CSP financing capacity is sufficient to sustain AI capex, with semiconductor momentum expected to continue through FY28e
AI summary card
CSP financing capacity is sufficient to sustain AI capex, with semiconductor momentum expected to continue through FY28e
Even if free cash flow comes under pressure from FY27 onward, large CSPs can still expand capital expenditure through explainable debt financing and continue to support AI semiconductor demand while keeping ROIC at acceptable levels.
- The market is concerned that CSP capital expenditure has peaked, but HSBC believes financing constraints are overstated.
- FY27e scenario capital expenditure is USD1.0-1.6trn, corresponding to a net debt-to-equity ratio of 8%-30%.
- FY28e scenario capital expenditure is USD1.9-2.5trn, corresponding to a net debt-to-equity ratio of 25%-43%.
- Under the base case, aggregate CSP ROIC is expected to decline from 22% to 19% in FY27e and from 21% to 17% in FY28e, but remains at acceptable levels.
- Preferred AI semiconductor names are Marvell, Intel, TSMC and ASML.
Report interpretation
Overview
The report assesses whether large cloud service providers can continue financing AI infrastructure investment against a backdrop of weakening free cash flow. HSBC believes Alphabet, Amazon, Microsoft, Meta and Oracle still have significant debt capacity; even if capital expenditure exceeds consensus expectations, group leverage and ROIC would not deteriorate to unsustainable levels. Therefore, recent market concerns about a peak in CSP capital expenditure and a downturn in semiconductor earnings may be excessive.
Core views
The CSP capital expenditure trend remains intact and can continue to support AI demand and semiconductor supply chain earnings. Consensus expectations show capex growth slowing from 95% in FY26e to 46% in FY27e and 11% in FY28e, with group free cash flow potentially turning negative in FY27e; however, HSBC’s scenario analysis indicates that CSPs can materially increase capital expenditure through incremental debt. Even under higher capex assumptions, aggregate ROIC in FY27e and FY28e can still be maintained at approximately 19% and 17%, respectively. Recent increases in capex or capacity plans by TSMC, Intel and ASML also provide industry validation for demand continuing through FY28e.
Analysis framework
The report combines Visible Alpha consensus expectations, company disclosures and HSBC forecasts. It first compares CSP free cash flow and capital expenditure paths, then constructs three FY27e and FY28e capital expenditure scenarios using operating cash flow plus incremental debt, and calculates the net debt-to-equity ratio. It then compares aggregate ROIC under consensus expectations and scenario assumptions, and assesses the sustainability of industry chain earnings based on semiconductor companies’ capital expenditure, capacity planning and target prices.
Methodology notes
Use operating cash flow and incremental debt to cover potential capital expenditure
The report sets three capital expenditure levels for FY27e and FY28e, assuming the portion exceeding operating cash flow is financed by new debt, and uses this to calculate the required incremental debt and net debt-to-equity ratio.
Assess the impact of capital expenditure expansion on return on invested capital
Because CSPs do not fully disclose standalone AI revenue, the report uses the company-level aggregate ROIC of Alphabet, Amazon, Microsoft, Meta and Oracle to measure capital efficiency.
Use supply chain capital expenditure and capacity expansion plans to validate end demand
The report references the capital expenditure and capacity expansion plans disclosed by TSMC, Intel and ASML in 2Q26 earnings to validate visibility into AI infrastructure demand continuing through FY28e.
Derive the target price using forecast earnings per share and a target price-to-earnings multiple
Taking TSMC as an example, HSBC applies a 25x target P/E multiple to FY27e EPS of TWD137.29 to derive a target price of TWD3,400.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Alphabet, Amazon, Microsoft, Meta, OracleMain financing and demand entities for AI capital expenditure
- Strengths
- The groups have large operating cash flow scale, their overall balance sheets still have room to increase leverage, and AI infrastructure investment can obtain continued financing.
- Weaknesses
- Combined FY27e free cash flow is expected to fall to -USD104bn, and further capex expansion will increase reliance on external financing and depress ROIC.
- Comparison
- Consensus capex growth is expected to slow to 11% in FY28e, while HSBC’s scenarios show that debt financing could enable a significantly higher level of capital expenditure.
- Risks
- AI commercialization slower than expected, insufficient infrastructure utilization, rising debt costs, weaker capital expenditure discipline, and ROIC below scenario assumptions.
- TSMC (2330 TT)Preferred AI semiconductor name and core supplier of advanced processes
- Strengths
- FY26e capital expenditure guidance was raised to USD60-64bn, and 2nm capacity may grow at a CAGR of over 70% in FY26-28, reflecting confidence in long-term AI demand and pricing power.
- Weaknesses
- Advanced process capacity expansion will bring higher depreciation costs and depends on new capacity ramp-up and yield improvement.
- Comparison
- Current price TWD2,380, target price TWD3,400, Buy rating, implied upside of approximately 43%.
- Risks
- 5nm orders weaker than expected, escalation of US-China trade tensions and export restrictions, advanced process ramp-up or yields slower than expected, and depreciation costs higher than expected.
- Intel (INTC US)Preferred AI semiconductor name and beneficiary of server CPU demand
- Strengths
- Server CPU demand is accelerating due to agentic AI, FY26e capital expenditure guidance was raised to above USD20bn, and FY27e is expected to continue increasing.
- Weaknesses
- Higher capital expenditure increases execution and capital return pressure, and earnings delivery depends on sustained demand and progress in foundry operations.
- Comparison
- Current price USD97.52, target price USD200, Buy rating, implied upside of approximately 105%.
- Risks
- AI demand or server CPU supply-demand improvement falls short of expectations, capacity expansion execution delays, and capital efficiency declines.
- Marvell (MRVL US)Preferred AI semiconductor name
- Strengths
- The report expects company-specific drivers to push FY27e results to continue exceeding market expectations.
- Weaknesses
- Earnings growth is relatively sensitive to sustained CSP investment and AI infrastructure demand.
- Comparison
- Current price USD208.56, target price USD300, Buy rating, implied upside of approximately 44%.
- Risks
- CSP capital expenditure slows, AI supply chain orders fall short of expectations, and earnings growth fails to materialize.
- ASML (ASML NA)Preferred AI semiconductor name and supplier of advanced lithography equipment
- Strengths
- Plans to increase EUV capacity by 30% in FY27e and is studying a further 30% increase in FY28e, indicating strong customer demand for capacity expansion.
- Weaknesses
- Growth depends on customers’ advanced process capital expenditure and capacity expansion plans being executed on schedule.
- Comparison
- Current price USD1,513.80, target price USD2,149, Buy rating, implied upside of approximately 42%.
- Risks
- Customer capital expenditure is delayed, EUV capacity expansion execution falls short of expectations, and trade and equipment export restrictions escalate.
Key data
- CSP consensus capital expenditureFY26e USD731bn; FY27e approximately USD1.1trn; FY28e approximately USD1.2trnCorresponding year-on-year growth rates are approximately 95%, 46% and 11%, respectively.
- CSP free cash flowFY26e USD34bn; FY27e -USD104bn; FY28e USD18bnCovers Alphabet, Amazon, Microsoft, Meta and Oracle.
- FY27e capital expenditure scenariosUSD1.0-1.6trnRequires new debt of USD38-638bn, with the net debt-to-equity ratio rising to 8%-30%, representing growth of 37%-119% versus FY26 consensus expectations.
- FY28e capital expenditure scenariosUSD1.9-2.5trnRequires new debt of USD698bn-1.298trn, with the net debt-to-equity ratio rising to 25%-43%; growth of 46%-92% relative to the FY27e base-case scenario.
- Consensus aggregate CSP ROICFY24 35%; FY25 29%; FY26e 22%; FY27e 22%; FY28e 21%Capital expenditure growth leads to a decline in returns, but consensus expectations still indicate relatively high capital efficiency.
- Base-case aggregate CSP ROICFY27e 19%; FY28e 17%Assumes FY27e and FY28e capital expenditure of USD1.3trn and USD2.2trn, respectively.
- TSMC capital expenditure guidanceFY26e USD60-64bnPreviously at the high end of the USD52-56bn range; also announced an additional USD100bn investment in Arizona, United States.
- Intel capital expenditure guidanceFY26e above USD20bnPrevious guidance was USD17.7bn, and the company expects FY27e capital expenditure to continue increasing.
- ASML EUV capacity planFY27e up 30%, and studying another 30% increase in FY28eCorresponds to FY28e low-NA EUV capacity of about 110 units, above the previous consensus expectation of 89 units.
- Recent market performanceThe SOX Index has fallen 16% since July 1, 2026, while the S&P 500 has risen 3% over the same periodReflects market concerns about AI infrastructure oversupply and a peak in CSP capital expenditure.
Impact & implications
If CSPs maintain or increase capital expenditure through the debt market, order and earnings visibility for AI server, advanced process, wafer foundry and lithography equipment suppliers is expected to extend through FY28e. The report believes the relative pullback in the SOX Index has already reflected substantial concerns about financing and demand peaking, while CSP balance sheets and capital returns can still support the investment cycle. However, the investment view still requires ongoing validation of AI commercialization, asset utilization, debt costs and actual company capital expenditure execution.
Risks
- CSP AI revenue and commercialization progress are slower than expected, resulting in insufficient returns on new infrastructure.
- Persistent cloud infrastructure oversupply or declining utilization prompts CSPs to cut capital expenditure.
- Incremental debt scale or financing costs are higher than assumed, causing the net debt-to-equity ratio and interest burden to rise rapidly.
- Aggregate CSP ROIC falls below the 17%-19% scenario level, weakening willingness to continue investing.
- Semiconductor capacity expansion outpaces actual demand, creating pressure on inventory, pricing and capacity utilization.
- US-China trade friction, export controls or economic sanctions escalate, affecting semiconductor equipment and advanced process supply chains.
- TSMC, Intel or ASML capacity expansion, process ramp-up and delivery progress fall short of expectations.
- Market valuations already reflect strong growth expectations, and earnings misses could trigger significant volatility.
What to watch
- Quarterly capital expenditure guidance from Alphabet, Amazon, Microsoft, Meta and Oracle and changes in FY27 free cash flow.
- Whether CSP new debt issuance scale, financing costs and net debt-to-equity ratios approach HSBC’s scenario ranges.
- Whether aggregate and individual CSP ROIC can remain at the acceptable levels estimated in the report.
- AI infrastructure utilization, agentic AI demand and related revenue commercialization progress.
- TSMC’s FY26 capital expenditure, 2nm capacity growth and progress on additional investment in Arizona.
- Intel FY27 capital expenditure and server CPU demand realization.
- ASML FY27-FY28 EUV capacity expansion plans, orders and customer delivery pace.
- Performance of the SOX Index relative to the S&P 500 and changes in market expectations for the AI capital expenditure cycle.