Abstract: This study examines how firms' adoption of industrial robots reshapes their value-added tax (VAT) burden under China's credit-invoice system,explaining why automation can both reduce and raise VAT over time. We propose and test a timing-based mechanism—“credit first,leverage later”. When a firm purchases robots,the equipment arrives with substantial input VAT that is immediately creditable,easing the net VAT burden in the short run. As robots are integrated,efficiency improves,capacity expands,and taxable sales grow; output VAT then rises faster than routine input credits from ordinary purchases,producing a rebound of the VAT burden in subsequent years. Whether robotization ultimately lowers or raises VAT depends on the relative speed and strength of these two forces. We build a large firm-level panel linking markers of robot adoption—identified from standardized product codes in customs and procurement records—to the Annual Tax Survey (ATS),China’s official micro database with invoice-based measures of output VAT,input VAT,and net VAT,plus detailed financials. This construction supports multiple definitions of “VAT burden”, including theoretical liabilities based on invoices and actual VAT paid relative to sales. Tracking firms around their first recorded robot purchase,our empirical strategy leverages staggered adoption across otherwise similar firms,includes firm and time fixed effects,anchors dynamics with event-time profiles,and probes credibility with design-consistent checks. Three core results emerge. First,in the purchase year,the net VAT burden falls meaningfully across normalizations—using theoretical VAT over sales,actual VAT paid over sales,or invoice-based netting. Flat pre-trends in event time support identification. Second,the decline is short-lived:in the first and second years after adoption， the VAT burden rebounds and often overshoots pre-adoption levels,aligning with rapid growth in taxable sales as production scales. Third,invoice-flow decomposition clarifies the mechanism:purchase-year input VAT credits jump due to large,lumpy capital buys,while output VAT normalized by sales barely moves,consistent with integration lags; after adoption,output VAT rises with sales while incremental input credits revert to routine levels,yielding the predicted dip-and-rebound path. Heterogeneity reveals where each force is strongest. Purchase-year easing is larger for higher-priced or precision-grade robots (bigger embedded credits,longer integration). Firms with tighter liquidity constraints see more pronounced short-run easing,underscoring cash-flow salience of timely input-VAT crediting. Financially stronger firms transition faster to the leverage phase,showing earlier rebounds. Regional capabilities matter:thinner knowledge ecosystems (lower research intensity or weaker global value-chain exposure) exhibit larger immediate relief and slower returns to baseline,indicating longer absorption lags. Complementary real-outcome evidence supports leverage:post-adoption,asset turnover,capacity-utilization proxies,and total factor productivity rise;employment,wage bills,and patenting increase with a lag. Robots are part of a broader reorganization that expands taxable activity and thickens the VAT base. The paper contributes along three dimensions. First,it brings indirect taxation to the center of the automation debate. Most existing work focuses on corporate income tax incidences or on aggregate fiscal effects. By exploiting invoice-level mechanics inside a credit-invoice regime,we provide micro evidence on how technology adoption maps into VAT flows,which are central to public finance in China and many other economies. Second,the timing framework'credit first,leverage later'reconciles claims that automation both eases and tightens tax burdens. Both statements can be true,but they refer to different horizons governed by accounting rules and real adjustment dynamics. Third,we highlight how technology complexity,financial frictions,and local capability endowments shape the path from purchase to performance,thereby explaining cross-sectional variation in the magnitude and speed of the rebound. Policy implications are direct. In the short run,administrative frictions that delay recognition of input-VAT credits can magnify financing stress exactly when firms are making lumpy frontier investments. Streamlined invoice processing,prompt and predictable crediting,and clear documentation standards can reduce those stresses and support diffusion. In the medium run,the observed rebound in VAT burden should be interpreted as the fiscal shadow of successful scaling rather than a policy failure. As firms integrate robots,taxable sales expand and the VAT base becomes broader and more resilient. Complementary policies that compress absorption lags support for systems integration,workforce upskilling,and process redesign can accelerate the transition from purchase-year relief to productivity-driven base expansion. For tax administrators,recognizing the predictable dip-and-rebound profile around adoption events can improve the targeting of taxpayer services,facilitate risk monitoring,and guide communication with stakeholders who may otherwise misread short-term fluctuations in filings. In sum,robot adoption triggers immediate relief via input-VAT credits,followed by integration-driven growth that lifts output VAT;understanding this temporal interplay is essential for evaluating tax consequences and designing policies that catalyze high-quality investment while sustaining a robust VAT base.