Insights · Forecast Analysis
US–China AI Competition: Twelve Open Forecasts
Part of Forecast Future 2026 · Forecast Analysis
Twelve probabilistic forecasts with horizons, resolution conditions, source families, and stress-test context.
Public and institutional sources only. Probabilities are scenario-calibrated research judgments, not certainties, investment advice, or claims of demonstrated forecasting accuracy. Forecasts remain open until their resolution horizons.
Public and institutional sources only. Probabilities are scenario-calibrated research judgments, not certainties, investment advice, or claims of demonstrated forecasting accuracy. Forecasts remain open until their resolution horizons.
Featured forecast dashboard
Eight decision-relevant forecasts. Each shows a probability with horizon, resolution condition, and review dates—never probability alone.
F01CapabilityOpen Near-parity without leadership handoff
By 2030, frontier AI language competition enters a near-parity regime without a leadership handoff: China closes within 5% of the US leader on LMArena text performance, while US organizations continue leading at least 50% of quarterly capability snapshots.
72%current probability · resolution horizon 2030
- Resolution condition
- China closes within 5% of the US leader on LMArena text performance AND US organizations lead at least 50% of quarterly capability snapshots in the project’s defined snapshot set.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Measures layered near-parity, not a claim that capability competition is resolved.
F02CapabilityOpen Leadership separates by capability layer
By 2027, AI leadership separates by capability layer rather than converging into a single frontier race: China reaches language-model parity within 5% of the US leader, while the United States maintains at least a 10% advantage in agentic or tool-use performance on LMArena.
64%current probability · resolution horizon 2027
- Resolution condition
- China within 5% of the US leader on language-model performance AND US retains ≥10% advantage on the project’s agentic/tool-use LMArena metric.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Tests whether capability leadership fragments by layer.
F05DeploymentOpen China leads embodied-AI deployment capacity
By 2030, China ranks first in embodied-AI deployment capacity among the United States, Germany, Japan, South Korea, and India using the project’s defined ADAS and industrial-deployment indicators.
82%current probability · resolution horizon 2030
- Resolution condition
- China ranks first on the frozen embodied-AI deployment-capacity index among US, Germany, Japan, South Korea, and India.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Highest-conviction deployment claim in the featured set.
F06DeploymentOpen China leads automation-hardware exports
By 2030, China becomes the primary industrial supplier for global embodied-AI deployment, ranking first in automation-hardware exports among the United States, Germany, Japan, South Korea, and India with at least an eight-percentage-point share advantage over the second-ranked economy.
69%current probability · resolution horizon 2030
- Resolution condition
- China ranks first in automation-hardware export share among the five-economy set with ≥8 percentage-point advantage over the second-ranked economy.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Export leadership is not equivalent to geopolitical dominance.
F07Financial CapacityOpen Hyperscaler AI CapEx remains expansionary
By 2028, the US hyperscaler AI investment cycle remains expansionary: at least two years from 2026 through 2028 record annual CapEx growth of 15% or more without a defined digestion year.
65%current probability · resolution horizon 2028
- Resolution condition
- At least two calendar years in 2026–2028 show ≥15% annual AI-related CapEx growth for the frozen hyperscaler set, with no digestion year under the project definition.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Expansionary CapEx is not a valuation recommendation.
F08Financial CapacityOpen Internal financing capacity retained
By 2028, hyperscalers maintain financial capacity to fund AI expansion internally, with peak AI CapEx-to-operating-cash-flow remaining at or below 75% during 2026–2028.
83%current probability · resolution horizon 2028
- Resolution condition
- Peak AI CapEx-to-operating-cash-flow for the frozen hyperscaler set remains ≤75% in every year 2026–2028.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Highest probability in the featured financial pair.
F10Resource ManagementOpen Bottleneck shifts toward physical delivery
By 2029, the bottleneck hierarchy flips toward physical delivery: HBM residual scarcity falls below 10% while interconnection timelines remain at least 42 months.
70%current probability · resolution horizon 2029
- Resolution condition
- HBM residual scarcity <10% under the project definition AND representative grid interconnection timelines remain ≥42 months.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Central illustration of bottleneck migration.
F11Geopolitical PowerOpenRequires refinement Selective fragmentation persists
By 2030, US–China AI technology trade remains selectively fragmented rather than fully decoupled, with strategic technology separation increasing while both economies retain measurable dependencies across the defined AI-relevant supply-chain basket.
60%current probability · resolution horizon 2030
- Resolution condition
- Strategic separation indicators rise AND critical import dependencies remain below 8% under a frozen product basket, denominator, trade source, and aggregation rule (each economy / each product / or aggregate index — not yet frozen).
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Featured for narrative importance, but measurement rule is incomplete. Treat as open and provisional.
Forecast register
All twelve forecasts with filters for system, horizon, probability range, and status.
Showing 12 of 12
F01CapabilityOpen Near-parity without leadership handoff
By 2030, frontier AI language competition enters a near-parity regime without a leadership handoff: China closes within 5% of the US leader on LMArena text performance, while US organizations continue leading at least 50% of quarterly capability snapshots.
72%current probability · resolution horizon 2030
- Resolution condition
- China closes within 5% of the US leader on LMArena text performance AND US organizations lead at least 50% of quarterly capability snapshots in the project’s defined snapshot set.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Measures layered near-parity, not a claim that capability competition is resolved.
F02CapabilityOpen Leadership separates by capability layer
By 2027, AI leadership separates by capability layer rather than converging into a single frontier race: China reaches language-model parity within 5% of the US leader, while the United States maintains at least a 10% advantage in agentic or tool-use performance on LMArena.
64%current probability · resolution horizon 2027
- Resolution condition
- China within 5% of the US leader on language-model performance AND US retains ≥10% advantage on the project’s agentic/tool-use LMArena metric.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Tests whether capability leadership fragments by layer.
F03CapabilityOpen Open-weight diffusion with US frontier releases
By the end of 2028, AI competition splits into two architectures: China leads open-weight AI diffusion, while US organizations continue producing at least 50% of frontier AI releases.
51%current probability · resolution horizon 2028
- Resolution condition
- China leads the project’s open-weight diffusion metric AND US organizations produce at least 50% of frontier AI releases in the frozen release set.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Appendix forecast. Diffusion metric incomplete — not featured.
F04CapabilityOpenRequires refinement Domestic share of China accelerator stock
By 2028-12-31, domestic Chinese AI-chip designers (including Huawei and Cambricon in the project set) hold at least 70% of China’s reported AI accelerator stock in H100e.
65%current probability · resolution horizon 2028
- Resolution condition
- Domestic designers in the frozen firm set hold ≥70% of China’s reported accelerator stock measured in H100e.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Requires refinement. Evidence quality and unit definition are not yet publication-ready.
F05DeploymentOpen China leads embodied-AI deployment capacity
By 2030, China ranks first in embodied-AI deployment capacity among the United States, Germany, Japan, South Korea, and India using the project’s defined ADAS and industrial-deployment indicators.
82%current probability · resolution horizon 2030
- Resolution condition
- China ranks first on the frozen embodied-AI deployment-capacity index among US, Germany, Japan, South Korea, and India.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Highest-conviction deployment claim in the featured set.
F06DeploymentOpen China leads automation-hardware exports
By 2030, China becomes the primary industrial supplier for global embodied-AI deployment, ranking first in automation-hardware exports among the United States, Germany, Japan, South Korea, and India with at least an eight-percentage-point share advantage over the second-ranked economy.
69%current probability · resolution horizon 2030
- Resolution condition
- China ranks first in automation-hardware export share among the five-economy set with ≥8 percentage-point advantage over the second-ranked economy.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Export leadership is not equivalent to geopolitical dominance.
F07Financial CapacityOpen Hyperscaler AI CapEx remains expansionary
By 2028, the US hyperscaler AI investment cycle remains expansionary: at least two years from 2026 through 2028 record annual CapEx growth of 15% or more without a defined digestion year.
65%current probability · resolution horizon 2028
- Resolution condition
- At least two calendar years in 2026–2028 show ≥15% annual AI-related CapEx growth for the frozen hyperscaler set, with no digestion year under the project definition.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Expansionary CapEx is not a valuation recommendation.
F08Financial CapacityOpen Internal financing capacity retained
By 2028, hyperscalers maintain financial capacity to fund AI expansion internally, with peak AI CapEx-to-operating-cash-flow remaining at or below 75% during 2026–2028.
83%current probability · resolution horizon 2028
- Resolution condition
- Peak AI CapEx-to-operating-cash-flow for the frozen hyperscaler set remains ≤75% in every year 2026–2028.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Highest probability in the featured financial pair.
F09Financial CapacityOpenRequires refinement Scarce-input supplier returns
By 2028, firms controlling scarce AI inputs continue earning sufficient economic returns to sustain the next wave of AI infrastructure investment.
72%current probability · resolution horizon 2028
- Resolution condition
- Frozen scarce-input firm set meets the project’s return threshold for sustaining reinvestment (threshold not yet frozen).
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Requires refinement. Appendix-quality until firm set and threshold are frozen.
F10Resource ManagementOpen Bottleneck shifts toward physical delivery
By 2029, the bottleneck hierarchy flips toward physical delivery: HBM residual scarcity falls below 10% while interconnection timelines remain at least 42 months.
70%current probability · resolution horizon 2029
- Resolution condition
- HBM residual scarcity <10% under the project definition AND representative grid interconnection timelines remain ≥42 months.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Open forecast. Central illustration of bottleneck migration.
F11Geopolitical PowerOpenRequires refinement Selective fragmentation persists
By 2030, US–China AI technology trade remains selectively fragmented rather than fully decoupled, with strategic technology separation increasing while both economies retain measurable dependencies across the defined AI-relevant supply-chain basket.
60%current probability · resolution horizon 2030
- Resolution condition
- Strategic separation indicators rise AND critical import dependencies remain below 8% under a frozen product basket, denominator, trade source, and aggregation rule (each economy / each product / or aggregate index — not yet frozen).
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Featured for narrative importance, but measurement rule is incomplete. Treat as open and provisional.
F12Geopolitical PowerOpen Incomplete export-control coordination
By 2028, AI technology competition operates under an incomplete containment regime: Japan maintains advanced semiconductor equipment controls on China, while equivalent restrictions remain limited to at most three of six US-aligned economies.
64%current probability · resolution horizon 2028
- Resolution condition
- Japan maintains advanced semiconductor equipment controls on China AND equivalent restrictions apply in at most three of the six US-aligned economies in the frozen comparison set.
- Initial date
- 2026-07-31
- Last reviewed
- 2026-08-06
Appendix forecast. Comparison set not yet frozen.
Stress-test explorer
Alternate regimes change which systems matter and which forecasts strengthen or weaken.
Scenario A: AI remains primarily software-based
- Capability
- Financial Capacity
F01 · F02 · F07 · F08
F05 · F06 · F10
- Share of AI value accruing to cloud and enterprise software versus physical systems
- Hyperscaler CapEx guidance versus industrial robotics CapEx
- Pace of ADAS and factory-automation adoption
- US frontier-model and cloud advantages become more decisive.
- China’s embodied-AI deployment advantage matters less.
- Grid and physical-delivery bottlenecks develop more slowly.
A. AI remains primarily software-based
- Capability
- Financial Capacity
F01 · F02 · F07 · F08
F05 · F06 · F10
- Share of AI value accruing to cloud and enterprise software versus physical systems
- Hyperscaler CapEx guidance versus industrial robotics CapEx
- Pace of ADAS and factory-automation adoption
- US frontier-model and cloud advantages become more decisive.
- China’s embodied-AI deployment advantage matters less.
- Grid and physical-delivery bottlenecks develop more slowly.
B. Embodied AI becomes the primary productivity engine
- Deployment
- Resource Management
F05 · F06 · F10
F01 · F07
- Industrial robot density and ADAS penetration differentials
- Automation-hardware export shares
- Whether deployment leadership translates into international adoption under trust constraints
- China’s manufacturing and integration advantages become more valuable.
- Automation-hardware exports become a stronger indicator of global diffusion.
- Deployment leadership still does not automatically create geopolitical dominance.
C. Monetization disappoints
- Financial Capacity
F08
F07 · F09 · F10
- Hyperscaler CapEx growth versus operating cash flow
- Supplier margins in scarce AI inputs
- Whether CapEx enters a digestion period under the project definition
- Hyperscaler CapEx enters a digestion period.
- Internal cash generation determines whether investment pauses or contracts.
- Supplier returns and reinvestment incentives weaken.
D. Energy becomes the dominant constraint
- Resource Management
- Deployment
F10 · F05
F07
- Interconnection queue lengths and power availability for data centers
- HBM residual scarcity versus grid delivery timelines
- Comparative advantage in infrastructure execution
- Data-center expansion becomes limited by power availability and interconnection.
- Semiconductor leadership remains important but is no longer sufficient.
- Infrastructure execution becomes a larger comparative advantage.
E. Geopolitical decoupling accelerates
- Geopolitical Power
- Capability
F04 · F12
F06 · F11
- Breadth of export-control coordination across aligned economies
- Domestic substitution rates for controlled inputs
- International adoption barriers for Chinese AI-enabled hardware
- Export controls and domestic substitution broaden.
- System redundancy increases while global efficiency declines.
- China’s domestic technology development becomes more important, but international diffusion becomes harder.
Limitations · read before citing
Disclosures and limits
Probabilities are open research judgments. They are not investment advice, backtested accuracy claims, or resolved outcomes.
- The work uses public and institutionally accessible sources rather than proprietary datasets.
- Several concepts—agentic capability, deployment efficiency, geopolitical influence, and supply-chain dependence—require contestable proxies.
- Probabilities are structured judgments informed by evidence and scenarios; they are not outputs from a fully validated statistical forecasting model.
- Forecast wording, thresholds, data vintages, and resolution sources must be frozen and recorded before publication.
- Future updates must preserve the original probability and timestamp rather than silently replacing prior judgments.
- Company and market discussion is research commentary, not investment advice.
- AI tools supported research organization, hypothesis development, coding, data processing, stress testing, and editing. Charlene remains responsible for source selection, analytical decisions, interpretation, verification, and final claims.