Sections
00:00 What we're covering today
00:21 1. AI backers are putting money behind data-center fights
01:29 2. The Pentagon launched ChatGPT Mil and Grok for Government
02:38 3. HUMAIN says its AMD and Cisco compute is live
03:41 4. SLB agreed to buy data-center cooling company Kelvion for $4.1 billion
04:38 5. Europe signed a €387.8 million contract for LUMI-AI
05:53 Visit Hot Tea

Disclosure
Narration uses an AI-generated voice.

Transcript
Welcome to Hot Tea for Tuesday, September 1, 2026. Today starts with data center politics in Kansas, Ohio, and Wisconsin. After that, the briefing turns to Pentagon model access, Saudi compute, data center cooling, and Europe's public supercomputer contract.

A I backers are putting money behind data center fights. Axios reported that Build American A I has about $50 million in cash. The group will spend millions on ads, local outreach, research, and public education. The campaign starts in Kansas, Ohio, and Wisconsin, where data center development has become an election issue. TechCrunch reported that the group is affiliated with Leading the Future. Marc Andreessen, Ben Horowitz, and OpenAI president Greg Brockman have funded that super PAC. OpenAI has said it does not direct Leading the Future or see its operations. The pressure: The campaign has announced its budget and target states. It has not published complete ad buys or evidence that it changed voters. A super PAC affiliation does not prove that every funder controls this campaign. OpenAI has publicly separated the company from Leading the Future's work. What to watch: Watch public ad libraries, state campaign filings, utility hearings, and 2027 legislative proposals in Kansas, Ohio, and Wisconsin. Compare promised jobs and tax revenue with binding power, water, and ratepayer terms. Ad reach is not the useful result. Clearer local project rules and visible costs are.

The Pentagon launched ChatGPT Mil and Grok for Government. The Pentagon launched ChatGPT Mil and Starshield AI's Grok for Government on GenAI.mil on August 31. TechCrunch reported that both tools join Google's Gemini in the secure portal. The department said Grok has Impact Level 5 approval for controlled unclassified information. The department also said GenAI.mil has more than 1.7 million unique users across a workforce of more than 3 million people. OpenAI says data processed through its government deployment stays inside the government environment and does not train its public models. The Pentagon's claims about faster work and better missions are official claims. Public releases do not show independent security tests, refusal logs, incident records, or model-level audit results. Impact Level 5 approval does not establish safe use on classified networks or in weapons. The launch also makes Anthropic's absence from the portal harder to ignore. Watch usage reports, security incidents, audit rules, retention policies, and model-specific controls. The Pentagon should separate routine drafting and logistics from decisions that affect surveillance, targeting, or force. Independent tests need to show whether the same controls work for every vendor in the portal.

HUMAIN says its AMD and Cisco compute is live. AMD, Cisco, and Saudi state-backed HUMAIN said their AI system is now serving customers. It uses AMD Instinct MI355X GPUs, AMD EPYC CPUs, and Cisco networking. HUMAIN will offer the system as a GPU service for model training and inference. The companies plan to start a second phase of up to 250 megawatts in 2027. Their joint venture still targets up to one gigawatt by 2030. This is a company claim in two versions of the same joint announcement. The companies did not disclose live megawatt capacity, GPU count, customer names, utilization, service prices, or measured performance. The 250-megawatt and one-gigawatt figures are plans, not operating capacity. Watch for named customers, live capacity, utilization, pricing, and independent performance tests. The second phase should disclose construction milestones and how much capacity has firm demand. Export approvals and power commitments will decide how much of the 2030 target becomes usable compute.

SLB agreed to buy data center cooling company Kelvion for $4.1 billion. SLB agreed to acquire Kelvion for about $3.4 billion in cash and assume about $700 million of debt. Reuters reported a total value of $4.1 billion including debt. Kelvion makes thermal-management and heat-exchange equipment. SLB expects the deal to close in the first half of 2027 after customary approvals. The transaction has not closed. SLB's revenue, margin, and synergy figures are company projections. The purchase adds cooling equipment to SLB's modular data center work. SLB still has to integrate Kelvion and win orders at the prices behind the deal. Watch regulatory approvals, financing, customer retention, and the closing date. After closing, compare actual data center revenue and margins with SLB's 2028 targets. Order growth and installed cooling capacity will show whether SLB bought real demand or a larger sales pitch.

Europe signed a €387.8 million contract for LUMI-A I. The European High Performance Computing Joint Undertaking signed a €387.8 million contract with Bull to deliver LUMI-A I in Kajaani, Finland. The EU will fund half, and the six-country LUMI A I Factory group will fund the rest. AMD says the system will use Instinct MI430X GPUs and 256-core sixth-generation EPYC CPUs. Deployment is scheduled for the second half of 2027. The system is meant for research, startups, industry, A I training, inference, and scientific simulation. The pressure: The contract is real, but no one has measured the performance yet. EuroHPC and AMD expect ten times the A I capacity and almost twice the high-performance computing capacity of the current LUMI system. They have not published the GPU count, power draw, benchmark method, or delivered performance. Installation is still more than a year away. What to watch: Watch the final system design, GPU count, power budget, access rules, and build schedule. Public users need clear prices and allocation rules. The first acceptance tests in 2027 should measure training, inference, and scientific-computing performance against the current LUMI system.

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.