China-US Trade War 2026: Every Major Tariff and Chip Ban

The China-US trade conflict has moved in two simultaneous directions since January 2025: tariff escalation on goods, and tightening export controls on advanced semiconductors. Both tracks are running concurrently. This timeline covers every major development from the start of the Trump second term through June 2026, with a developer-specific analysis of what each event means for infrastructure costs, chip availability, and the global AI supply chain split.

January–February 2025: Second Term Starts With Tariff Threats

Donald Trump returned to the White House on January 20, 2025, having campaigned explicitly on tariff escalation against China. Within weeks, the administration moved from campaign promise to policy.

January 2025: Executive orders directing a review of US-China trade relationships and all existing trade agreements. The Commerce Department began drafting updated export control rules covering AI accelerators, advanced memory, and semiconductor manufacturing equipment.

February 2025: The first round of additional tariffs went into effect: 10% on Chinese goods, layered on top of existing Section 301 tariffs from the first Trump administration that had been maintained through the Biden years. At this stage, the additional tariff burden on most Chinese electronics reached approximately 35-45% combined when stacked with pre-existing duties.

China's immediate response was measured: tariffs on selected US agricultural products (soybeans, LNG, coal), and an announcement that it was adding several US defense contractors to its Unreliable Entity List.

March–April 2025: Liberation Day and the 145% Escalation

April 2, 2025 is the date that changed global trade calculations. The Trump administration announced what it called "Liberation Day" tariffs — a sweeping tariff schedule covering most US trading partners, framed as reciprocal duties that matched each country's effective trade barriers against US goods.

For China, the Liberation Day tariffs set a new base rate of 34% on all Chinese goods. Combined with existing Section 301 tariffs, the total effective tariff burden on most Chinese goods reached 84-104% depending on the product category.

China immediately retaliated with its own 34% tariff on US goods, triggering a second round of US escalation. By mid-April, the administration had pushed the China-specific rate to 145% — the highest tariff level imposed on China in the post-WWII era.

What 145% means in practice: A server rack component that cost $1,000 at the Chinese factory gate faced $1,450 in tariffs before reaching a US data center. For consumer electronics, the effect was immediate and visible: laptop prices increased $100-300 within weeks, and server hardware costs rose 15-25% depending on the component mix. Cloud providers absorbed some of this on multi-year contracts but signaled price increases at next renewal.

April 9, 2025: The administration announced a 90-day pause on Liberation Day tariffs for most trading partners, dropping them back to 10%. China was explicitly excluded from the pause. The 145% rate remained in place on Chinese goods.

May 2025: The US-China Trade Truce

May 12, 2025: After negotiations in Geneva, the United States and China announced a 90-day mutual pause in tariff escalation. The deal reduced US tariffs on Chinese goods from 145% to 30% and reduced Chinese tariffs on US goods from their peak retaliation level. The tariff rate of 30% is still substantially above the pre-Trump levels, but it represented a de-escalation significant enough to move markets: equities rallied, and shipping bookings between China and the US spiked as manufacturers rushed to move goods during the window.

The 30% rate was explicitly framed as temporary — a pause to allow negotiations, not a resolution. Neither side agreed to a framework for permanent tariff levels. The 90-day clock started May 12.

June–August 2025: Chip Export Controls Tighten Independent of Tariff Track

While tariff negotiations proceeded, the export control track continued independently. The Bureau of Industry and Security (BIS) at Commerce maintained and expanded restrictions on advanced semiconductor exports to China regardless of the tariff status.

The key distinction between tariffs and export controls: Tariffs apply across-the-board to categories of goods and can be adjusted by executive action. Export controls on semiconductors are embedded in US law through the Export Administration Regulations (EAR) and the CHIPS Act framework — they require regulatory process to change and operate separately from tariff negotiations. A trade truce does not affect export controls.

H20 chip status: The Nvidia H20, which Nvidia designed specifically to meet the performance thresholds of prior export control rules, was already under restriction by early 2025. BIS had determined that the H20's combination of memory bandwidth and interconnect speed exceeded the effective ceiling for unrestricted export. Nvidia's ability to sell any meaningful AI accelerator into China for AI training applications is effectively zero under the current rules.

The capability ceiling: BIS sets export thresholds based on combinations of performance parameters. The rules target the "combined interconnect bandwidth" metric that determines how efficiently chips can work together in training clusters. Nvidia's attempts to design under-the-line chips (first A800/H800, then H20) have each been swept up in successive rule revisions as BIS raised the analytical sophistication of the restrictions.

September 2025–January 2026: China Responds on Rare Earths and Materials

China's strategic response to semiconductor export controls was not mirror-for-mirror chip restrictions. China does not produce competitive AI accelerators to export-ban. Instead, it moved on materials:

Rare earth export restrictions: China controls approximately 60-70% of global rare earth processing capacity. Starting in late 2025, it imposed licensing requirements on exports of gallium, germanium, graphite, and other materials critical to semiconductor manufacturing. These are not full bans — they are a license-and-restrict framework that creates friction, delays, and pricing pressure for chip manufacturers outside China.

Impact on chip fabs: TSMC, Samsung, and Intel use gallium and germanium in compound semiconductor processes. The licensing requirements added 2-4 weeks to material procurement timelines and increased spot prices 15-30%. ASML (which makes the EUV lithography machines that produce advanced chips) flagged the restrictions as a concern for long-term supply planning.

January 2026: The Commerce Department published an updated Export Administration Regulations rule tightening controls on "advanced computing items" — a category that captures AI accelerators above specific performance thresholds and now explicitly includes inference-optimized chips (not just training accelerators). This rule affected not just Nvidia but also the category of chips Nvidia designs for inference markets (the L40S, H100 NVL) and covered AMD MI300X variants above the threshold.

February–May 2026: China's $295 Billion National AI Grid

May 2026: Beijing formally approved nine categories of domestically developed AI chips for deployment across government and security-sensitive sectors — the foundational step for the national grid. The chips approved include Huawei's Ascend 910B and 910C series, Biren Technology's BR100, Moore Threads' MTT S80, and Alibaba Cloud's Hanguang 800.

The approval was followed immediately by the announcement of the $295 billion (2 trillion yuan) national AI computing grid planned for completion by 2028. The grid requires 80% domestic chip sourcing, which structurally excludes Nvidia and AMD regardless of export control status. Even if the US tomorrow removed all export controls on AI chips to China, the Chinese national grid mandate would prevent those chips from being the primary supply.

This represents the most significant structural consequence of the trade and export control conflict: China is not simply waiting for restrictions to lift. It is building an independent AI infrastructure stack designed to function without US semiconductor input.

For the full analysis: China $295 billion AI grid: what it means and who wins.

Full Trade War Timeline — Key Dates

What This Means for Developers and Infrastructure

The two-track conflict — tariffs on goods, export controls on chips — creates distinct infrastructure impacts:

Hardware cost: Server components from Chinese manufacturers (which supply a large fraction of rack hardware, cooling components, and lower-level electronics) are effectively 30-145% more expensive depending on whether the truce holds. Cloud providers with multi-year hardware procurement contracts absorbed the initial shock; spot hardware purchases for self-hosted infrastructure reflect the tariff directly.

Chip availability: For AI workloads, the chip impact is bifurcated. Developers building on Nvidia and AMD in the US, EU, Japan, or allied markets have no access restrictions — supply is constrained only by production capacity, not policy. Developers building for China-accessible deployment cannot use the same hardware stack. The Chinese market now routes through Huawei Ascend and domestic alternatives.

Supply chain split: The export control regime has accelerated a bifurcation that was structurally inevitable. By 2026, there are two AI hardware ecosystems: the US-and-allies ecosystem centered on Nvidia/AMD/TSMC, and the China ecosystem centered on Huawei Ascend/SMIC. Applications and models optimized for one do not run efficiently on the other. This is not a temporary artifact of trade tensions — it is a durable architectural divide.

The HBM constraint on China: Even with domestic chip approvals, China's AI buildout faces a materials ceiling. High-bandwidth memory (HBM) — the stacked memory that AI accelerators require — is produced almost entirely by SK Hynix, Samsung, and Micron, all US-allied companies. China's domestic HBM capacity is minimal. This means the Chinese national grid runs at lower memory bandwidth than equivalent Nvidia-based clusters, limiting training throughput for large models. See the developer-focused breakdown of the chip supply chain split.

Our Analysis: Is the Trade War Ending or Deepening?

The 30% tariff rate from the May 2025 truce is where things stood as of June 2026. The 90-day window from May 12 expired in August 2025 — but both sides have maintained the status quo without formal renewal or formal collapse. Tariff policy is now in a de facto holding pattern.

The export control track is a different story. BIS rule updates have continued regardless of tariff status. The direction is unambiguous: each revision tightens the performance threshold, captures more chip categories, and reduces Nvidia and AMD's ability to design products that can legally sell into China's AI market. This is not a trade negotiation — it is industrial policy with bipartisan US support.

China's response — building a $295 billion domestic compute grid and approving domestic chips for government procurement — is the most significant strategic move of 2026. It signals that China is not betting on the export controls being lifted. It is investing in the assumption that the controls are permanent, and building the industrial capacity to make that assumption survivable.

For developers: the practical consequence is that if you build AI infrastructure today, you are implicitly choosing a camp. Nvidia-based infrastructure is the global standard for performance, but it is structurally excluded from the Chinese market. Huawei Ascend-based infrastructure can access the Chinese market and the national grid procurement, but runs at lower performance on current benchmarks and has HBM supply constraints.

For cloud providers: AWS, Azure, and Google Cloud can serve the US and allied markets on Nvidia hardware without restriction. None of them can build a meaningful China-targeted GPU cloud on the same chips. The hyperscalers with China exposure (Azure China, Alibaba Cloud, Tencent Cloud) are building on domestic silicon.

Key Takeaways

• Liberation Day (Apr 2, 2025): Trump imposed 34% tariffs on China, escalated to 145% within weeks — the highest rate since the post-WWII era
• US-China Truce (May 12, 2025): Tariffs pulled back to 30% for a 90-day window; rate has remained in de facto status quo into 2026
• Server hardware impact: Component costs from Chinese manufacturers rose 15-25% at peak 145%; at 30%, the increase is more modest but still material for large-scale infrastructure
• AI chip export controls operate independently of tariff negotiations: Nvidia H100/H200/B100, AMD MI300X all restricted from China AI market regardless of tariff status
• China's strategic response: $295 billion national AI grid mandating 80% domestic chips — building infrastructure designed to function without US silicon, not waiting for controls to lift
• The HBM bottleneck: China's domestic chip production is improving but still limited by SK Hynix/Samsung/Micron HBM dependency that export controls indirectly constrain
• For developers: two AI hardware ecosystems are forming — US-and-allies on Nvidia/AMD, China on Huawei Ascend/domestic; optimize for one, work harder on the other