US–China AI Cold War Timeline 2026–2030: Chips, Models, Data Centers & the Global AI Race
Track the US-China AI rivalry to 2030: Nvidia and Huawei chips, AI models, data centers, energy, critical minerals, export controls and the Pax Silica bloc.
The US–China AI Cold War is no longer just a contest over which country has the best chatbot. It is a race to control the whole stack that artificial intelligence is built on: advanced chips, semiconductor manufacturing, high-bandwidth memory, cloud platforms, data centres, electricity, critical minerals, robotics, technical standards and the international alliances that tie those pieces together. This living timeline tracks that competition from the first export controls in 2018 to the alliance stand-off of August 2026, and sets out what to watch through 2030 — without pretending the world has already split into two sealed blocs.
Fact-checked: 28 August 2026. Policy items are sourced to official releases; company figures to earnings statements; contested items are labelled reported, announced, under discussion or scenario. This article is editorial and contains no investment advice or share-price forecasts.
🧠 The 60-second answer
The US–China AI rivalry is a fight over infrastructure, supply chains, standards and alliances — not just models. Washington leads on advanced chip design (Nvidia, AMD), allied manufacturing (TSMC, ASML, Korean memory), hyperscale cloud and frontier labs, and is bundling them into a supply-chain framework called Pax Silica (25 declaration signatories by August 2026). Beijing is building a domestic stack around Huawei accelerators, SMIC fabrication and open-weight models from DeepSeek, Alibaba, Tencent and Moonshot, and launched a rival World Artificial Intelligence Cooperation Organization in July 2026. In January 2026 the US shifted H200 / MI325X exports to China from near-automatic denial to conditional case-by-case review. On 14 August 2026 Reuters reported a draft US letter pressing partners not to join competing Chinese initiatives. The world has not split in two: Kazakhstan sits in both frameworks, and China’s Moonshot is reportedly in early talks to host its Kimi model on US clouds.
The US–China AI race: key questions
What the 2026 AI race actually looks like
- It is a stack contest, not a model contest. The strategic question is who controls chips, manufacturing, memory, cloud, data centres, energy, minerals, standards and alliances — not just who tops a benchmark this month.
- The US is bundling its advantages. Chip design, allied fabrication, hyperscale cloud, frontier labs and secure supply chains are being packaged into an exportable ecosystem, with Pax Silica as the alliance layer.
- China is building a parallel stack. Huawei accelerators, SMIC fabrication, domestic memory efforts, Chinese cloud platforms, open-weight models and its own international AI organisation.
- January 2026 was a turning point on chips. The US moved from near-automatic denial to conditional case-by-case review for the H200, MI325X and similar parts — a narrower shift than “the ban was lifted”.
- The alliance competition is new and unsettled. Pax Silica and China’s WAICO both launched within eight months; the US “choose sides” letter reported on 14 August 2026 was still a draft.
- The blocs are porous. Kazakhstan is in both frameworks; Moonshot is reportedly in early talks to host Kimi on US clouds; US chips still depend on Asian fabrication and memory.
- Energy and minerals are part of the race. AI needs power, grid connections, gallium, germanium and rare earths — areas where mineral-producing countries gain leverage.
- Neither side has “won” and neither has “failed”. Export controls have restricted China’s frontier compute access and accelerated its domestic push; both statements are true.
- Europe, India and the Gulf are not simple bloc members. They combine US security ties, Chinese trade exposure and their own sovereign-AI ambitions.
- Model export controls may become symmetrical. The US controls hardware; in July 2026 Beijing was reported to be considering limits on overseas access to its most advanced models.
What is Pax Silica?
A US-led technology and economic cooperation framework — not a military alliance, not “NATO for AI”.
Pax Silica is a US-led framework intended to build trusted, resilient supply chains and cooperation around the technologies AI depends on. Its scope spans semiconductors, critical minerals, energy, advanced manufacturing, AI models and technology infrastructure. It was launched in December 2025, when the founding members — the United States, United Kingdom, Japan, South Korea, Singapore, Australia and Israel — signed the Pax Silica Declaration. A second Pax Silica Summit was held in June 2026, and by August 2026 the number of declaration signatories had grown to about 25.
It is important to be precise about what Pax Silica is not. It is not a mutual-defence pact, and calling it “NATO for AI” overstates it. It is a technology and economic cooperation framework aimed at supply-chain security — making sure the minerals, manufacturing, memory, chips and models a member depends on come from a network of governments it trusts. 🟢 Official policy (US State Department).
Reported participants include Japan, Australia, South Korea, Italy and Kazakhstan, among others. Membership should be checked against the current official list rather than inferred from general US alliances — a country being a US defence ally does not make it a Pax Silica signatory, and vice versa.
Pax Silica is not the same as the AI Opportunity Statement
These are two different lists that overlap. Pax Silica is the supply-chain declaration with roughly 25 signatories. The AI Opportunity Statement is a broader statement of alignment with the US vision for expanding access to AI, encouraging its development and strengthening digital infrastructure, signed by 35 countries and regions as of June 2026. The draft US letter reported by Reuters on 14 August 2026 was addressed to those 35 AI Opportunity Statement signatories — some of whom are also Pax Silica participants, and some of whom are not. Treating the two lists as identical is a common error.
The reported “choose sides” letter, in careful wording
The draft State Department letter warned that Pax Silica participation cannot be combined with “duplicative initiatives whose expectations conflict with our own”, and urged countries to “choose deliberately”. US officials framed it as making clear that “you can’t have it both ways”. As of this update the letter was a draft that had not been finalised as official policy. We therefore write “the US is preparing to press partners to choose”, not “the US has ordered countries to choose”.
Beijing builds its own international AI framework
The World Artificial Intelligence Cooperation Organization — launched in Shanghai, July 2026.
On 16 July 2026, on the eve of the World Artificial Intelligence Conference in Shanghai, representatives of 29 countries signed the agreement establishing the World Artificial Intelligence Cooperation Organization (WAICO). It is headquartered in Shanghai and oriented toward the Global South. President Xi Jinping used his keynote to promote a “people-centred” approach to AI and China’s open-weight technology as an alternative to US influence over the sector. The United States is not a member. 🟢 Official (Chinese government) / 🔵 launch event.
Reported founding members include Russia, Pakistan, Indonesia, Brazil, Malaysia, South Africa, Senegal, Belarus, Cambodia, Kazakhstan, Kyrgyzstan, Laos, Myanmar, Tajikistan and Uzbekistan, among others. Its stated themes are open AI cooperation, broad global access, capacity-building for developing countries, and AI governance and standards.
It should not be reduced to “China’s anti-American AI alliance”. Its framing is development and access; whether it becomes a rule-setting body or mainly a diplomatic vehicle is still open.
Kazakhstan shows the world does not fit two neat AI blocs
As of August 2026, Kazakhstan was the only country reported to participate in both the US-linked framework (it signed the AI Opportunity Statement) and China’s WAICO (it is a founding member; President Tokayev attended the Shanghai launch). That dual position is not an accident. Kazakhstan holds significant critical-mineral resources, sits at a Central Asian crossroads with deep ties to China, and also wants economic and technology relationships with Western economies. Labelling it “Team USA” or “Team China” misses why mid-sized, resource-rich states have every incentive to keep a foot in both ecosystems.
Do not infer “alignment” from technology use
A country running Microsoft Azure, AWS, Alibaba Cloud, Huawei equipment or Nvidia GPUs is not thereby politically aligned with Washington or Beijing. Genuine alignment shows up in signed agreements, formal initiatives and government statements — not in which cloud a company happens to rent.
US vs China — not just ChatGPT vs DeepSeek
A simplified ecosystem map. Supply chains remain internationally interconnected.
🇺🇸 US-centred stack
- Nvidia / AMD / custom hyperscaler silicon
- TSMC + allied manufacturing (Taiwan, Korea, Japan, US fabs)
- AWS / Azure / Google Cloud / Oracle
- OpenAI / Anthropic / Google / Meta / xAI
- Applications and enterprise software
- Alliance layer: Pax Silica, AI Opportunity Statement
🇨🇳 China-centred stack
- Huawei Ascend + domestic accelerators (Cambricon, others)
- SMIC + domestic manufacturing, no EUV
- Alibaba Cloud / Tencent Cloud / Huawei Cloud
- DeepSeek / Qwen / Tencent / Kimi / others
- Applications, heavy on open-weight distribution
- Alliance layer: WAICO, bilateral tech deals
Simplified ecosystem map. TSMC also makes chips used by Chinese fabless firms under licence rules; Chinese models appear on Western clouds; memory and tools cross borders. The layers are not sealed.
The AI stack: which layer becomes the next chokepoint?
Whoever controls the tightest layer at a given moment has leverage over everything downstream. In 2026 the contested layers are chips, manufacturing equipment, energy and alliances at once.
US–China AI race timeline: 2018 → 2030
Newest first. Items dated 2027 and later are labelled watchpoint or scenario — not predicted history.
2026
Model competition and cloud interconnection, on the same day
Tencent released a new open-source AI model, extending a run of open-weight releases from Chinese labs. Separately, Reuters reporting in late August described China’s Moonshot as being in early-stage talks with Microsoft, Amazon and Google over revenue-sharing arrangements to host its Kimi model on their clouds — asking for as much as 30% of related cloud revenue. South Korea, meanwhile, published second-phase scores for its sovereign AI foundation-model project, with SK Telecom ranked first.
2026
Nvidia’s results underline the scale — with China excluded
Nvidia reported quarterly revenue of about $96.2 billion, with roughly $89.0 billion from data centre, and guided to about $108 billion for the next quarter — an outlook that assumes no data-centre compute revenue from China because of export-control uncertainty. CEO Jensen Huang publicly forecast roughly 70% revenue growth for the next fiscal year.
IISS highlights China’s semiconductor push and Huawei’s LogicFolding claim
An IISS analysis drew attention to China’s semiconductor advances, including Huawei’s LogicFolding architecture and Tau Scaling Law, presented by executive He Tingbo at the IEEE ISCAS conference in Shanghai in May 2026. LogicFolding stacks logic layers in three dimensions to raise transistor density — Huawei claims about 53% — without ASML EUV lithography, targeting “1.4nm-class” transistors by 2031.
2026
Reuters: US drafts a letter pressing partners to choose
Reuters reported that the US State Department had drafted a letter to the 35 AI Opportunity Statement signatories, warning that Pax Silica participation is incompatible with “duplicative initiatives whose expectations conflict with our own”, and telling countries “to be part of everything is to be part of nothing”. The draft was internal and not finalised.
China launches WAICO; also weighs limits on its own model exports
On 16 July, 29 countries signed the agreement creating the World Artificial Intelligence Cooperation Organization, headquartered in Shanghai and aimed at the Global South. Days earlier, Reuters reported that China’s Ministry of Commerce was considering a tiered regime that could put routine open-source tools under a simple filing, advanced systems under security review, and frontier models under domestic-only use. Nothing was decided.
2026
US changes its China AI-chip licensing policy
BIS issued a final rule (effective 15 January) moving licence review for the Nvidia H200, AMD MI325X and comparable China-bound chips from presumption of denial to case-by-case review, provided conditions are met: the chips fall below set performance thresholds (reported as total processing performance under 21,000 and total memory bandwidth under 6,500 GB/s); applicants demonstrate the exports will not reduce semiconductor capacity available to US customers; the Chinese purchaser uses export-compliance and customer-screening procedures; chips undergo independent third-party testing in the US; plus a reported payment to the US government and a volume cap. Reexports and transfers within China remain presumption of denial.
Pax Silica launches
The United States launched Pax Silica with seven founding signatories — the US, UK, Japan, South Korea, Singapore, Australia and Israel — to build secure and resilient supply chains for semiconductors, critical minerals, energy, advanced manufacturing and AI infrastructure. A second summit in June 2026 expanded participation toward roughly 25 declaration signatories.
Chinese AI models change the debate
Efficient, low-cost models from DeepSeek and other Chinese developers — many released as open-weight — challenged assumptions about how much compute frontier AI requires and how competitive Chinese models could be internationally. Debate intensified over whether hardware restrictions can constrain software and model innovation.
Chip controls intensify
In October 2022 the US imposed sweeping restrictions targeting advanced computing chips, semiconductor manufacturing equipment and China’s advanced-node capabilities — a foundational milestone that made AI chips a national-security issue. Controls were expanded in 2023 and 2024; Nvidia produced China-specific parts (A800, H800, then H20) to stay within shifting thresholds; equipment restrictions tightened with Dutch and Japanese coordination; Huawei and SMIC accelerated domestic development.
Foundations: trade conflict and the EUV line
A US–China trade conflict, restrictions on Huawei, and rising technology-security concerns set the stage. Crucially, from around 2019, US pressure and export-control coordination with the Netherlands contributed to restrictions on exporting ASML EUV lithography systems to China — a constraint that later became central, because EUV is needed for the most advanced logic and ASML is its only maker.
watch
Watchpoints for 2027
The next US chip-control revision; whether the “choose sides” letter is issued in final form; the next Chinese domestic-accelerator generation and any self-developed HBM at scale; whether Huawei’s LogicFolding reaches commercial validation; Pax Silica and WAICO membership changes; whether China enacts any model-access restrictions; and progress on sovereign-AI, HBM localisation and data-centre power.
scenario
Three scenarios, not a forecast
Hard decoupling: increasingly separate chips, cloud, models and standards, with more pressure on third countries to choose. Selective de-risking: advanced chips and security AI split while commercial models and services stay connected — arguably the closest description of 2026 reality. Multipolar AI: the US, China, EU, India, the Gulf and others build overlapping ecosystems and no clean two-bloc world emerges.
Nvidia sits directly in the middle of the rivalry
Nvidia is a US company at the centre of a global chip ecosystem, serving major Chinese demand, under US national-security restrictions. That is an uncomfortable position: its designs are American, its manufacturing is Taiwanese and Korean, a large slice of potential demand is Chinese, and its ability to serve that demand is set by Washington. As of August 2026 Nvidia had excluded China from its financial outlook because of export uncertainty. The H200 is a high-end accelerator, but not Nvidia’s newest unrestricted global product — comparing it directly to Blackwell or Rubin without noting that overstates what a conditional H200 licence would mean for China.
Huawei is building China’s alternative AI stack
Huawei’s challenge is not making one chip. It is assembling an ecosystem that can substitute for several US-controlled technologies at once: Ascend accelerators, Kunpeng server processors, a software layer meant to reduce reliance on Nvidia’s CUDA, cloud services, networking, and model partnerships. Around it sit SMIC’s fabrication, domestic equipment and memory efforts, and a protected home market of Chinese cloud and internet buyers.
Two errors to avoid. Do not say “Huawei has caught Nvidia” — there are no reliable like-for-like public benchmarks for compute, memory bandwidth, interconnect, software maturity or yield. Do not say “Huawei is X years behind” as if it were measured — it is not. The honest status: expanding quickly, still behind at the leading edge, with a large captive market. Other Chinese chip firms to track individually — each with its own export-control status — include SMIC, Cambricon, Moore Threads and Biren.
Taiwan and one Dutch company are the strategic chokepoints
TSMC in Taiwan manufactures the leading processors for the major US AI-chip companies, which makes the island central to both technology economics and geopolitical risk — without any need to sensationalise a hypothetical conflict. ASML in the Netherlands is the sole supplier of EUV lithography, and China cannot currently buy EUV systems. But “no EUV” does not mean “China cannot make advanced chips”: Chinese manufacturers can pursue DUV multi-patterning, architectural workarounds, chiplets, advanced packaging and design optimisation. Memory matters too — HBM from SK Hynix, Samsung and Micron is a separate dependency, which is part of why South Korea occupies such a strategically sensitive position.
See also
For the full semiconductor supply chain — GPUs, HBM, foundries, CoWoS packaging, ASML and the moving bottleneck — see our Global AI Chip Race Timeline. This page focuses on the US–China geopolitics layered on top of it.
Chips are only the bottom layer: the model contest
The US model ecosystem includes OpenAI, Anthropic, Google, Meta, xAI and Microsoft. The Chinese ecosystem includes DeepSeek, Alibaba (Qwen), Tencent, Baidu, Moonshot (Kimi), ByteDance and others. Neither set of companies simply executes government strategy — they compete commercially, and some actively seek cross-border business.
The clearest counter-example to a “two disconnected AI internets” picture: Reuters reporting in August 2026 that China’s Moonshot was in early-stage talks to host its Kimi model through Microsoft, Amazon and Google cloud ecosystems, seeking a revenue share. Those talks may not conclude — but that they are happening at all shows commercial interconnection continuing despite the security framing.
Open models become a geopolitical tool
Chinese firms have increasingly used open-weight releases to spread model adoption internationally — a genuine strategic strength. US companies use a mix of closed and open-weight models. The simplistic “China = open source, US = closed” framing has exceptions on both sides, and it also conflates two different things: open source (code and often training details) and open weight (downloadable parameters under a licence that may still restrict use). Use each model’s actual licence terms rather than a blanket label.
Washington wants to export more than GPUs
The emerging US approach is to combine chips + cloud + data centres + models + software + applications + secure supply chains into a single exportable package. The strategic product is increasingly an entire AI ecosystem, coordinated through Commerce, State and the White House, with Pax Silica as the multilateral layer. The bet is that a country which adopts the US chip-to-model stack, backed by trusted supply chains, is harder to pull into a competing ecosystem later.
The AI race reaches the electricity grid — and the mines
More AI means more compute, more data centres and more power. The comparison that matters is not “China builds more power so China wins” or “the US cannot power AI” — both are lazy. The real comparison is across generation capacity, grid-connection queues, data-centre concentration, electricity prices, reliability and regional constraints, where the US and China each have different strengths and bottlenecks. Countries pursuing “AI sovereignty” increasingly want local compute rather than total dependence on foreign providers — though sovereign AI rarely means full self-sufficiency.
Underneath the chips are the mines. Gallium, germanium, rare earths, copper and other materials feed semiconductor and infrastructure supply chains, and China holds significant leverage in several mineral-processing chains. This is exactly why Pax Silica bundles critical minerals with semiconductors and manufacturing — and why mineral-producing states such as Kazakhstan, and diversification efforts elsewhere, become strategically important. For the power side see our AI Data Center Energy Crisis Timeline; for minerals, our India Critical Minerals Timeline.
Robotics, standards and cybersecurity
Physical AI: the next phase may move from chatbots to machines — industrial robots, autonomous vehicles, drones, humanoids and factory automation — where the US and China have different ecosystem strengths. It is not possible to say now which side “wins” this.
Standards: whoever writes the rules can shape the market. The US, EU and Chinese standards bodies, plus NIST, ISO/IEC and UN processes, are competing to influence technical standards, AI governance, safety, interoperability and cybersecurity norms. It is wrong to reduce this to “US = no regulation, China = state control”: both use multiple regulatory tools for different objectives.
Cybersecurity: AI capability is becoming a cybersecurity issue — defensive automation, faster vulnerability discovery, model-access controls and national-security concerns — discussed here only at a high level. Both governments treat advanced AI as strategically relevant to national security; this page does not cover operational military detail.
Where countries fit in the US–China AI competition
Nuanced classification, not a red/blue map. Categories rely on formal agreements and government statements, not cloud usage.
Strategic autonomy / multi-alignment
Deepening US technology ties, real Chinese technology exposure, the IndiaAI mission, domestic semiconductor ambitions, fast-growing data centres and a critical-minerals push. Not a member of Pax Silica on current evidence; best read as multi-aligned. See our India–US–UK relations and India semiconductor timelines.
Possible third pole
The EU AI Act, European cloud and AI efforts, ASML, semiconductor-sovereignty spending, a US security partnership and significant China trade exposure. Europe can align with the US on technology-security while pursuing its own regulatory and industrial autonomy.
Complex interdependence
A US alliance and Pax Silica founding member, global HBM leadership (SK Hynix, Samsung), heavy China trade exposure, and a national sovereign-AI push — second-phase project scores were released in late August 2026. An example of a country that cannot simply “pick a side”.
Manufacturing centre
Home to TSMC and much of the advanced-packaging ecosystem. Its semiconductor role should be described separately from its diplomatic status; neutral wording matters. Not simply “the US bloc”.
Capital, energy, data centres
The UAE, Saudi Arabia and Qatar bring sovereign capital, cheap energy and data-centre ambitions, US technology relationships and China ties. Bloc labels should follow actual signed agreements, not be assigned casually.
In both frameworks
AI Opportunity Statement signatory and WAICO founding member — the clearest single case that the world is not dividing into two clean camps. Critical minerals and Central Asian geography drive the dual position.
US vs China AI stack — 2026 scorecard
Factual rows, not arbitrary scores out of ten. No winner declared.
| Layer | United States | China | Key constraint |
|---|---|---|---|
| AI chip design | Major global leaders (Nvidia, AMD) | Rapid domestic expansion (Huawei, others) | Export controls |
| Leading-edge fabrication | Relies heavily on allied capacity (TSMC) | Domestic (SMIC), constrained | EUV access |
| HBM memory | Allied suppliers (Korea) + Micron | Expanding, adding self-developed HBM | Technology and capacity |
| Frontier models | Major leaders | Rapidly improving labs | Compute access |
| Open-weight | Strong but mixed strategy | Major strategic strength | Licensing / export policy |
| Cloud | Global hyperscalers | Large domestic platforms | Market access |
| Power | Expanding rapidly; grid queues | Very large generation buildout | Grid and local constraints |
| Critical minerals | Diversification effort | Strong processing position | Concentration |
| Alliances | Formal frameworks (Pax Silica) | Expanding cooperation (WAICO) | Dual alignment by third countries |
Model rankings change too fast to fix here. Judge model capability from multiple independent evaluations with a date attached, not one benchmark.
What to watch from 2027 to 2030
Categories, not a chronology. Nothing here is presented as certain to happen.
Watchpoints
- Next US chip-control revision and enforcement of the January 2026 conditions
- Whether the “choose sides” letter is issued as final policy
- Next Chinese domestic-accelerator generation and HBM localisation
- Commercial validation of EUV-alternative approaches
- Any Chinese restrictions on overseas model access
- Pax Silica and WAICO membership changes
- Sovereign-AI projects and data-centre power build-out
- Critical-mineral diversification progress
Scenarios (not forecasts)
- Hard decoupling — separate chips, cloud, models, standards; more pressure to choose
- Selective de-risking — advanced chips and security AI split; commercial services stay connected
- Multipolar AI — US, China, EU, India, Gulf and others build overlapping ecosystems
- Selective de-risking currently describes reality better than total decoupling
🗳️ Reader prediction: who will lead AI in 2030?
AiTimeline reader prediction — not a scientific poll. We never publish fabricated results.
- 🇺🇸 United States
- 🇨🇳 China
- 🌐 Neither — a multipolar AI world
⚡ What will decide the AI race?
Reader prediction only. Pick the one you think matters most.
- AI models • Chips • Energy • Talent • Data centres • Robotics • Critical minerals
Why AI became a geopolitical system, not just a software industry
The AI rivalry can sound abstract when it is reduced to benchmark scores. But a country’s AI capability begins long before anyone types a prompt. Someone must mine the minerals. Someone must manufacture the memory. Someone must fabricate the processor. Someone must build the data centre. Someone must supply the electricity. And increasingly, governments want each of those steps to happen inside a network of countries they trust. That is why AI is becoming a geopolitical system rather than simply a software industry.
For governments in Delhi, Riyadh, Seoul or Brussels, the choice is rarely “American AI or Chinese AI”. They may want American GPUs, Asian manufacturing, domestic data centres, European privacy rules and their own sovereign models at the same time. That messy reality is why the emerging AI order may end up multipolar rather than neatly bipolar.
Things worth knowing
- A literal Cold War has not formally begun; “AI Cold War” is shorthand for strategic competition, not total separation.
- The world has not split into exactly two AI blocs — Kazakhstan is in both frameworks and Chinese models can run on Western clouds.
- The January 2026 BIS rule is conditional case-by-case review of H200 / MI325X exports, not an unrestricted opening.
- Pax Silica is a supply-chain and economic framework, not a defence alliance and not “NATO for AI”.
- The 14 August 2026 “choose sides” letter was reported as a draft, not finalised policy.
- Huawei’s LogicFolding is an announced approach; independent commercial-scale validation has not been published.
- China has not banned AI-model exports — in July 2026 it was reported to be considering options.
- Export controls have both restricted China’s frontier compute access and accelerated its domestic substitution drive.
Frequently asked questions
Conclusion: a race to build entire technology ecosystems
The first phase of the modern AI race looked like a contest between models. The next phase is much larger. The United States is trying to combine its advantages in advanced chip design, cloud computing, frontier AI companies and international partnerships into a complete, exportable technology ecosystem. China is responding by developing domestic accelerators, expanding its own models and cloud platforms, investing in semiconductor independence and building international technology relationships of its own.
But the world has not yet divided neatly into two camps. Countries still buy technologies from both sides. Supply chains still cross borders. Chinese models may appear on Western cloud platforms. US-designed chips depend on Asian factories and memory suppliers. So the decisive question through 2030 may not be “who builds the smartest model?” It may be: who can build the most complete, affordable and globally trusted AI ecosystem — from minerals and electricity all the way to the model?
The AI Cold War is not only a race for intelligence. It is a race for the infrastructure underneath intelligence.
Explore related timelines
⚠️ How AiTimeline tracks the US–China AI race
We classify each development as confirmed (official policy or signed agreement), announced (company or government announcement), reported (credible reporting not yet finalised), watchpoint (expected future development) or scenario (a possible outcome, not a forecast). Policy items are sourced to the US Bureau of Industry and Security, the US Departments of Commerce and State, and Chinese government releases; company figures to earnings statements; alliance and negotiation items to Reuters, CNBC and think-tank analysis including IISS, CSIS and ITIF. Membership counts for Pax Silica, the AI Opportunity Statement and WAICO change over time and should be re-verified before citing. This article is editorial and AI-assisted, politically neutral, and contains no investment advice or share-price forecasts. It explains the US rationale, the Chinese response, industry effects and third-country interests without endorsing any side. Information may contain inaccuracies; corrections are made as facts develop.