IT Infrastructure, Digital Economy, Enterprise Systems, and Crowdsourcing Notes

Core IT Infrastructure and Hardware Architecture

Computer Processors: CPU vs. GPU

Computer processing relies on two primary processor architectures designed for distinct operational workloads:

• Central Processing Unit (CPU):

  • Architecture: Optimized for serial processing and execution of sequential algorithms.

  • Core Count: Typically contains 4 to 8 high-performance cores.

  • Primary Functions: Handles system-level operations, executes fundamental operating system tasks, and manages sequential code logic.

• Graphics Processing Unit (GPU):

  • Architecture: Optimized for massive parallel processing across matrix-based calculations.

  • Core Count: Contains hundreds or thousands of smaller, specialized cores.

  • Primary Functions: Handles parallel workloads including high-performance gaming, photo/video rendering, vector processing, and machine learning/AI model training.

Semiconductor Manufacturing and Transistor Scaling

• Transistors: The foundational building blocks of microchips. A single transistor acts as an electronic switch controlling a binary signal ($\text{on/off}$). Chip performance is directly proportional to total transistor density.

• Historical Scaling:

  • Intel 4004 (1971): Contained 2,250 transistors.

  • Apple M3 Ultra (2025): Contains 184,000,000,000 ($184 \times 10^9$) transistors.

• Physical Scale: Modern transistor gate widths are manufactured at sub-micron scales. For scale comparison, the average width of a human hair is approximately $100\,\mu\text{m}$.

iPhone 16 Pro Cost Structure (Bill of Materials)

The hardware manufacturing breakdown for an iPhone 16 Pro demonstrates the dominance of semiconductor components in modern consumer devices:

• Total Component Cost (Bill of Materials / BOM): Approximately \\$550

• Retail Price: \\$999

Technological Scaling Laws

Technological progress in computing hardware and software is mapped across several empirical observations and industrial benchmarks:

• Moore's Law:

  • Statement: The number of transistors integrated on a microchip doubles approximately every two years.

  • Nature: An empirical industrial observation and planning standard, rather than an unchangeable physical law of nature.

  • Historical Baseline: The Apollo Guidance Computer (AGC) from 1969 compared to modern silicon microchips highlights a speed and performance increase of over $100,000,000\times$.

  • Strategic Importance: Serves as a predictable timeline for stakeholder planning across three groups:

    1. Semiconductor Manufacturers (e.g., Intel, AMD, Nvidia): Guides R&D cycles and competitive positioning.

    2. IT & Consumer Electronics: Informs multi-year product design cycles.

    3. Consumers: Influences purchasing and device upgrade cycles.

• Huang's Law (2018): Formulated by Jensen Huang (Nvidia CEO). Asserts that GPU and AI computing performance more than doubles every two years. This acceleration is achieved through full-stack optimization—combining chip architecture, low-precision floating-point formats, tensor cores, NVLink, CUDA, and software algorithms—rather than relying solely on transistor shrinkage.

• Gates's Law: Formulated by Bill Gates (Microsoft co-founder). States that "software gets slower faster than hardware gets faster," referring to software bloat consuming hardware performance gains.

• Zuckerberg's Law (2008): Formulated by Mark Zuckerberg (Meta CEO). Stated that the volume of information shared online by users doubles every year (held empirically through the 2010s).

• Bezos's Law (2014): Formulated by Jeff Bezos (Amazon founder). Stated that the unit cost of cloud computing power halves approximately every 3 years (stalled in the mid-2020s due to global GPU supply constraints).

• Altman's Law / "Moore's Law for Everything" (2021–2025): Formulated by Sam Altman (OpenAI CEO). Asserts that the cost to run a given benchmark level of AI intelligence decreases by approximately $10\times$ every 12 months, with intelligence capabilities following an exponential cost curve.

Evolution of Enterprise Computing and Cloud Infrastructure

Mainframe Computing to Virtualization

• 1950s Mainframes: Computing was centralized in large mainframe installations (e.g., the IBM 360 weighing 1,700 lbs). Users interacted with mainframes via batch terminals processing queue jobs strictly one at a time.

• 1970s Virtualization: Systems like the IBM System/370 introduced virtual memory and Virtual Machines (VMs). A VM software abstraction layer enables a single physical computer to host and execute multiple independent, isolated virtual operating systems simultaneously, optimizing physical hardware utilization.

Fundamentals of Cloud Computing

• Formal Definition: The delivery of hosted computing services—including compute, storage, networking, and software—over the internet ("the cloud") enabling on-demand access without direct user management of physical infrastructure.

• Business Model: Hardware resources are pooled into vast data centers and sold to enterprise clients on a pay-as-you-go, utility-style usage fee model (pioneered by Andy Jassy at Amazon Web Services starting in 2003).

• Economic Impact: Cloud computing decentralizes computing capacity in a manner analogous to the industrial rollout of the electric grid in the late 19th century. High upfront infrastructure CapEx by cloud hyperscalers provides scalable, variable OpEx flexibility for end-user businesses.

Hyperscaler Financial Performance

Cloud operations represent the primary profit driver for major technology conglomerates:

Amazon Web Services (AWS)

• Operating Income (2024–2025):

  • AWS 2024 Operating Income: \\$39,834\,\text{million}

  • AWS 2025 Operating Income: \\$45,606\,\text{million}

  • Amazon Consolidated Total (2025): \\$79,975\,\text{million}

• Quarterly Growth: AWS quarterly revenue expanded from \\$1.1\,\text{billion} in Q1 2014 to \\$24.2\,\text{billion} in Q4 2023, representing a 9-year Compound Annual Growth Rate (CAGR) of $37\%$.

Google Cloud (Alphabet)

• Revenues: \\$33,088\,\text{million} (2023) $\rightarrow$ \\$43,229\,\text{million} (2024) $\rightarrow$ \\$58,705\,\text{million} (2025).

• Operating Income: \\$1,716\,\text{million} (2023) $\rightarrow$ \\$6,112\,\text{million} (2024) $\rightarrow$ \\$13,910\,\text{million} (2025).

Microsoft Intelligent Cloud

• Revenues: \\$87,907\,\text{million} (2023) $\rightarrow$ \\$105,362\,\text{million} (2024), representing a $20\%$ annual growth rate.

• Operating Income: \\$37,884\,\text{million} (2023) $\rightarrow$ \\$49,584\,\text{million} (2024), representing a $31\%$ growth rate out of Microsoft's consolidated operating income of \\$109,433\,\text{million}.

Physical Data Center Infrastructure

• Global Scale: Approximately 9,000 active data centers operate worldwide (as of December 2025), a count projected to triple by 2030.

• Hyperscale Data Center Composition:

  • Server Rooms: Houses server racks; single computing facilities frequently contain over 100,000 servers.

  • Cooling Systems: Essential for managing thermal output. HKUST launched the largest IT liquid immersion cooling system in Hong Kong, expanding it into a dedicated 8-story high-performance computing data center on campus in 2026.

  • Network Rooms: Links server clusters to core transit providers. Microsoft alone maintains over $600,000\,\text{km}$ of fiber-optic cabling ($\sim 15\times$ the Earth's circumference).

Telecommunications Infrastructure and Network Vulnerabilities

Subsea Fiber-Optic Networks

Over $95\%$ of global transoceanic internet traffic travels through undersea fiber-optic cables anchored at coastal landing stations.

Primary Internet Infrastructure Vulnerabilities

1. Physical Damage: Undersea cables are vulnerable to marine hazards. In 2008, a ship anchor dragged along the Mediterranean floor severed two primary subsea cables connecting Europe to the Middle East and Asia, severing connectivity across Egypt and India.

2. Government Interference & Chokepoints: Centralized international gateways allow single state actors to shut down national internet access. During the 2011 Tahrir Square protests, the Egyptian government severed all national connections to global networks by targeting a small number of international landing points.

3. Natural Disasters: Extreme weather destroys localized server infrastructure. In October 2012, Superstorm Sandy made landfall in New Jersey, causing over $30\%$ of local web servers in affected zones to drop offline.

Satellite Internet Constellations

To bypass physical terrestrial chokepoints and provide global connectivity, Low-Earth Orbit (LEO) satellite constellations (e.g., SpaceX Starlink) establish direct-to-cell data networks. Satellite signals route directly between LEO satellite arrays, ground stations, partner operator networks, and unmodified mobile cellular devices.

Enterprise Information Systems and Operational Simulation

Business Software Market Dynamics

Custom enterprise software commands the largest share of the global software development market (projected to reach \\$295.17\,\text{billion} by 2030), outpacing web-based applications and mobile apps.

Core Enterprise Information Systems (EIS)

Modern organizations integrate four specialized information system layers:

1. Customer Relationship Management (CRM): Manages front-office interactions, marketing campaigns, customer records, and sales funnels.

2. Enterprise Resource Planning (ERP): Integrates core internal business processes including accounting, human resources, manufacturing, inventory management, and financial records.

3. Supply Chain Management (SCM): Coordinates information, materials, orders, and logistics flow between suppliers, manufacturers, wholesalers, and retailers.

4. Business Intelligence (BI): Extracts, aggregates, and analyzes enterprise data to support data-driven decision-making.

Supply Chain Mismatches: The Bullwhip Effect

• Definition: The Bullwhip Effect describes how small fluctuations in end-consumer retail demand amplify into progressively larger swings in demand forecasts and inventory orders as information moves upstream through retailers, wholesalers, manufacturers, and raw material suppliers.

• Root Cause: Information asymmetry, delayed reporting, and uncoordinated local optimization across organizational tiers.

HKUST McDonald's Lunch Crisis Case Study

When uncoordinated, local optimization across individual functional roles creates operational failure:

Takeaway: Information must be recorded, analyzed, and shared across departments simultaneously through integrated ERP/SCM systems to prevent operational breakdowns.

Digital Twins

Digital Twins create exact virtual software representations of physical assets, facilities, or operational processes (e.g., Nvidia Omniverse). By running real-time simulations fed by sensor data, companies stress-test supply chains, model fleet robotics, and resolve warehouse bottlenecks virtually before physical deployment.

The Open-Source Economy

Proprietary vs. Free Operating Systems

• Proprietary Software: Closed-source code controlled exclusively by a single corporate owner (e.g., UNIX released in the 1960s by Bell Labs; Apple Mac OS released in 1984; Microsoft Windows released in 1985).

• Free & Open-Source Software (FOSS):

  • Origins: Launched by Richard Stallman via the GNU Project (1984) and the Free Software Foundation (FSF, 1985).

  • Definition of "Free": Refers to freedom of control, modification, and redistribution ("free speech"), not zero financial cost ("free beer"). Users retain full rights to run, inspect, modify, and redistribute the underlying source code.

Legal Permissions and Copyleft (GNU GPL v3.0)

The GNU General Public License (GPL) uses "copyleft" legal mechanisms to guarantee software freedoms:

• Source Code Provision: Distributors must provide access to complete, readable source code.

• Passing Freedoms: Anyone distributing GPL-licensed software (gratis or for a fee) must pass along the exact same freedoms to recipients.

• No Re-Enclosure: Users cannot modify GPL code and re-release it inside closed, proprietary commercial programs.

• Patents: Developers cannot use patent claims to restrict user freedoms.

Open-Source Commercial Ecosystems and Brand Conditions

Commercial enterprises leverage open-source models while enforcing strict intellectual property conditions. For example, Meta releases its Llama AI models under a custom license that requires any fine-tuned, quantized, or derivative AI model built on Llama to explicitly include "Llama" at the beginning of the derivative model's name (e.g., Llama-3.2-1B-Instruct-GGUF, Llama-Breeze, Dolphin3.0-Llama3.2).

Motivations for Contributing to Open-Source

• Individual Level:

  • Intrinsic: Sense of autonomy, community belonging, peer recognition, and technical mastery.

  • Extrinsic: Cost reduction, personal portfolio development, and career advancement opportunities.

• Organizational Level:

  • Setting industry standards and expanding developer ecosystems.

  • Crowdsourcing bug fixes and security audits across global developer pools.

  • Lowering foundational R&D costs by sharing base infrastructure.

The Crowdsourcing Economy

Crowdsourcing replaces internal employee execution with inputs distributed across an online public crowd.

Crowdsourcing Ideas (Open Innovation)

• NASA LunaRecycle Challenge: A \\$3\,\text{million}, two-track, two-phase global competition focused on designing recycling solutions to process non-metabolic waste during long-duration lunar missions. Led by the Kennedy Space Center and the University of Alabama (Phase 2 final demonstrations scheduled for August 2026).

• Kaggle Data Science Competitions: Organizations post datasets online for public ML model development:

  • American Express Default Prediction: \\$100,000 prize pool for predicting credit default risk.

  • Zillow Home Value Prediction (Zestimate): \\$1,200,000 total prize pool.

  • Intellectual Property Terms: Kaggle competition rules enforce strict IP grants. First-round winners grant non-exclusive, perpetual commercial licenses to sponsors. Second-round participants must fully assign all right, title, interest, source code, patents, algorithms, and trade secrets to the corporate sponsor.

• LEGO Ideas: Platform where community members submit custom LEGO set builds. Designs reaching 10,000 public votes are reviewed for official commercial mass production.

Crowdsourcing Capital (Crowdfunding)

Platforms like Kickstarter allow entrepreneurs to raise early-stage capital directly from public backers:

• Bambu Lab 3D Printer: Raised \\$7,019,236.

• Coolest Cooler: Raised \\$13,285,226.

Specialized Crowdsourcing Verticals

• Crowdsourced Content: ViralHog acts as an agent licensing user-submitted viral videos to media outlets, splitting ad revenue with creators.

• Crowdsourced Mapping: Hivemapper incentivizes drivers to capture real-time street mapping data using connected dashcams.

• Crowdsourced Voice AI Data: Silencio Network gathers audio recordings from over 2.5 million contributors across 180+ countries paid in stablecoin to train voice AI models.

• Crowdsourced Labor for AI: RentAHuman enables AI software agents to hire human workers to perform physical real-world tasks.

• Crowdsourced Physical Gaming: Frodobots enables online players to remotely drive small physical rovers worldwide.

• Crowdsourced AI Security: Agent Breaker gamifies prompt injection and adversarial attacks on LLM applications to crowdsource vulnerability discovery.

The Sharing Economy

Asset-Sharing Evolution: Heavy vs. Light Models

• Origins: Car-sharing originated in Switzerland in 1948 (later rebranded as Mobility in 1997).

• Heavy-Asset Model (Zipcar): Founded in 2000 in Cambridge, MA (IPO in 2011). The company owned and maintained its fleet of rental vehicles, incurring high capital expenses and vehicle depreciation burdens.

• Light-Asset Platform Model (Uber & Airbnb): Platforms connect independent third-party asset owners with buyers without owning the physical assets:

  • Uber (Founded 2008 in Paris by Travis Kalanick & Garrett Camp): Owns zero passenger vehicles and classifies drivers as independent contractors. Annual global rides grew from 1.5 billion in 2015 to 6.5 billion in 2016; net revenue scaled from \\$0.1\,\text{billion} (2013) to \\$37.28\,\text{billion} (2023).

  • Airbnb (Founded 2007 by Brian Chesky & Joe Gebbia): Owns zero real estate. Annual revenue scaled from \\$0.4\,\text{billion} (2014) to \\$11.1\,\text{billion} (2024).

Data Science & Machine Learning in Sharing Platforms

• Uber ETA Prediction: Custom ML routing engines integrate real-time traffic telemetry and spatial mapping data to generate arrival predictions superior to baseline mapping APIs.

• Airbnb Demand Analytics: Academic research ("What Makes a Good Image? Airbnb Demand Analytics Leveraging Interpretable Image Features" by Shunyuan Zhang, Dokyun Lee, Param Vir Singh, and Kannan Srinivasan) demonstrated that high-quality host photos utilizing composition techniques such as the Rule of Thirds (ROT) and Diagonal Dominance directly drive higher booking conversion rates and revenue.

Niche Asset-Sharing Ecosystem

• Swimply: Hourly rental of private backyard swimming pools.

• Neighbor: Peer-to-peer self-storage marketplace utilizing empty residential garages/basements.

• Bounce: Luggage storage network using excess retail shop space.

• Sniffspot: Private dog park rentals on private land.

• TULU: On-demand rental kiosks in residential buildings for household appliances and tools.

• Wingly: Flight-sharing platform matching private pilots with passengers.

• Eatwith: In-home dining experiences hosted by local cooks.

• TimeBanks.org: Currency-free barter exchanges trading equal hours of human labor.

The Gig Economy and Digital Challenges

Historical Evolution of Gig Work

1. 1920s: The term "gig" coined by jazz musicians to describe single performance engagements.

2. 1940s: Emergence of temporary staffing agencies offering contract work as an alternative to corporate employment.

3. Late 1990s–2000s: Digital platforms launch (Upwork, Uber, Airbnb, TaskRabbit).

4. 2020+: Global pandemic accelerates remote work and independent contractor arrangements. McKinsey research indicates 1 in 6 traditional employees desires to transition into a primary independent earner.

Contractual Realities and Labor Vulnerabilities

Gig platforms operate under Independent Contractor Service Agreements (e.g., Foodpanda, Keeta) that legally shift operational risks onto workers:

• Platform Role Disclaimer: Contracts explicitly define platforms as technology matching software rather than employers.

• Complete Liability Transfer: Contractors absorb all costs for vehicle maintenance, insurance coverage, third-party injuries, and property damage.

• Lack of Benefits: Zero sick leave, health insurance, paid holiday, or pension contributions.

• Labor Disputes: Hong Kong couriers organized strikes against Keeta in May 2025 following pay cuts that reduced per-order earnings from \text{HK}\\$40\text{--}50 down to \text{HK}\\$30 alongside the elimination of peak-hour bonuses.

Macro Challenges of the Digital Economy

• Digital Divide: Socioeconomic exclusion of underprivileged populations lacking access to high-speed devices or AI interfaces.

• Information Overload: Average global daily screen time reached 6 hours and 58 minutes per person.

• Cognitive & Brain Health: Constant consumption of short-form, high-dopamine content reduces attention spans and impinge deep focus and memory consolidation.

• Mental Health Isolation: Worker isolation, stress, and anxiety resulting from algorithmic management, continuous performance tracking, and app-driven automated evaluation.

Administrative and Course Information

Industry Speaker Series Schedule

• Stanley Sum: Partner at KPMG Hong Kong

  • Schedule: Tuesday, 3 Nov 2026 | 3:00 PM – 4:30 PM

• Nelson Chow: Partner at Argon & Co Asia, President of the Institute for Supply Management (ISM) Hong Kong, HKUST BBA ISMT Alumnus (Class of 2001)

  • Schedule: Wednesday, 4 Nov 2026 | 12:00 PM – 1:30 PM

• Dr. Sheng Qiang: Director of E-commerce Operations at ByteDance

  • Schedule: Thursday, 5 Nov 2026 | 9:00 AM – 10:30 AM

Assessment and Logistics

• Mid-Term Examination:

  • Date: Tuesday, October 6th, 2026

  • Time Window: 19:30 – 21:30 (Exam duration is 1 hour within this window)

  • Venue: LTA

• Class Participation: Accounts for $10\%$ of the final course grade; evaluated exclusively via in-class contribution.