Tuesday, August 18, 2026HotTea archive editionVerified 12:23 AM PDT

7 minutes. Facts before narrative.

AI's center of gravity moved to financed capacity, cyber control and policy distribution.

The day put hard numbers around the AI stack. OpenAI's Ohio lease moved from reported talks to an announced Nvidia-backed capacity plan, Anthropic revenue reporting reset lab-scale comparisons, OpenAI framed cyber defense as a short operating window, and policy fights spread from federal model testing to state mental-health rules.

Published daily by 3:00 AM Pacific. No forced optimism. No manufactured panic.

Listen to today’s briefing

AI's center of gravity moved to financed capacity, cyber control and policy distribution.

The sourced HotTea edition, condensed into a chaptered morning podcast with verified audio and a full transcript.

OpenAI, Nvidia and SB Energy made the Ohio PORTS-Pike AI data-center plan official.

OpenAI said on August 17 that SB Energy will build, own and operate the PORTS-Pike campus under a 20-year lease, with OpenAI using capacity that will exclusively host Nvidia compute.

What happened

OpenAI said the first 800 megawatts are expected in 2028 using existing AEP infrastructure, while further development will require new power plants, transmission and financing. Nvidia said the initial deployment is designed for 4.25 IT-GW of AI-factory capacity, with OpenAI as customer for 8 IT-GW, and that Nvidia will invest $1.5 billion in SB Energy. Reuters, via Moneycontrol, reported Nvidia agreed to provide up to $105 billion of guarantee support for the project and that SoftBank and SB Energy plan at least 10 gigawatts of new power generation plus $4.2 billion of regional grid investment.

Why it matters

The deal makes the AI supply chain look less like software scaling and more like project finance: land, power, shells, grid upgrades, residual-value guarantees, chip exclusivity and long-dated lease commitments all have to clear before model demand becomes usable capacity.

What to watch

Whether permits, power plants, grid upgrades and financing close on schedule; whether the first 800 megawatts actually becomes available in 2028; whether Nvidia's credit support changes how lenders price AI infrastructure; and whether OpenAI demand produces enough cash flow to justify the fixed capacity.

The caveat

OpenAI and Nvidia are interested primary sources for capacity, community and demand claims. The Reuters account adds independent financing detail, but the final capital stack, project debt and return profile are not yet public.

Read this story on its own →

Worth knowing

The rest of the morning

Facts, pressure point, next evidence.

02

Anthropic was reported to have topped a $65 billion annualized revenue run rate before a possible IPO.

Reuters reported on August 17 that Anthropic's annual revenue run rate topped $65 billion by the end of July, citing a person familiar with the matter. Axios, citing Bloomberg figures, said preliminary second-quarter revenue exceeded $11.5 billion and noted OpenAI's latest reported run rate at $40 billion.

Pressure point Run rate is an extrapolation, not audited annual revenue. The figures come through people, investor materials and secondary reporting, and OpenAI and Anthropic may classify usage, credits, enterprise commitments and revenue timing differently.

Watch Whether Anthropic files IPO documents with audited revenue, gross margin, compute commitments and customer concentration; whether enterprise revenue remains durable after discounting; and whether model efficiency becomes a margin metric rather than only a benchmark claim.

WTVB / Thomson ReutersAxiosMarketWatch
Read article →
03

OpenAI told defenders they have a narrow window to upgrade before AI-enabled attackers scale.

OpenAI president Greg Brockman published 'The Defender's Window' on August 17, saying the OpenAI-Hugging Face incident showed the company had underestimated real-world cyber capabilities and that organizations should use AI to find and fix longstanding security flaws. Business Insider reported the post as a 10-point urgent defense agenda, while FT coverage placed the episode inside a broader debate over autonomous AI cyber risk.

Pressure point This is a high-risk security story. OpenAI is both the source of the disclosure and a vendor for defensive AI tools, so its urgency is not independent proof of product effectiveness. The edition does not publish exploit mechanics, live targets, or operational attack instructions.

Watch Whether OpenAI publishes more incident evidence, whether customers can verify AI-assisted remediation without adding new privilege risk, whether open-weight cyber-capable models ship at the end of August, and whether regulators move from voluntary testing to incident-reporting requirements.

OpenAIBusiness InsiderFinancial Times
Read article →
04

OpenAI funded 14 independent policy projects tied to economic opportunity and resilience.

OpenAI said on August 17 that it is awarding grants to 14 projects led by independent organizations to test, challenge and build on its April industrial-policy agenda. Daily.dev summarized the package as $1 million in grants and up to $1 million in API credits across organizations in the U.S., EU, Brazil, Singapore and South Korea.

Pressure point The projects may be led by independent organizations, but the funding source is OpenAI. That makes this a company-funded policy program, not neutral evidence that the proposed policies are sufficient for labor displacement, concentration or resilience risks.

Watch Which organizations receive money, whether grant terms preserve independence, whether outputs criticize OpenAI's own policy positions, and whether governments treat the work as evidence or lobbying-adjacent input.

OpenAIdaily.dev
Read article →
05

The White House was reported to be preparing to expand its frontier AI framework to open models.

WIRED reported that White House officials expect to update the administration's AI framework so sufficiently advanced open models would be subject to federal pre-release testing, alongside closed models from labs such as OpenAI and Anthropic. The report said the framework remains voluntary, but officials are weighing national-security concerns against the risk of disadvantaging U.S. open-model development.

Pressure point The framework itself has not been made public, and WIRED's account relies on people familiar with the matter. Until text, scope thresholds and test results are public, companies and open-model developers cannot audit whether this is safety evaluation, market signaling or informal permissioning.

Watch Whether the White House publishes the framework, whether open-weight developers receive the same review path as closed labs, whether testing is voluntary in practice, and whether Chinese open models become the political trigger for tighter U.S. controls.

WIREDEuropean Commission
Read article →
06

California's AI therapy bill showed states drawing a line between clinical support and chatbot psychotherapy.

CalMatters reported that California SB 903 would ban companies from advertising chatbots as therapy, block AI from making therapeutic decisions without licensed-professional review, and require disclosure and permission when AI is used for therapy-session recording or mental-health triage. Becker's reported that five states enacted AI-therapy chatbot restrictions in 2026, in addition to earlier Illinois and Nevada measures.

Pressure point The California bill is not yet a signed law, and state approaches differ. The policy premise is plausible, but it does not itself prove which AI mental-health tools are unsafe, clinically useful or properly supervised.

Watch Whether SB 903 advances, whether wrongful-death litigation produces discoverable safety evidence, whether state boards define permitted administrative uses, and whether federal agencies set a mental-health chatbot baseline.

CalMattersBecker's Behavioral Health
Read article →

The whole AI power map

AI is no longer a tech beat.

HotTea follows where AI moves power, money, labor, security, and state capacity—not only where a new model scores higher.

01

Politics & regulation

Elections, procurement, courts, surveillance, lobbying, and state power.

02

Economics & labor

Productivity, wages, employment, capital spending, concentration, and who captures the gains.

03

War & security

Autonomy, cyber operations, intelligence, targeting, export controls, and escalation risk.

04

AI geopolitics

Chips, energy, alliances, sovereign capability, supply chains, and strategic competition.

05

Markets & companies

Funding, revenue, margins, model economics, enterprise adoption, and infrastructure bets.

06

Science & society

Medicine, education, climate, culture, research, rights, and measurable public outcomes.

Financed control

The day's through-line was that AI scale now depends on institutions that can verify, finance and constrain it.

The strongest stories were not raw model demos. They were capacity contracts, revenue claims, cyber-control programs, policy funding, review frameworks and health-sector rules, each asking who bears risk when AI moves from product usage into infrastructure and public systems.

1

Infrastructure claims now need project-finance evidence: power, permits, grid upgrades, lease terms and residual-value exposure.

2

Cyber capability claims now need incident proof and access controls, not only vendor assurance that defensive use will dominate.

3

Policy and labor claims now need independence tests because the labs funding the debate are also beneficiaries of the outcomes.

The watchlist

Signals that could change the read

WatchlistWhether PORTS-Pike publishes permit, power, financing and phase-delivery milestones beyond the first 800 megawatts.Tracking
WatchlistWhether Anthropic's eventual IPO filing confirms reported revenue, margins and compute obligations.Tracking
WatchlistWhether OpenAI releases more auditable evidence from the Hugging Face incident without exposing operational details.Tracking
WatchlistWhether the White House publishes a frontier-model framework that treats open and closed models consistently.Tracking

How HotTea works

No optimism quota. No negativity quota. Just the honest read.

Every reported item links to its source. Company claims remain company claims. High-risk stories require stronger corroboration. Material caveats, conflicts, and unknowns stay in the story. HotTea’s interpretation is visibly separated so readers can disagree without losing the facts.

Edition validated · 6 stories · 15 unique sources

Audit today’s sources →

Tomorrow’s signal, before tomorrow’s noise

Open HotTea. Know what changed.

A new verified edition every morning. If the evidence or release gate fails, the last verified briefing stays live.

Back to today’s top ↑