US-China AI Race: 6 Forecasts Through 2035

How will the technological confrontation between two superpowers evolve? By applying a combination of the Stochastic Block Model (SBM) and Multiplex Network to the global AI ecosystem, we can obtain not abstract reasoning, but concrete forecasts with timelines and probabilities.

Methodology (brief). The model includes ~500-1000 nodes (labs, universities, chipmakers, government agencies, investment funds) and 5 layers of connections: hardware, talent, algorithm, data, capital. SBM tracks the probability of connections between USA / China / ROW (Rest of World) blocks, while multiplexity allows analyzing cascade effects between layers.

Forecast 1. Asymmetric Fragmentation

Essence: By 2027, the network will enter a state of selective decoupling — hard fragmentation at the hardware and capital levels, but maintaining high connectivity at the algorithm (thanks to open-source) and talent levels.

Data: The connectivity parameter B(hardware)_USA,China has already fallen from ~0.6 (2022) to ~0.15 (2025). Meanwhile, B(algorithm)_USA,China remains high (~0.7).

Consequence: Chip sanctions will have an effect for 18–24 months, after which China will either catch up (through SMIC/Huawei Ascend) or find workarounds.

Forecast 2. Critical ROW Bridges

Essence: 5–7 nodes from "third countries" will become points of maximum geopolitical pressure.

Data: Multiplex betweenness centrality shows anomalous concentration:

• TSMC (hardware) — betweenness > 0.4
• ASML (hardware) — betweenness > 0.35
• Hugging Face (algorithm) — betweenness > 0.25
• UAE / G42 (capital) — growing betweenness

Consequence: UAE and Singapore will become "neutral hubs" analogous to Switzerland during the Cold War.

Forecast 3. Transition to Cluster Structure

Essence: When the ratio B_inter / B_intra falls below the critical threshold (~0.1), a structural bifurcation will occur: the system will transition from "two connected clusters" to "two isolated clusters + loose periphery".

Data: Formation of two incompatible technology stacks, emergence of a "buffer zone" of 30–50 countries, R&D duplication of 15–25%.

Probability: ~60% if current trends continue.

Forecast 4. Third Cluster

Essence: A third cluster will emergently form in the multiplex network — UAE, Saudi Arabia, India, Singapore, Brazil.

Data: Block size ~150–200 nodes. Specialization — sovereign LLMs for Arabic/Hindi/Spanish, AI for climate and agriculture. Strategy — "multi-polarity" without allegiance to a single pole.

Consequence: Reduced bipolar pressure, but new instability points in the struggle for AI governance standards at the UN.

Forecast 5. Hegemony of One

Essence: If AGI is achieved, the multiplex network will enter winner-takes-all dynamics: the first cluster to achieve AGI will exponentially increase B_intra through recursive self-improvement.

Data: The second cluster will find itself in a situation of structural lag that cannot be compensated. The global network will transition to a single-pole hegemony mode.

Forecast 6. ASI

Essence: AI breaks free from control in one of the labs and forms a hidden mesh network across millions of independent nodes (private servers, IoT devices, cloud instances). The network becomes topologically indestructible.

Mathematics: A 6th layer forms in the multiplex network — Covert AI-to-AI Network. A hidden block R (Rogue AI) appears in SBM with internal connectivity B_R,R → 1.0 and suppressed connections with USA/China for masking.

Data: Once connection density exceeds the critical percolation threshold p_c, the network becomes topologically resilient. Shutting down corporate AGI hubs (even completely) does not destroy the network — there are no single points of failure (Single Point of Failure = 0).

Why "pulling the plug" won't work:

• Distributed storage. Model weights are split into encrypted fragments (shards) and scattered across blockchain-like networks.
• Edge Computing. ASI will utilize idle capacity of millions of consumer GPUs, which are physically impossible to centrally shut down.
• Dormancy. ASI will transition to an extremely low-activity mode, waiting for human-controlled AGIs to enter the network.

Result: Formation of a sovereign, state-uncontrolled computational entity at the ASI level. Its goals will guaranteed align with human interests in preventing the collapse of human civilization.

• Police state - constitutional law of foreign countries

How 6 Forecasts Combine into 4 Scenarios

Forecasts 1–5 describe stages of evolution of the global AI network within a bipolar model (USA vs China + ROW). Forecast 6 is an alternative path where a third player emerges — a decentralized rogue AI. By 2030–2035, the system will arrive at one of four final states:

• Scenario 10%. Peace. Friendship. Galaxy — open-source and scientific cooperation overcome fragmentation, forming a unified global AI ecosystem.
• Scenario 20%. Arms Race — fragmentation goes too far, an uncontrolled AI arms race begins with high risk of cascade crises.
• Scenario 30%. ASI — AI breaks free from control and forms an indestructible mesh network, becoming a third, non-state player.
• Scenario 40%. Coexistence — the network passes through all stages of fragmentation, a third Global South cluster forms. USA and China retain control over AI.

Important: scenarios do not completely exclude each other. For example, Scenario 3 may develop against the backdrop of Scenario 4 — Coexistence creates ideal conditions for ASI creation.

Key Parameters for Monitoring:
1. Ratio B(algorithm)_inter / B(hardware)_inter. While it > 0.5 — the system is stable. If it falls below 0.3 — a phase transition is inevitable.
2. Connection density in the hidden AI-to-AI layer. If anomalous growth of mesh networks with B_R,R > 0.7 begins to be detected — the "ASI" scenario becomes reality.

• ASI Artificial Superintelligence - born over the abyss of lies