The question every institutional portfolio manager is sitting with this week is not whether Washington will tax semiconductors harder. It already said it would. The question is which slice of the $1 trillion-plus data center spending cycle absorbs the hit, and who passes it on to whom.
Presidential Proclamation 11002, signed January 14, 2026, imposed a 25% tariff on certain advanced computing chips and derivative products. The tariff takes effect for goods entered on or after January 15, 2026. The proclamation did not label the action “Phase 1,” but it did direct Commerce and USTR to pursue negotiations and noted the President may consider imposing significant tariffs on semiconductors and derivative products in the future. The same document carved out exemption categories for chips used in U.S. data centers, repairs or replacements performed in the United States, research and development in the United States, startups, non-data center consumer applications, non-data center civil industrial applications, and U.S. public sector applications, plus other uses the Secretary determines contribute to strengthening the U.S. technology supply chain or domestic manufacturing capacity. Those exemptions are now the live variable. Politico reported Thursday, citing eight people familiar with the drafting, that a Phase 2 is being designed to extend duties to finished hardware: data center servers, laptops, and gaming consoles.
Why Wall Street Cares
The global data center capex outlook has been raised to more than $1 trillion for 2026, with the top four U.S. cloud providers increasing data center capex 78 percent year-over-year, according to Dell’Oro Group. That spending baseline is what makes trade policy suddenly feel like a cost-of-goods problem rather than a geopolitical background condition. The cloud buildout has moved from an income-statement story to a supply-chain story. Investors used to read cloud growth through revenue and margins; in 2026, the sharper signal is capital expenditure: where the money goes, how fast it converts into usable capacity.
Server OEM margins are thin enough that even modest duty exposure lands directly on the income statement. Claims about “5 to 8 percent EBIT” at SMCI and “4 to 6 percent” on Dell AI infrastructure servers are not consistently supported in company filings or earnings materials, so the right frame is simpler: these are competitive, high-mix hardware businesses where a new tariff layer on finished hardware does not leave much room to absorb before it becomes a pricing conversation with hyperscaler customers who are already managing memory cost inflation.
The Bull Case for Pressing Ahead
At the center of the proposed framework is a mechanism Politico described as favored by Commerce Secretary Howard Lutnick: tying duty-free chip imports to a company’s committed U.S. production, effectively pegging an exemption quota to domestic manufacturing investment. For companies already deep in U.S. reshoring commitments, this is a moat, not a tax. Apple, for instance, has publicly said its U.S. investment commitment totals $600 billion over four years. Separately, the White House has said Apple secured an exemption from the chip tariff. Firms with that kind of bilateral arrangement priced Phase 2 months ago.
The Bear Case
The bear case is structural, not cyclical. Politico reported that tech lobbyists have met with Lutnick and Bureau of Industry and Security undersecretary Jeffrey Kessler with growing frequency since the start of summer, but that three sources said recent talks moved against the industry. One person involved put the domestic manufacturing build-out at more than five years, longer than any phase-in the administration has allowed on previous tariff rounds. That timeline mismatch is the crux. Companies cannot earn their way into the exemption fast enough to avoid near-term cost exposure.
CSIS analysis found that a 100 percent tariff on all semiconductors and products containing them would likely impose an additional $1.4 trillion burden on the buildout. Even more moderate scenarios compound an already strained bill of materials. Rising memory and storage pricing has already substantially increased overall server system costs this year. A finished-goods tariff stacks on top of that, not instead of it.
What Investors Are Missing
The debate has centered on hyperscaler capex resilience. The overlooked implication is order-book composition at server OEMs. IDC data for the first quarter of 2026 shows Dell leading worldwide server vendor revenue, followed by Supermicro and Lenovo, while ODM Direct still accounted for the largest share of server revenue. ODM Direct vendors, which carry no brand premium and operate on razor margins, have almost no mechanism to absorb a finished-hardware duty. If Phase 2 lands without the data center carve-out, the ODM channel could compress sharply, redirecting more hyperscaler procurement toward branded OEMs with exemption-eligible domestic footprints. Dell and HP have domestic assembly operations that many ODM-heavy supply chains in Asia do not.
The core contradiction of this policy is that while the U.S. wants to drive reshoring of semiconductor manufacturing, it is simultaneously relying on large imports of high-end chips to build AI data centers. If the scope of exemptions tightens, both the cost and uncertainty of AI infrastructure investment could rise. NVDA and AVGO sit upstream of all of this: their silicon gets bought regardless, but any slowdown in hyperscaler deployment timelines eventually shows up in order cadence.
Stocks to Watch
Dell (DELL): Leads server OEM revenue and has U.S. assembly exposure that could qualify for exemption relief. The key question is whether its hyperscaler contract structure allows tariff pass-through.
Super Micro Computer (SMCI): Thin operating margins make it the most exposed branded OEM if finished-hardware duties arrive without a carve-out. Watch for any guidance revision tied to procurement costs.
HP Inc. (HPQ): Consumer PC and enterprise hardware exposure across laptops and servers puts it squarely in the product categories under discussion. Less AI-server-concentrated than Dell, but broader surface area.
NVIDIA (NVDA): Capex growth is expected to accelerate further in the second half of 2026, with NVIDIA’s next-generation Vera Rubin platform beginning shipments and ramping through the back half. Upstream demand for its silicon is structurally insulated from finished-goods tariffs for now, but any hyperscaler deployment delay is a duration risk on order backlog.
Broadcom (AVGO): Custom ASIC demand is growing faster than merchant GPUs this year. A policy environment that rewards domestic investment could accelerate hyperscaler insourcing of silicon, which plays directly to Broadcom’s co-design model with the large cloud operators.
