how current businesses should transform. For example, the future form of computing and storage was Azure.
• Extrapolation: The current AI wave requires similar thinking—don't ask "how to make money with AI," but "how will AI reshape industries." Falsification Condition: If companies still embed AI as a "feature" rather than "infrastructure," they will not gain long-term competitive advantage.

4. The Intersection of AI and IoT: From "Data Cleaning" to "Data Deluge"

Ozzie believes that traditional machine learning requires manual data labeling and dataset cleaning, which is costly and hard to scale. In contrast, LLMs' large context window capability allows companies to directly collect raw data (without cleaning) and let the model learn anomaly patterns on its own. He is experimenting with "dashboard by query"—users describe needs in natural language, and the LLM automatically generates queries and charts.

• Data Chain: Blues's customer True Manufacturing (commercial refrigeration equipment) uses connected devices to monitor refrigerator temperature, energy consumption, and door openings in real time, enabling predictive maintenance and optimized energy management. Another customer, SoFar Ocean, collects real-time ocean current data via marine buoys, helping shipping companies optimize routes and save fuel costs.
• Mechanism Breakdown: AI's role in IoT falls into two categories: human interface accelerators (e.g., natural language queries) and anomaly detection engines (e.g., equipment failure prediction). The latter requires large amounts of unlabeled data, and LLMs' context window capability makes them ideal tools.
• Extrapolation: Over the next five years, data volume will replace algorithm accuracy as a competitive moat—whoever can collect the most raw data at the lowest cost will train the most effective anomaly detection models. Risk: Data privacy and transmission bandwidth may become bottlenecks.

5. "Software-ization" of Hardware Development is the Next Wave

Ozzie points out that current hardware development is still in the "card deck era"—requiring multiple prototype iterations, physical testing, and certification, with long cycles. AI and LLMs have the potential to shorten this cycle, for example, by automatically generating circuit schematics and optimizing power management algorithms.

• Historical Analogy: Ozzie compares hardware development to software development in the 1970s—programmers then had to write cards, submit jobs, and wait for results. Today, software has achieved "instant iteration," while hardware remains on "physical world time scales."
• Mechanism Breakdown: The bottleneck in hardware development is the scarcity of talent with systems thinking—few people can simultaneously understand chips, firmware, cloud, and business logic. AI may accumulate the knowledge of these experts to assist newcomers in cross-layer design.
• Extrapolation: When AI can automate 80% of routine hardware design tasks (e.g., power budgeting, antenna matching), the barrier to hardware entrepreneurship will drop significantly. Signal: The emergence of a platform like "GitHub for Hardware," or hardware prototype iteration cycles shortening to under one week.

---

Position Moves

Ticker	Analyst View	Key Data
Blues Wireless (Private)	Bullish (Founder's company)	Clients include True Manufacturing (commercial refrigeration), American Crane (large cranes), SoFar Ocean (ocean sensors)
SafeCast (Non-profit)	Bullish (Co-founded)	World's largest open radiation dataset; RadNote device solar-powered with 10-year battery life
Microsoft Azure	Bullish (Previously led creation)	Originated from a 2005 memo; vision currently executed by Satya Nadella's team
John Deere	Neutral (Innovative but closed)	Early IoT innovator, but closed system triggered farmer backlash
True Manufacturing	Bullish (Client case)	Connected commercial refrigeration enables preventive maintenance and customer experience optimization
SoFar Ocean	Bullish (Client case)	World's largest private ocean sensor network; data transmission via satellite + cellular
American Crane	Neutral (Client case)	Large cranes embedded with smart systems for defense and energy sectors
Skylo / AST SpaceMobile	Bullish (Technology trend)	Satellite technology poised to achieve 100% global coverage

---

Judgments Worth Remembering

1. “The core of intelligent machines is not AI, but the convergence of cloud and edge.” (Ray Ozzie) — RadNote devices achieve adaptive sampling through swarm intelligence, proving that inter-machine collaboration is more important than single-machine intelligence.

2. “Commercial IoT will explode; consumer IoT will face another wave of disappointment.” (Ray Ozzie) — Enterprise device connections have clear ROI (predictive maintenance, customer experience), while smart home “connections for the sake of connections” will lead to user fatigue.

3. “Azure was born from the methodology of ‘jumping to the future and looking back.’” (Ray Ozzie) — Assume technology has been commoditized 10 years from now, then reverse-engineer how the current business should transform, rather than pursuing incremental improvements.

4. “AI will shift enterprises from ‘data cleaning’ to ‘data deluge.’” (Ray Ozzie) — The large context window capability of LLMs makes raw data directly usable without manual labeling, turning data volume into a competitive moat.

5. “Hardware development is still in the ‘card era’; AI is expected to software-ize it.” (Ray Ozzie) — Current hardware iteration cycles are long and talent is scarce; AI may automate routine tasks such as power budgeting and antenna matching.

6. “Commercial value is proportional to domain specificity.” (Ray Ozzie) — General-purpose platforms will eventually be commoditized, while deep solutions in vertical domains (e.g., healthcare, logistics) can create long-term value.

7. “Young entrepreneurs should first ‘serve’ a few years at large companies.” (Ray Ozzie) — Understanding corporate organizational dynamics, procurement logic, and complex system management is more valuable than jumping straight into consumer applications.

8. “Mitch Kapor’s trust was the biggest leverage in my career.” (Ray Ozzie) — Even without understanding your vision, being willing to support your passion—this kind of “irrational” support is a key catalyst for innovation.