Goldman Sachs

AI Adoption, China Self-Reliance, and Talent Risks

A July 5, 2026 Exponential View newsletter reviews new research and data on AI’s labor effects, geopolitics of AI model development, and supply-chain risks. Ramp and Revelio Labs’ analysis of over 21,000 U.S. firms finds high-AI adopters increased overall employment by ~10% over two years, with entry-level roles rising ~12%. Papers argue U.S. chip export controls have pushed Chinese developers toward open-source, locally runnable model infrastructure (raising GitHub forking among China-linked developers), while other research documents growing Chinese self-reliance in the science underpinning domestic patents. The newsletter also notes sectoral concerns such as “never skilling” in medicine and global tungsten supply concentration (China mines ~80%). The piece synthesizes data-driven findings and academic papers to highlight nuanced, mixed impacts of AI adoption.

Goldman Sachs Names Undervalued Stocks to Outperform

Goldman Sachs identified several stocks it considers undervalued and likely to outperform in July, highlighting O’Reilly Automotive, NetEase, Tradeweb and Liftoff Mobile. Analysts at Goldman argued O’Reilly’s store positioning and resilient comps reduce demand volatility; NetEase was described as a resilient, non-AI-focused compounder with potential margin expansion and a new game launch; Tradeweb was upgraded to buy from neutral despite revenue-growth and tokenization concerns; and Liftoff Mobile was noted for scaling its Cortex AI platform and exposure to in‑app advertising and AI-driven automation across the ad tech value chain. The article was published July 5, 2026 on CNBC.

Data Engineering Take-Home Tests Become 20-Hour Unpaid Work

An opinion piece argues that data-engineering take-home assignments have grown from short, bounded exercises into unpaid consulting projects that routinely demand 10–20 hours of candidate time. The author cites industry statistics showing widespread use and scope creep of take-homes, rising AI use during assessments, inconsistent or unenforced AI bans, and a pervasive lack of post-rejection feedback. Companies that have reduced take-home scope in favor of live debugging, pair-programming, or bounded exercises report better signal and completion rates. The article warns the current hiring practice harms candidates (mental-health impacts, opportunity cost, reduced diversity) and calls for time-bounded assessments, feedback, and interviewer changes that evaluate engineering judgment rather than long unpaid deliverables.