The AI economy produced scale without a clean productivity verdict. Alphabet's cloud growth arrived with negative free cash flow, while Google's usage study found broad but shallow workplace adoption. Water constrained a proposed data center, local time savings faded before aggregate output, synthetic books crowded a market, and Uber linked support cuts to AI without disclosing a service design.
1. Alphabet's AI buildout pushes quarterly free cash flow below zero
Alphabet reported $119.8 billion in second-quarter revenue, including $24.8 billion from Google Cloud. Operating cash flow was about $39.1 billion, while capital expenditure reached $44.9 billion, producing negative free cash flow of roughly $5.8 billion. The company also raised its 2026 infrastructure spending ceiling from $190 billion to $205 billion.
Alphabet remains profitable and holds more than $100 billion in cash, so one negative free-cash-flow quarter is not a solvency problem. It is evidence that AI has altered the timing of returns: cloud demand is growing while data-center outlays arrive first. Future quarters must distinguish durable AI revenue from capacity purchased ahead of uncertain utilization.
Sources: Ars Technica on Alphabet's second-quarter cash flow · Alphabet's second-quarter earnings release
2. Google's ATLAS finds workplace AI adoption is broad but task automation is rare
Google's first ATLAS report analyzes 15 million aggregated, de-identified interactions across the Gemini app, AI Mode, and Gemini API. The sample covers more than 150 countries, 140 languages, 800 occupations, and 4,000 tasks. Workplace use appeared in 68% of occupations representing 90% of US employment, but touched only about 21% of tasks in a typical job.
Fewer than 10% of workplace interactions fully automated a task, and 86% of all sampled interactions occurred outside work. Those findings favor assistance over wholesale job execution. ATLAS observes only selected Google products, excludes major surfaces such as Workspace and Cloud, and measures interactions, not productivity outcomes, so it maps usage without establishing causal economic impact.
Sources: Google's overview of the first AI and Economy ATLAS
3. OpenAI's proposed Sydney data center abandons recycled-water cooling
NEXTDC dropped a plan to cool the proposed 612-megawatt S7 data center with recycled water because permission was unavailable for the required pipeline. The OpenAI partner now plans to circulate liquid coolant around chips and reject heat with fans, avoiding drinking water but using a method that outside specialists told Reuters would consume more energy.
The facility is still proposed, and NEXTDC did not provide a target power-usage effectiveness figure. Sydney's grid operator already reports limited capacity for large new data-center connections. The change exposes a local infrastructure tradeoff: eliminating operational water demand can increase electricity demand when recycled-water networks and transmission upgrades do not arrive on the same schedule as compute projects.
Sources: Reuters investigation of the S7 cooling change
4. Stripe's economist finds task-level AI gains have not reached aggregate productivity
Stripe economist Ernie Tedeschi synthesized studies reporting faster writing, consulting, customer service, and email work alongside US labor-productivity growth of roughly 2.5% over the past year. Yet estimated total-factor-productivity growth remained near zero, and the relationship between industry AI adoption and recent productivity disappeared after accounting for pre-pandemic sector trends.
The analysis argues that AI accelerates bounded tasks while review, integration, incentives, and release processes remain bottlenecks. It is an economic interpretation, not a causal estimate of AI's national contribution, and many cited experiments use older models. Firms can record local time savings without greater shipped output unless workflows redistribute the saved time toward whatever constrains production next.
Sources: Stripe Economics on AI and productivity evidence · Bureau of Labor Statistics productivity data
5. A preprint measures AI-generated fiction gaining sales through volume
A July 22 preprint applied full-text AI detection to 14,419 self-published genre-fiction books sold on Amazon from 2023 through June 2026 and matched them to daily sales. Books with more than 25% detected AI text held a smaller share of sales than catalog listings but gained share over time and occupied more scarce top-rank positions.
The number of books recording quarterly sales grew 19.2-fold while quarterly revenue grew 8.9-fold, consistent with more titles dividing a slower-growing pool. This is a working paper, its Amazon sample is not the entire book market, and detector classifications cannot prove a title's production history. The evidence supports a scale effect, not a finding that AI text caused every displacement.
Sources: Version one of the generative-AI book-market preprint
6. Uber cuts 10% of its customer-service organization while invoking AI
Uber cut 10% of Community Operations, its global customer-support organization, and required some remote staff to move to hub offices. A company spokesperson said the changes would simplify operations, strengthen in-person collaboration, and continue the company's embrace of AI. Uber did not disclose the number of affected workers or the support functions AI would assume.
The missing operating detail prevents the reorganization from demonstrating that automation can resolve customer problems safely or cheaply. Support quality turns on escalation rates, language coverage, refunds, fraud handling, and resolution time, not headcount alone. The disclosed facts establish an AI-linked labor decision while leaving its service design and measured productivity unproven.
Sources: Bloomberg on Uber's Community Operations cuts · Engadget on the layoffs and return-to-office change