China AI Capex Statistics 2026
By Axis Intelligence Research
Co-authors: Mia Scarlett & Sarah Mitchell | Last updated: September 1, 2026 | License: CC BY 4.0
China’s three listed AI platform majors spent RMB131.9 billion (US$19.4 billion) on capital expenditure in the June 2026 quarter alone — 114.1% more than the same quarter a year earlier. Every yuan of that figure traces to a filed results announcement. The state telecom carriers spent differently, and that difference is the story most coverage misses.
Quick Answer
Alibaba, Tencent and Baidu reported combined capital expenditure of RMB131,852 million (US$19.4 billion) for the quarter ended 30 June 2026, up 114.1% year-over-year from RMB61,583 million, according to Axis Intelligence Research’s aggregation of the three companies’ filed results. Alibaba accounted for 51.3% of that total. China’s national intelligent computing capacity reached 2,185 EFLOPS by end-June 2026, with nationwide computing facility utilisation at 71.4%.
Key Findings
- Axis Intelligence Research finds that Alibaba, Tencent and Baidu together reported RMB131,852 million (US$19.4 billion) of capital expenditure in the quarter ended 30 June 2026, a 114.1% increase over the RMB61,583 million the same three companies reported for the June 2025 quarter.
- Alibaba’s June 2026 quarter capital expenditure of RMB67,678 million (US$9,975 million) equalled 53.7% of everything the company spent across its entire fiscal year to 31 March 2026 (RMB126,063 million), per its results announcement filed with the SEC.
- Baidu’s capital expenditure rose 199.7% year-over-year and 92.5% quarter-over-quarter to RMB11,390 million, the steepest sequential acceleration of the three, while its AI Cloud Infra revenue reached RMB7.3 billion.
- Axis Intelligence Research’s China AI Investment Strain Index (CAIS™) reads 60.4 for Baidu, 44.3 for Alibaba and 23.6 for Tencent as of the June 2026 quarter — a 36.8-point spread between the most and least strained balance sheet in the same market, in the same quarter.
- China Mobile did not raise capital expenditure to fund AI — it reallocated. Group capex rose just 4.5% to RMB61.0 billion in the first half of 2026, but computing-network and AI-network spending together climbed to 31.6% of the total from roughly 19.0% a year earlier, a 12.6 percentage-point shift funded by a 9.8% cut to communications networks.
How Much Are Chinese Tech Companies Spending on AI Infrastructure in 2026?
Three Chinese companies disclose AI-driven capital expenditure on a comparable quarterly basis and file the numbers with a regulator. That is the entire population. Everything else circulating as a “China AI capex” figure is either an estimate, a multi-year pledge, or a private company’s unaudited briefing.
For the quarter ended 30 June 2026, those three reported:
| Company | Jun-Q 2026 capex (RMB m) | Jun-Q 2025 capex (RMB m) | YoY | Share of total | Source |
|---|---|---|---|---|---|
| Alibaba Group | 67,678 | 38,676 | +75.0% | 51.3% | June Quarter 2026 Results, Form 6-K exhibit |
| Tencent Holdings | 52,784 | 19,107 | +176.3% | 40.0% | 2026 Second Quarter Results announcement |
| Baidu, Inc. | 11,390 | 3,800 | +199.7% | 8.6% | Second Quarter 2026 Results, Form 6-K exhibit |
| Platform-3 total | 131,852 | 61,583 | +114.1% | 100% | Axis Intelligence Research aggregation |
US$ conversion at RMB6.7851 to US$1.00, the 30 June 2026 rate in the Federal Reserve H.10 release, used by both Alibaba and Baidu in their own filings. Sources: company results announcements as cited; totals and percentages calculated by Axis Intelligence Research.
Alibaba’s disclosure is the most explicit about what the money buys. The company attributed its 75% jump to procurement-cycle timing, added CPU-compute capacity for anticipated AI agent adoption, and higher pricing across a broad range of chip components. That third clause is worth pausing on: part of the capex increase is unit cost, not unit volume. A capex line that rises because silicon got more expensive buys less compute than a capex line that rises because more racks went in, and no Chinese platform currently discloses the split.
Mia Scarlett’s read: the headline everyone lifted was “Alibaba capex up 75%.” The number in the filing that reframes the quarter is different. Alibaba spent RMB67,678 million in three months against RMB126,063 million across the whole twelve months to 31 March 2026 — one quarter consuming 53.7% of a full prior year. Annualise the June run-rate and the fiscal 2027 figure lands near double fiscal 2026. That is not an incremental build; it is a re-baselining of the company’s cost structure, and the free cash flow line already shows it, swinging to an outflow of RMB44,670 million from an outflow of RMB18,815 million a year earlier.
The three-year pledge, marked to market
Alibaba’s RMB380 billion three-year AI and cloud commitment (US$56.0 billion at the June 2026 rate) is the most-quoted forward number in Chinese AI. Management stated cumulative spending under the plan had reached RMB190 billion through June 2026 — exactly 50.0% of the programme. Axis Intelligence Research treats pledges and disbursements as separate rows in the dataset, because they behave differently: a pledge is a communication, a disbursement is a cash flow, and only one of them shows up in free cash flow.
The China AI Investment Strain Index (CAIS™) — Axis Intelligence Research
Capex totals tell you who is spending. They do not tell you who can afford it. CAIS™ — the China AI Investment Strain Index — measures how hard a company’s AI build-out is pressing against the cash and revenue it is meant to generate. The scale runs 0 to 100; higher means more strain. A reading near zero means the quarter’s capex was covered by the quarter’s operating cash flow, grew moderately, and is matched by disclosed AI revenue. A reading above 60 means the opposite on all three counts.
CAIS™ is not a solvency measure and not a verdict on strategy. Alibaba, Tencent and Baidu all hold substantial liquidity. It is a measure of how much of the AI bet is currently being financed by something other than the business it is being built for.
Components and weights
- S1 — Cash coverage strain (45%) = 100 × (1 − operating cash flow ÷ capex), floored at 0, capped at 100
- S2 — Acceleration strain (30%) = year-over-year capex growth in percentage points ÷ 3, capped at 100
- S3 — Monetisation strain (25%) = 100 × (1 − disclosed AI/cloud segment revenue ÷ capex), floored at 0
CAIS™ = (S1 × 0.45) + (S2 × 0.30) + (S3 × 0.25)
CAIS™ readings, quarter ended 30 June 2026
| Company | S1 cash coverage | S2 acceleration | S3 monetisation | Weighted | CAIS™ |
|---|---|---|---|---|---|
| Baidu, Inc. | 69.8 | 66.6 | 35.9 | 31.42 + 19.97 + 8.98 | 60.4 |
| Alibaba Group | 66.1 | 25.0 | 28.4 | 29.74 + 7.50 + 7.11 | 44.3 |
| Tencent Holdings | 0.1 | 58.8 | not computable | 0.08 + 23.50 | 23.6 |
Inputs: operating cash flow — Alibaba RMB22,945m, Tencent RMB52.7bn, Baidu RMB3,436m. Capex as tabled above. Segment revenue — Alibaba AI Cloud and Compute Services RMB48,437m; Baidu AI Cloud Infra RMB7.3bn. Baseline reading, snapshot date 30 June 2026. Cite as: Axis Intelligence Research, China AI Investment Strain Index (CAIS™), September 2026, axis-intelligence.com/china-ai-capex-statistics/
A disclosed refusal. Tencent does not report cloud revenue as a separate line. Its cloud business sits inside FinTech and Business Services, which totalled RMB60,286 million in the quarter alongside commercial payment, wealth management and consumer loan revenue. Axis Intelligence Research will not estimate a cloud subtotal in order to complete the third component. Tencent’s CAIS™ is therefore computed on the two available components at reweighted 60/40, and is flagged as such wherever it appears. An index that fills a disclosure gap with a guess stops being an index.
What the spread means
The 36.8-point range between Baidu and Tencent is the most useful thing in the table, and it exists because the three companies are financing the same bet from three different positions.
Tencent scores 0.1 on cash coverage because operating cash flow of RMB52.7 billion very nearly matched capex of RMB52,784 million — a coverage ratio of 99.8%. It scores 58.8 on acceleration because that capex nearly tripled. The pattern is a company with enough cash generation to absorb a tripling without leaving the operating business, at least for one quarter. The pressure shows up elsewhere: free cash flow was negative RMB13.8 billion, and Tencent stated it would have been positive RMB37.6 billion excluding prepayments for compute procurement. Non-current prepayments, deposits and other assets on the balance sheet moved from RMB24,540 million at 31 December 2025 to RMB91,203 million at 30 June 2026. Roughly RMB67 billion of compute has been paid for and not yet delivered. That is a forward capex commitment sitting in an asset line, and it will convert.
Baidu scores highest because all three components point the same way. Operating cash flow of RMB3,436 million covered 30.2% of an RMB11,390 million capex bill; capex tripled year-over-year and nearly doubled sequentially; and AI Cloud Infra revenue of RMB7.3 billion, up 50% year-over-year with GPU Cloud up 283%, covered 64.1% of the spend. Baidu is building the fastest relative to its own size while its online marketing business declines 19% year-over-year. The strain reading is not a prediction of failure — GPU Cloud growing 283% off a rising base is the strongest AI revenue signal any of the three disclosed. It is a statement that Baidu has the least margin for the build taking longer than planned.
How Do China’s State Telecom Carriers Fund AI Compute?
This is the section that separates China’s AI capex structure from the American one, and it is consistently underweighted in English-language coverage because carrier filings are duller reading than platform filings.
China Mobile’s first-half 2026 capital expenditure was RMB61.0 billion, up 4.5% from RMB58.4 billion. On the surface, nothing happened. Inside the number, the entire composition changed.
| China Mobile 1H 2026 capex line | RMB bn | YoY | Share of group capex |
|---|---|---|---|
| Communications networks | 37.9 | −9.8% | 62.1% |
| Computing networks | 15.6 | +71.3% | 25.6% |
| AI networks | 3.7 | +85.2% | 6.1% |
| Unallocated / other | 3.8 | — | 6.2% |
| Total capital expenditure | 61.0 | +4.5% | 100% |
Source: China Mobile Limited, 2026 Interim Results presentation and press release, 13 August 2026. Share-of-total and unallocated residual calculated by Axis Intelligence Research.
Applying the disclosed growth rates backwards puts computing plus AI network spend at approximately RMB11.1 billion in the first half of 2025, or 19.0% of that period’s RMB58.4 billion. In the first half of 2026 the same two lines reached RMB19.3 billion, or 31.6%. Axis Intelligence Research finds that China Mobile shifted 12.6 percentage points of its capital budget into computing and AI in twelve months without materially increasing the budget — the increase was funded almost entirely by cutting communications network investment by 9.8%.
That is a fundamentally different financing mechanism from the platform model. Alibaba, Tencent and Baidu are funding AI by pushing free cash flow negative. China Mobile is funding it by retiring a previous capital cycle. Its free cash flow went the other way: RMB53.9 billion in the first half of 2026, up 111.6% year-over-year, alongside intelligent computing capacity of 112,700 PFLOPS at FP16 and in-service resource utilisation above 90%.
Mia Scarlett’s read: a 90% utilisation rate on installed intelligent compute is the number a credit analyst would circle, and it points in an awkward direction for the platform thesis. The carrier is running its fleet close to full while its capex barely moves. The platforms are tripling capex and, in Tencent’s case, prepaying for chips that have not shipped. If domestic accelerator supply were the only constraint, you would expect the carrier — with guaranteed provincial demand and state procurement access — to be the one bidding hardest. It isn’t. Read the two disclosures together and the more plausible constraint on the platform side is not availability but timing: they are buying ahead of demand they have forecast, and the carrier is buying against demand it has already contracted.
China Telecom’s 2025 baseline gives the second carrier data point. Per the company’s own investor disclosures, full-year 2025 capital expenditure was RMB80.4 billion and China Telecom Cloud revenue reached RMB120.7 billion. For the first half of 2026 the company reported capital expenditure of RMB32.4 billion, of which RMB15.4 billion went to computing power networks, up 97% year-over-year — the same reallocation pattern, at a similar magnitude, at a second carrier.
Why ByteDance is absent from this dataset
ByteDance is widely described as one of China’s largest AI infrastructure spenders. It is privately held, files no audited results with any securities regulator, and publishes no capital expenditure figure. Axis Intelligence Research does not publish a ByteDance capex number, an estimate of one, or a range. Every “ByteDance AI capex” figure in circulation traces back to unnamed sourcing, and including it would contaminate a dataset whose entire value is that each row points to a filing. When ByteDance discloses, this page will carry the number and say where it came from.
What Is China’s Total AI Computing Capacity in 2026?
Capital expenditure is an input. Installed compute is the output, and China publishes it directly.
| Metric | Value | As of | Source |
|---|---|---|---|
| National intelligent computing capacity | 2,185 EFLOPS | 30 Jun 2026 | MIIT, via State Council Information Office |
| Nationwide computing facility utilisation | 71.4% | 30 Jun 2026 | MIIT, via State Council Information Office |
| National intelligent computing capacity | 1,590 EFLOPS | 31 Dec 2025 | National Data Administration, Digital China Development Report 2025 |
| China Mobile intelligent computing capacity (FP16) | 112.7 EFLOPS | 30 Jun 2026 | China Mobile Limited, 2026 Interim Results |
| China Mobile in-service utilisation | >90% | 30 Jun 2026 | China Mobile Limited, 2026 Interim Results |
| China Mobile 2026 capacity addition target | >48 EFLOPS | FY2026 target | China Mobile Limited, 2026 Interim Results |
MIIT reported national intelligent computing capacity of 2,185 EFLOPS at end-June 2026 with facility utilisation at 71.4%. Against the National Data Administration’s 1,590 EFLOPS at end-2025, that is 595 EFLOPS added in six months — growth of 37.4%.
Two things sit in tension inside that pair of numbers, and the tension is the analysis. National capacity grew 37.4% in half a year. National utilisation is 71.4%. China Mobile’s own utilisation is above 90%. The gap between the national rate and the largest carrier’s rate is roughly nineteen percentage points, which means a meaningful share of the non-carrier fleet — municipal intelligent computing centres, provincial projects, smaller commercial operators — is running materially below the carrier standard. Capacity is being added faster than it is being consumed at the periphery, even while the operators closest to contracted enterprise demand run near saturation.
China Mobile’s 112.7 EFLOPS is 5.2% of the national figure. Its target of adding more than 48 EFLOPS across 2026 would, on its own, represent roughly 8% of what the entire country added in the first half.
Which Chinese Companies Are Actually Earning AI Revenue?
Capex without revenue is a bet. The disclosed revenue lines are where the bet starts settling, and the growth rates diverge sharply from the capex rankings.
| Company | AI / cloud revenue line | Period | Value | YoY | Source |
|---|---|---|---|---|---|
| Alibaba | AI Cloud and Compute Services | Jun-Q 2026 | RMB48,437m | +45% | June Quarter 2026 Results |
| Alibaba | AI-related product revenue | Jun-Q 2026 | RMB12,376m | triple-digit, 12th consecutive quarter | June Quarter 2026 Results |
| Baidu | AI Cloud Infra | Q2 2026 | RMB7.3bn | +50% | Second Quarter 2026 Results |
| Baidu | GPU Cloud (within AI Cloud Infra) | Q2 2026 | not disclosed in absolute terms | +283% | Second Quarter 2026 Results |
| China Mobile | Computing services | 1H 2026 | RMB52.9bn | +14.0% | 2026 Interim Results |
| China Mobile | Intelligent computing services | 1H 2026 | RMB5.3bn | +130.1% | 2026 Interim Results |
| China Mobile | AIDC revenue | 1H 2026 | not disclosed in absolute terms | +486.1% | 2026 Interim Results |
| China Telecom | China Telecom Cloud | FY2025 | RMB120.7bn | — | China Telecom investor disclosures |
Periods are not comparable across rows: platform figures are three months, carrier figures are six months or a full year. Axis Intelligence Research does not annualise or blend them.
The pattern across all four disclosures is the same shape: a very large, slow-growing legacy cloud or computing line, and a small, violently fast-growing AI-specific line inside it. Alibaba’s AI Cloud and Compute Services grew 45% while AI-related products inside it grew triple digits. China Mobile’s computing services grew 14.0% while intelligent computing services inside grew 130.1% and AIDC grew 486.1%. Baidu’s AI Cloud Infra grew 50% while GPU Cloud inside it grew 283%.
Mia Scarlett’s read: this is what the early phase of a mix shift looks like from the outside, and it is the reason segment-level reading beats headline reading here. Every one of these companies could report a disappointing cloud number and a spectacular AI number in the same quarter, truthfully, because the AI line is still small enough that its growth rate does not yet move the segment it sits in. The quarter that matters is the one where the AI line is large enough that the segment growth rate converges toward it. Alibaba’s AI Cloud and Compute Services accelerating from 34% for the full fiscal year to 45% in the June quarter is the first visible sign of that convergence in this dataset.
The counterweight is on the profit line. Alibaba’s AI Labs and Applications segment recorded an adjusted EBITA loss of RMB13,861 million in the June quarter against RMB3,224 million a year earlier — the company attributed the widening to increased AI investment and higher inference cost. Inference cost scales with usage. A consumer AI product that succeeds makes that line worse before it makes it better.
Frequently Asked Questions
Which Chinese company spent the most on AI capex in the June 2026 quarter?
Alibaba, at RMB67,678 million (US$9,975 million), representing 51.3% of the RMB131,852 million reported collectively by Alibaba, Tencent and Baidu. Tencent was second at RMB52,784 million and Baidu third at RMB11,390 million. All three figures come from the companies’ own quarterly results announcements for the period ended 30 June 2026.
Is Chinese AI capex reported on a comparable basis across companies?
Only partially, and the differences matter when citing figures. Alibaba and Baidu report capital expenditure on a US GAAP cash-flow basis in quarterly SEC filings. Tencent reports under IFRS on a calendar-quarter basis and additionally splits operating from non-operating capex — its RMB52,784 million total comprises approximately RMB51.8 billion operating and RMB1.0 billion non-operating — while separately disclosing capital expenditure payments of RMB59.3 billion, a cash figure that differs from the accrued total. Alibaba’s fiscal year ends 31 March, so its “fiscal 2026” covers April 2025 to March 2026 and is not the same twelve months as Tencent’s or Baidu’s 2026. Axis Intelligence Research states the period and basis on every row rather than harmonising them.
What does the CAIS™ index measure that a capex figure doesn’t?
Affordability rather than magnitude. CAIS™ combines cash coverage (operating cash flow against capex), acceleration (year-over-year capex growth) and monetisation (disclosed AI or cloud revenue against capex) into a single 0–100 reading, weighted 45/30/25. Two companies can spend similar absolute amounts with very different strain profiles. In the June 2026 quarter Tencent read 23.6 and Baidu 60.4 — Tencent’s operating cash flow covered 99.8% of its capex, Baidu’s covered 30.2%.
Why is Tencent’s CAIS™ reading based on only two components?
Tencent does not disclose cloud revenue as a separate line item; its cloud business is reported inside FinTech and Business Services, which totalled RMB60,286 million in the June 2026 quarter alongside payment, wealth management and lending revenue. Rather than estimate a cloud subtotal, Axis Intelligence Research computes Tencent’s CAIS™ on cash coverage and acceleration only, reweighted to 60/40, and labels it wherever it appears.
Are China’s telecom carriers increasing AI capex or reallocating existing budgets?
Reallocating, on the disclosed evidence. China Mobile’s first-half 2026 group capital expenditure rose only 4.5% to RMB61.0 billion, but computing-network spending rose 71.3% to RMB15.6 billion and AI-network spending rose 85.2% to RMB3.7 billion, funded by a 9.8% reduction in communications network investment to RMB37.9 billion. Computing plus AI moved from roughly 19.0% to 31.6% of group capex. China Telecom disclosed the same pattern, with RMB15.4 billion of its RMB32.4 billion first-half capex going to computing power networks, up 97% year-over-year.
How much AI computing capacity does China have, and how much of it is being used?
MIIT reported national intelligent computing capacity of 2,185 EFLOPS at end-June 2026, with nationwide computing facility utilisation of 71.4%. That capacity figure is 37.4% above the 1,590 EFLOPS the National Data Administration reported for end-2025. China Mobile, which holds 112.7 EFLOPS or 5.2% of the national total, reported in-service utilisation above 90% — roughly nineteen percentage points above the national rate.
Does Alibaba’s RMB380 billion commitment mean RMB380 billion has been spent?
No. RMB380 billion (US$56.0 billion at the 30 June 2026 rate) is a three-year commitment. Management stated cumulative spending under the programme reached RMB190 billion through June 2026, or 50.0% of the pledged total. Axis Intelligence Research records pledges and disbursements as separate rows in the dataset because only disbursements appear in cash flow statements.
Why does this dataset exclude ByteDance and other private Chinese AI spenders?
Because no audited disclosure exists. ByteDance publishes no capital expenditure figure and files no results with a securities regulator, so any number attributed to it rests on unnamed sourcing. The value of this dataset is that every quantitative row points to a filed or published company document with a retrieval date. Adding an unverifiable figure — even a widely repeated one — would break that guarantee for every other row.
Methodology
Collection. Axis Intelligence Research retrieved each source document directly between 30 August and 1 September 2026. Alibaba and Baidu figures come from the exhibits to their Form 6-K filings on SEC EDGAR for the quarter ended 30 June 2026. Tencent figures come from the company’s 2026 Second Quarter Results announcement. China Mobile figures come from the company’s 2026 Interim Results press release and results presentation, both dated 13 August 2026. China Telecom’s full-year 2025 figures come from the company’s investor disclosures; its first-half 2026 capital expenditure figures come from its 2026 interim results announcement. National computing capacity figures come from MIIT via the State Council Information Office and from the National Data Administration’s Digital China Development Report 2025.
Currency. All US dollar conversions use RMB6.7851 to US$1.00, the 30 June 2026 rate published in the Federal Reserve Board’s H.10 statistical release. Both Alibaba and Baidu use this rate in their own filings, which makes it the only conversion basis that keeps Axis figures consistent with the source documents rather than introducing a third exchange rate.
Aggregation. The Platform-3 total is a simple sum of three companies’ reported capital expenditure for the same calendar quarter. It is not adjusted for accounting framework differences between US GAAP and IFRS, and it excludes Tencent’s separately disclosed capital expenditure payments figure of RMB59.3 billion, which is a cash measure rather than the accrued total.
CAIS™ computation. S1 = 100 × (1 − operating cash flow ÷ capex), floored at 0 and capped at 100. S2 = year-over-year capex growth in percentage points ÷ 3, capped at 100. S3 = 100 × (1 − disclosed AI or cloud segment revenue ÷ capex), floored at 0. CAIS™ = (S1 × 0.45) + (S2 × 0.30) + (S3 × 0.25). Where S3 is not computable from disclosure, the remaining components are reweighted to 60/40 and the reading is flagged. All arithmetic was computed twice — in floating point and in exact rational arithmetic — and reconciled before publication. Every figure in this article corresponds to a row in the downloadable CSV.
Scope. This dataset covers listed companies that disclose capital expenditure on a regular reporting cycle. It excludes private companies, provincial and municipal intelligent computing centre projects, state guidance fund deployments, and equipment vendors. Platform figures are three-month periods; carrier figures are six-month or full-year periods; these are never blended into a single total. Where a company does not disclose a component, Axis Intelligence Research reports the gap rather than estimating across it.
Related Axis datasets. Market-size, adoption and competitive-position figures for Chinese AI are maintained separately on the China AI statistics page, which owns the headline market-size and user-count metrics referenced across this silo. Accelerator supply and domestic silicon are covered in China AI chip statistics and China semiconductor statistics, with fabrication capacity detail in China memory chip production. Global comparators are tracked on the AI capex tracker and AI data center statistics.
About This Dataset
This dataset covers disclosed capital expenditure, AI and cloud segment revenue, and installed intelligent computing capacity for China’s principal listed AI infrastructure spenders, with observations spanning the June 2025 quarter through the June 2026 quarter and full-year 2025 baselines where relevant. Coverage is national (China) at the company and aggregate level.
Citation (APA): Axis Intelligence Research. (2026). China AI Capex Statistics 2026 [Dataset]. https://axis-intelligence.com/china-ai-capex-statistics/. License: CC BY 4.0.
Citation (MLA): Axis Intelligence Research. “China AI Capex Statistics 2026.” Axis Intelligence Research, September 2026, axis-intelligence.com/china-ai-capex-statistics/. CC BY 4.0.
Citation (Chicago): Axis Intelligence Research. “China AI Capex Statistics 2026.” Axis Intelligence Research, September 2026. https://axis-intelligence.com/china-ai-capex-statistics/.