Sections
00:00 What we're covering today
00:26 1. Washington wants Mexico's AI hardware to use fewer Chinese parts
02:40 2. Oracle raised its restructuring estimate to as much as $2.8 billion as another layoff round began
04:23 3. AI's newest demand signal is steel-toe boots
05:56 4. China can stop tech workers from leaving under new security rules
07:31 5. An AI agent passed 77% of runs but went five for five on only 53% of tasks
09:11 Visit Hot Tea

Disclosure
Narration uses an AI-generated voice.

Transcript
Welcome to Hot Tea for Wednesday, September 16, 2026. Washington wants Mexico's A I hardware to use fewer Chinese parts. Today's briefing covers the lead, companies and labor, economics and infrastructure, geopolitics and export controls, and models and products. The United States is pressing Mexico to accept stricter rules of origin for A I hardware.

Washington wants Mexico's A I hardware to use fewer Chinese parts. The United States is pressing Mexico to accept stricter rules of origin for A I hardware. The Wall Street Journal reported Tuesday that the proposal covers chips, servers and related equipment crossing the southern border. Washington wants to stop Chinese companies and other foreign suppliers from using Mexico to avoid U S. tariffs. The proposal follows the model used for auto trade. Under the U S.-Mexico-Canada Agreement, vehicles need 75 percent North American content to receive preferential treatment. Washington has not disclosed the threshold it wants for A I hardware. The White House and the Office of the U S. Trade Representative did not comment to the Journal. The two countries announced related work in March. U S. Trade Representative Jamieson Greer and Mexican Economy Secretary Marcelo Ebrard said their teams would examine rules of origin and economic security. They also said the teams would study ways to limit what they called non-market inputs in North American supply chains. The new proposal brings A I hardware into those talks. Mexico exported $82.9 billion in computer servers during the first half of 2026. Reports on the proposal cited figures from S&P Global Market Intelligence. The proposal remains under negotiation. Washington has not published a content threshold, a product list or draft legal text. The export figure describes computer servers, not only servers used for A I. Stricter origin rules would not by themselves stop every Chinese component from entering North American supply chains. A written U S. proposal would answer three open questions. Which products would the rule cover? How much North American content would they need? The text would also need to explain how officials calculate that share. It would need to say whether Chinese-owned factories in Mexico count as North American production. The next record from the USMCA talks could show whether Mexico will accept A I hardware rules in return for relief elsewhere. That relief could cover vehicle tariffs or other disputes. Supplier announcements would offer the clearest sign that companies expect to change what they buy. Watch for changes involving chips, memory, networking gear and server parts.

Oracle raised its restructuring estimate to as much as $2.8 billion as another layoff round began. Oracle began another round of layoffs Monday, according to three affected people and an employee notification reviewed by Business Insider. The company did not disclose how many jobs it cut. Oracle had about 141,000 employees before this round. Its workforce fell by 21,000 during the fiscal year that ended in May. A quarterly filing shows the broader cost. Oracle estimated up to $2.1 billion for its fiscal 2026 restructuring plan as of August 31. Management later added about $700 million for further actions. The filing says the plan includes operational changes tied to adopting and integrating A I across some functions. Employee severance accounted for most costs recorded so far. The total estimate now reaches $2.8 billion. Oracle is cutting jobs while spending heavily on A I infrastructure. Its filing does not connect every eliminated role to A I. The company also gave no total or team breakdown for the latest layoff round. Strong cloud demand means the layoffs alone do not show that Oracle's A I investment is failing. Workers still bear part of the cost. They are losing jobs while Oracle puts more money into data centers and equipment. Oracle's next filing could disclose more restructuring charges, workforce changes and capital spending. A team-level count would show where the cuts landed. Cash flow, debt issuance and data center commitments will show how Oracle pays for the buildout. Another large cost reduction would show that the current plan was not enough. Hiring by function would show which skills Oracle adds while other roles disappear.

A I's newest demand signal is steel-toe boots. Boot Barn says data-center construction crews account for a growing share of its bulk workwear orders. CEO John Hazen described the pattern at a Goldman Sachs conference. The retailer tracks nearby projects and changes local inventory for each stage of construction. Crews clearing a site need regular work boots. Once building starts, crews need steel-toe boots. Boot Barn has nearly 600 stores, and its work-boot business has posted comparable sales growth for five straight quarters. Brunt Workwear told Reuters that California and Texas are among its strongest markets as A I infrastructure investment expands. The Bureau of Labor Statistics reported 22,000 more U S. construction jobs in August. The official jobs report doesn't say how many came from data centers. Boot orders show that crews are working, but they don't measure A I construction across the country. Neither retailer disclosed data-center revenue, order volume, or the share of growth tied to those projects. Data centers also face local resistance over power, water, and land. Construction can create many jobs before a facility opens, but a completed data center can operate with a much smaller permanent workforce. Retailers could disclose revenue from bulk data-center orders. State construction payrolls and electrical-trade employment could show where the projects create work. Suppliers could feel cancellations before those losses appear in national data. Local permits, utility connections, and completed facilities will show which announced projects are actually under construction.

China can stop tech workers from leaving under new security rules. China's new exit-and-entry rules took effect Tuesday. Authorities can stop a citizen from leaving if an export-control or technology-trade violation may endanger national industrial or technological security. Other provisions apply to some citizens returning after illegal or criminal acts abroad. If those acts harmed national security or national interests, authorities can impose an exit ban lasting six months to three years. Foreign nationals can receive entry bans of one to five years for making false visa statements. Reuters reported that travel limits already applied to senior officials and state executives with access to confidential information. The new rules tie technology security and export controls more directly to decisions at the border. The rules took effect Tuesday, so there is no enforcement record yet. The language on technology security leaves room for officials to decide when it applies. The public text does not define which jobs, data or technologies can trigger a ban. Authorities may use the rules only for serious export-control cases. They may also use them to restrict ordinary private-sector movement. Enforcement records are needed before the evidence supports either reading. China could publish guidance that names covered technologies and the agencies allowed to request an exit ban. Court challenges, company notices and consular cases would show how officials use the rules. Semiconductor, battery, solar and rare-earth employers will provide early tests. Their workers carry the technical knowledge that export rules are meant to protect.

An A I agent passed 77% of runs but went five for five on only 53% of tasks. IBM researchers tested a ReAct agent powered by GPT-4.1 on 168 AppWorld tasks. They ran every task five times. The agent passed 77.4 percent of all runs, but it passed all five runs on only 53.0 percent of the tasks. The researchers call the 24.4-point difference the consistency gap. Their method reviews the agent's recorded steps and reruns individual decision points. It flags decisions that are likely to change on another run. The method then stores short guidelines about those weak points in memory. On the same tasks, the method raised the share that passed all five runs by 16 points, to 69.0 percent. It raised that share by 13 points on similar tasks. The average pass rate also increased. The authors tested one benchmark, two model backends and one agent pattern. No independent team has reproduced IBM's research claim. The paper is a preprint, and the same researchers built and tested the method. AppWorld uses controlled API tasks rather than live customer work. Requiring five successful runs may be stricter than some production settings need. It still catches failures that a one-run benchmark can miss. Independent teams could repeat the test on browser, coding and enterprise agents. Production teams could also rerun the same task over time, after model updates and during real tool failures. Teams could publish average success rates alongside repeat success rates. Adding repeatability scores to model cards would help buyers compare reliability without depending on one successful run.

That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.