Interview, Fireside Chat
Inside The $2.2B AI Research Accelerator | Turing
- Financial Performance & Valuation: Turing reports over $300 million in revenue, profitability, and $225 million in total funding, with a last valuation of $2.2 billion.
- Customer Base: The company serves 7 to 8 out of 9 frontier labs, including OpenAI, NVIDIA, Anthropic, Google, Microsoft, Meta, Salesforce, and Amazon.
- Network Scale: Turing operates a global network of 4 million engineers utilized to generate high-quality data for AI training.
- Market Opportunity: The company targets a $30 trillion addressable market representing the automation of global knowledge work across four dimensions: industry, function, role, and workflow.
- Strategic Pivot: The industry has shifted from commodity "sweatshop data labeling" to a demand for expert human data and "strategic research acceleration" to advance superintelligence.
- Core Value Proposition: Turing differentiates itself not just as a talent marketplace or data factory, but as a research-first partner that proactively engineers data to break models and improve performance in coding, reasoning, and multi-modality.
- Targeted AI Pillars: The company focuses on developing data to advance four specific pillars of superintelligence: multi-modality, reasoning, tool use, and coding.
- ASI Definition: Turing defines Artificial Superintelligence (ASI) as the point where AI automates 90% of the tasks performed by 90% of humans in knowledge work.
- Competitive Landscape: Founders identify the primary competitors as pure-play data labelers (e.g., Scale AI) and talent marketplaces (e.g., Mercore, Surge), positioning Turing as the only entity excelling at both smart human sourcing and proactive data research.
- Revenue Sustainability: Unlike traditional service-based data firms, Turing argues for durable revenue by acting as "picks and shovels" for the AGI era, serving both the scaling frontier labs and enterprises building proprietary intelligence.
- Model Evolution & Data Scarcity: Pre-training data (internet corpus) is largely exhausted; future scaling depends on generating hard, domain-specific expert data (STEM, healthcare, finance) that is not available on the open web.
- Training Methodology: Frontier model training now involves a shift from simple Supervised Fine-Tuning (SFT) and RLHF to Reinforcement Learning in verifiable domains (coding, math), where models learn through self-play and automated verification rather than human imitation.
- GPT-5 Perspective: Founder Jonathan views GPT-5 as "awesome" and rejects the "rapid takeoff" narrative, predicting instead a steady, continuous progression driven by scaling laws for compute and data.
- Safety Narrative Shift: The probability of AI research pausing for safety reasons has dropped to approximately 9% (down from ~40%), as the industry adopts "human-in-the-loop" partial autonomy to mitigate risks.
- Enterprise Strategy: Turing advises enterprises to build "proprietary intelligence" by fine-tuning smaller models (0.5B–10B parameters) on internal unstructured data to create secure, domain-specific moats, rather than relying on general-purpose SaaS.
- Future Outlook: The company expects to automate and replace traditional consulting and services firms (e.g., McKinsey, Accenture) by deploying agentic systems that handle complex enterprise workflows.
- Adoption Risks: The primary risk identified is not AI safety, but the failure of enterprises to adopt AI quickly enough, citing a 95% failure rate for GenAI pilots due to data and expertise gaps.
- Research Roadmap: Turing's internal R&D focuses on "Horizon One" (immediate high-quality data generation) and "Horizon Two" (predicting data needs three months out), currently focusing on embodied AI and robotics benchmarks.
- Recruitment Criteria: To win in this space, Turing argues companies must excel at two axes: finding smart humans rapidly and conducting frontier research to understand what data will most benefit model advancement.