Technical Deep Dive
The claim that AI creates jobs rests on a technical foundation: the entire AI stack requires human labor at every layer. At the base, training large language models (LLMs) like GPT-4, Claude, or open-source alternatives such as Llama 3 requires massive datasets. This has spawned the role of data engineer—professionals who curate, clean, and label training data. For example, the Common Crawl dataset, a primary source for web text, contains over 250 billion pages, but only a fraction is usable. Companies like Scale AI and Surge AI employ thousands of annotators and quality assurance specialists to filter toxic content, align outputs, and create instruction-tuning datasets. A single model training run for a 70B-parameter model can require 10,000+ person-hours of data work.
Above the data layer sits the model alignment pipeline. Reinforcement Learning from Human Feedback (RLHF) is now standard practice. OpenAI, Anthropic, and Google DeepMind all employ armies of contractors to evaluate model outputs, ranking responses for safety and helpfulness. This is not a transient need; as models become more capable, alignment becomes more complex. The open-source repository trl (Transformer Reinforcement Learning) on GitHub has over 10,000 stars and is used by startups to implement RLHF, but the human-in-the-loop component remains labor-intensive.
Then there is inference infrastructure. Deploying an LLM at scale requires GPU clusters, load balancers, and monitoring systems. Companies like Together AI and Replicate run managed inference services, employing DevOps engineers specialized in GPU optimization. The rise of AI agents—autonomous systems that execute multi-step tasks—has created the role of agent operator or workflow architect. These professionals design prompt chains, integrate APIs, and handle error recovery. For instance, the open-source framework LangChain (over 100,000 GitHub stars) allows developers to build complex agent workflows, but productionizing these systems requires constant human oversight. A single agent failure—like a hallucinated API call or incorrect tool selection—can cascade, requiring a human to intervene.
Finally, hardware is a direct job creator. NVIDIA’s GPU supply chain involves chip designers (VLSI engineers), software engineers for CUDA and TensorRT, and data center technicians. The company’s H100 GPU alone requires a complex global supply chain: TSMC fabricates the dies, SK Hynix supplies HBM3 memory, and Foxconn assembles the modules. Each step employs thousands. The global AI chip market is projected to grow from $53 billion in 2023 to $227 billion by 2032, according to industry estimates, directly correlating with employment in semiconductor design, manufacturing, and deployment.
| Layer | Job Roles | Example Companies | Estimated Global Employment (2024) |
|---|---|---|---|
| Data Curation | Data engineer, annotator, QA specialist | Scale AI, Surge AI, Appen | 500,000+ |
| Model Alignment | RLHF contractor, safety researcher | OpenAI, Anthropic, Google | 100,000+ |
| Inference & Agents | Agent operator, workflow architect, DevOps | Together AI, Replicate, LangChain | 200,000+ |
| Hardware | Chip designer, fab technician, data center ops | NVIDIA, TSMC, Foxconn | 1,000,000+ |
Data Takeaway: The AI job ecosystem is not monolithic; it spans from low-skill data annotation to high-skill chip design. The largest employment is in hardware, but the fastest growth is in agent-related roles, which are projected to double by 2026.
Key Players & Case Studies
Several companies exemplify Huang’s thesis. NVIDIA itself is the most direct example. The company’s workforce has grown from 13,000 employees in 2019 to over 36,000 in 2024, a 177% increase. This growth is driven by AI demand. NVIDIA’s data center revenue alone hit $47.5 billion in fiscal 2024, up from $15 billion the prior year. The company is now building new campuses in Taiwan and the US, each requiring thousands of construction workers, engineers, and technicians.
Scale AI is a case study in data-driven job creation. Founded in 2016, it now employs over 1,200 full-time staff and contracts with over 100,000 remote workers globally for data labeling and RLHF. Its valuation reached $14 billion in 2024. The company’s platform supports everything from autonomous vehicle data to LLM fine-tuning. Scale AI’s CEO, Alexandr Wang, has stated that the demand for high-quality training data is “insatiable,” directly contradicting the notion that AI eliminates jobs.
Anthropic, the AI safety company behind Claude, employs over 1,000 people, including a large team of “constitutional AI” researchers and red-teamers. The company’s safety-focused approach requires ongoing human evaluation, creating roles that did not exist five years ago. Similarly, OpenAI has grown from a small research lab to over 3,000 employees, with job postings for roles like “prompt engineer” and “AI alignment researcher” becoming commonplace.
On the open-source side, Hugging Face has become a hub for model sharing and collaboration. The platform hosts over 500,000 models and employs 200+ people. The community around Hugging Face includes thousands of independent developers who fine-tune models for niche applications, creating a cottage industry of AI consultants.
| Company | 2020 Employees | 2024 Employees | Growth % | Key AI Job Roles Created |
|---|---|---|---|---|
| NVIDIA | 13,000 | 36,000 | 177% | GPU architect, CUDA engineer, data center ops |
| Scale AI | 200 | 1,200+ (full-time) | 500% | Data annotator, RLHF contractor, QA specialist |
| OpenAI | ~200 | 3,000+ | 1,400% | Prompt engineer, alignment researcher, safety analyst |
| Anthropic | 50 | 1,000+ | 1,900% | Constitutional AI researcher, red-teamer |
Data Takeaway: The fastest-growing AI companies are not just adding engineers; they are creating entirely new job categories. The compound annual growth rate (CAGR) for AI-related employment at these firms exceeds 50%, far outpacing the general tech sector.
Industry Impact & Market Dynamics
The AI job creation thesis has profound implications for labor markets and business models. First, it challenges the assumption that automation leads to net job loss. A 2023 study by the World Economic Forum estimated that AI could displace 85 million jobs by 2025 but create 97 million new ones—a net positive. Huang’s argument aligns with this: the displacement is real, but the creation is larger.
Second, the rise of generative AI tools is democratizing production. Platforms like Midjourney, Runway, and ElevenLabs allow individuals to create professional-grade content without teams. This is creating a new class of “solo entrepreneurs” who hire freelancers for specific tasks—editing, distribution, legal—effectively generating new employment. For instance, a YouTuber using AI for scriptwriting might hire a human editor to polish the final cut. This is a shift from “jobs” to “gigs,” but it is still employment.
Third, the AI agent economy is nascent but growing rapidly. Companies like Cognition Labs (creator of Devin, the AI software engineer) and Adept (building AI agents for enterprise workflows) are hiring agent operators and workflow designers. The market for AI agents is projected to reach $30 billion by 2028, according to industry analysts. This will create roles like “agent supervisor” who monitors multiple AI agents, handles exceptions, and retrains models.
Fourth, education and training is a new sector. Coursera, Udacity, and LinkedIn Learning report massive demand for AI-related courses. NVIDIA’s own Deep Learning Institute has trained over 500,000 developers. This creates jobs for instructors, curriculum designers, and certification evaluators.
| Sector | 2023 Market Size | 2028 Projected Size | CAGR | New Job Roles |
|---|---|---|---|---|
| AI Data Services | $8B | $25B | 25% | Data annotator, curator, QA |
| AI Agent Platforms | $2B | $30B | 72% | Agent operator, workflow architect |
| AI Education | $5B | $20B | 32% | Instructor, curriculum designer |
| AI Hardware | $53B | $227B | 18% | Chip designer, fab engineer |
Data Takeaway: The fastest-growing sector is AI agents, with a 72% CAGR, suggesting that the most significant job creation will come from managing and orchestrating autonomous systems, not just building them.
Risks, Limitations & Open Questions
Huang’s optimism is not without caveats. First, skill mismatch is a critical risk. Many displaced workers—e.g., call center agents, translators, graphic designers—may not easily transition to roles like “prompt engineer” or “GPU architect.” The new jobs often require technical literacy that the displaced workforce lacks. Retraining programs are slow and uneven.
Second, job quality matters. Many new AI jobs, particularly in data annotation and RLHF, are low-paid, precarious contract work. Scale AI’s contractors, for example, earn as little as $10 per hour in some regions. This is not the high-quality employment Huang envisions. The gig economy model may replicate existing inequalities.
Third, automation of the new jobs is a looming threat. As AI improves, even the new roles may be automated. Prompt engineering, for instance, is already being replaced by automated prompt optimization tools like DSPy. Agent operators may be replaced by self-healing AI agents. The half-life of these new jobs could be short.
Fourth, geographic concentration is a concern. Most high-value AI jobs are concentrated in San Francisco, New York, and a few global hubs. This exacerbates regional inequality. Countries without strong tech ecosystems may see job losses without corresponding gains.
Finally, ethical questions around AI-generated content and deepfakes could lead to regulatory backlash, potentially slowing job creation. If governments impose strict liability on AI outputs, companies may hire fewer humans to avoid legal risk.
AINews Verdict & Predictions
Jensen Huang is right in the aggregate: AI is creating more jobs than it destroys, and the trend will continue for the next 5–10 years. However, the distribution is uneven. The winners are highly skilled workers in tech hubs; the losers are low-skill workers in routine cognitive tasks. The policy challenge is not to stop AI but to manage the transition.
Our predictions:
1. By 2027, the role of “AI agent operator” will be one of the fastest-growing job categories, with over 1 million positions globally. Companies will hire dedicated teams to manage AI agents in customer service, software development, and logistics.
2. By 2028, the data annotation market will peak and then decline as synthetic data and self-supervised learning reduce the need for human labeling. The next wave of job creation will be in model alignment and safety, not data prep.
3. NVIDIA’s own workforce will surpass 50,000 employees by 2026, driven by demand for AI infrastructure. The company will become a bellwether for the AI job market.
4. The gig economy will expand as AI enables solo entrepreneurs. Platforms like Upwork and Fiverr will see a surge in AI-related project postings, but wages for basic tasks will fall, creating a two-tier market.
5. Regulation will be the wildcard. If the EU’s AI Act imposes strict requirements on high-risk AI systems, it could create a new compliance industry—lawyers, auditors, and ethicists—but also slow deployment.
What to watch: The next major indicator will be the employment reports from AI companies in Q3 2025. If hiring continues at the current pace, Huang’s thesis is validated. If hiring slows, the narrative may shift. For now, the data supports the optimist’s view: AI is not the end of work, but the beginning of a new kind of work.