Trump Chip Tariff Phase 2 Could Hit Data Center Servers, Reshaping AI Build Costs

The Trump administration is considering a second round of semiconductor tariffs that would eliminate the data center, startup, and consumer exemptions that have shielded the AI infrastructure buildout since January 2026. Phase 2 discussions, reported by Politico and confirmed by eight people familiar with the talks, would extend duties to data center servers, laptops, and gaming consoles built with the chips already subject to Phase 1, marking a shift from targeted tariffs to a broader regime that could reshape the economics of AI data center construction in the United States.

Phase 2 would reverse the January carve-outs that protected AI data centers

Presidential Proclamation 11002, signed January 14, 2026, imposed a 25% tariff on specific advanced AI accelerators like the Nvidia H200 and AMD MI325X, but carved out six exemption categories covering chips used in U.S. data centers, research and development, startups, repairs, consumer devices, and public-sector uses. According to the parent article, those exemptions were a deliberate structural choice to allow the AI infrastructure buildout to proceed without cost disruption. Commerce officials have indicated in private talks that the Phase 1 exemptions may not carry into Phase 2, according to four of the Politico sources.

The same day the Phase 2 deliberations were reported, SK Hynix broke ground on its $4-billion-plus advanced-packaging facility at Purdue Research Park in Indiana, the company's first U.S. high-bandwidth memory production hub. That juxtaposition captures the central tension: the administration is courting foreign investment while weighing tariffs that could cost those same investors billions more per year.

How the quota-based tariff framework falls short

Commerce Secretary Howard Lutnick favors a framework where tariff-free chip import allowances are tied proportionally to each company's committed investment in U.S. semiconductor manufacturing. A company building domestic capacity would receive duty-free import rights scaled to its domestic output. But the investment-linked quota structure has a specific, quantifiable problem: the duty-free volumes the formula would generate are structurally insufficient to cover what American cloud companies are procuring during the current record AI spending run.

TSMC controls 73% of the foundry market and essentially 100% of the leading-edge logic chip production powering Nvidia's AI accelerators and custom silicon from Amazon, Google, and Microsoft. TSMC has committed $265 billion to Arizona, yet projects only about 30% of its most advanced fabrication capacity will be located there at full build-out. Under the existing Taiwan trade agreement, TSMC can import 2.5 times its current U.S. manufacturing capacity duty-free while plants are under construction, tightening to 1.5 times once operational. With only 30% of capacity in Arizona, the quota math leaves the majority of what hyperscalers need subject to potential tariff exposure.

On the memory side, SK Hynix controls roughly 50% of the high-bandwidth memory market, and the HBM market was sold out through 2026 as of recent reporting. The Indiana plant, with mass production of next-generation HBM4E chips targeted for the second half of 2029, will assemble DRAM chips from Korea, meaning even the U.S.-based HBM packaging will depend on Korean wafer production for years.

What this means for AI builders and hyperscalers

If Phase 2 eliminates the data center exemptions, every chip used in a U.S. data center server would become subject to import duties, directly raising the cost of building and operating AI infrastructure. Jonathan McHale, digital policy chief at the Computer and Communications Industry Association (whose membership includes Amazon, Google, and Meta), compared the data center buildout to "building the transcontinental railroad" and warned that added cost and unpredictability put that entire investment at risk.

The consumer electronics supply chain is already under pressure from a global memory shortage that pushed Nintendo, Sony, and Microsoft to raise gaming hardware prices. Additional tariffs on finished electronics would compound those cost increases at the retail level on top of existing cost pressures.

Caveats and uncertainties

The policy is still developing. The White House spokesperson told Politico that unannounced tariff plans should be treated as speculation. The Commerce Department did not respond to comment requests. The July 1 Commerce Department report on the data center semiconductor market, mandated by Proclamation 11002, has been completed but not made public. The specific mechanism, rates, exemption structure, and timeline remain in flux and could change substantially. One industry source placed the timeline for meaningful domestic manufacturing build-out at more than five years, longer than any phase-in period the administration has allowed on previous tariff rounds.

For now, the central question is whether the January exemptions survive Phase 2. The administration's two stated goals, reshoring and AI leadership, pull in opposite directions, and the Phase 2 framework will determine how much of the AI buildout's cost structure lands on the domestic side of the supply chain.