Anthropic’s IPO bet meets a warning on AI acceleration

China (6)

• U.S. intelligence reportedly identified Alice Huang, a Megaspeed executive and Jensen Huang’s romantic partner, as a Chinese Ministry of State Security agent. Megaspeed provided Chinese AI labs remote access to restricted NVIDIA GPUs; officials also suspect chip smuggling, an allegation not proved.
• Critics argue investigators must examine whether Huang’s personal relationship affected NVIDIA’s sales to Megaspeed or his advocacy against China chip restrictions. Dmitry Alperovitch alleges Megaspeed funneled billions in computing power to Chinese AI labs; the transcript presents that as his claim, not an established finding.
• Andrew Liu argues the scandal could undermine Huang’s opposition to tougher China chip controls, reverse H200 export approvals, and put NVIDIA under congressional oversight. Liu frames the reporting as a possible strategic move to limit the company’s autonomy over chip destinations.
• The New York Times reported that Commerce approved H200 exports to Chinese companies with links to sanctioned firms, including shell companies with little visible business. NVIDIA announced Southeast Asian data centers must join its fleet intelligence system, which tracks GPU locations.
• Marco Rubio reportedly oversaw the classified memo about Huang and Alice Huang before Trump’s China trip. Trump said he had not spoken to Jensen Huang, called him a fine person, and said he might invite him on a later Asia trip; Republican and Democratic lawmakers signaled scrutiny.
• Sean McGuire and Bill Ackman defended Huang while acknowledging his mistake; Alex Brunzini Vendor and Matthew Iglesias rejected the idea that lack of harmful intent resolves the issue. Samir argued Huang’s decisions could have endangered U.S. interests, exposing a risk if NVIDIA depends too heavily on one leader.

Big Tech (1)

• Jensen Huang admitted in a company-wide email that he had been unfaithful to his wife, but did not address the geopolitical allegations. NVIDIA’s board opened an independent investigation, said it had no reason to believe secrets or national security were compromised, and expressed confidence in Huang.

Markets (3)

• NVIDIA futures initially erased about $300 billion in market value, then recovered within minutes; a Kobeissi letter described a $650 billion implied-market-cap swing. Shares later fell as much as 8.8% before recovering roughly half the loss, with no clear spillover to broader tech.
• OpenAI closed a funding round below its initial target, securing commitments from institutional investors and planning to continue fundraising from wealthy individuals and retail investors. The round included KOTU, T. Rowe Price, SoftBank, Sequoia, Founders Fund, A16Z, and UAE backers.
• SoftBank invested only hundreds of millions in OpenAI’s latest round after investing heavily in prior rounds, while its shares fell sharply. A mystery seller initially linked to SoftBank was identified as ING; Elon Musk separately claimed SoftBank was defaulting on turbine orders.

AI Infrastructure (2)

• Microsoft declined to renew its Oracle compute rental contract, reportedly because its aging GPUs no longer met its needs. The move supports the view that powered sites with current-generation chips are more valuable than delayed projects, while Microsoft also reconsidered other costly, delayed data center agreements.
• Google took over Crusoe’s delayed Wyoming data center project after tenant, energy, construction, and cost problems. Google had been a prospective tenant and plans to invest in the site; Crusoe remains involved in separate Texas projects for Google and OpenAI.

Models (1)

• DeepMind’s Pushmeet Kohli, Meta’s Tom Mann, and OpenAI’s Greg Brockman posted about accelerated scientific discovery as new internal models come online. Andrew Curran argued that major labs had reached autonomous research, with models choosing research directions; Riley Brown expected visible developments soon.

Enterprise (6)

• Nathaniel Whittemore says business AI adoption is accelerating despite headlines about companies restricting employee use, including Apple limiting engineers’ daily AI spending. Agents are shifting enterprise spending from per-seat access toward tokens and work completed.
• Menlo’s enterprise AI estimates put annual spending at $33 billion in 2025 and nearly $177 billion in 2026; its reported growth multiples were 5x from 2023 to 2024 and 3.9x from 2024 to 2025. The survey covered about 400 U.S. enterprise decision makers responsible for roughly $4 billion in AI spend.
• Menlo found enterprise AI use rose from 55% to 65% in a year, while average spending per AI-using enterprise grew 4.7x. Whittemore reads the gap as evidence that agents let companies scale spending with work and parallel activity, not employee headcount.
• Companies spent about 2.5 times what they budgeted for AI: planned spending growth was 1.9x, but actual average spending rose 4.7x. Whittemore says seat-based budgets failed to anticipate token-heavy agent use, leaving CFOs to treat AI as a new spending category.
• AI-native vendors dominated coding, finance and operations, and design spending, while incumbents retained most other enterprise categories. IT was the clearest incumbent stronghold: established vendors captured 91% of IT AI-app spending, compared with 9% for AI-native companies.
• Menlo found SaaS application growth slowed from 13% in 2025 to 11% in 2026, but most new AI spending came from new budgets rather than existing software budgets. Whittemore warns that agents using software without conventional interfaces could still pressure SaaS.

Models (3)

• Enterprise buyers shifted spending toward more capable, more expensive frontier models: their share rose from 35% to 63%, while mid-tier, small, and older models lost share. Menlo also reported labs cutting small-model prices by more than 80%, showing that cost competition and demand for capability coexist.
• Anthropic gained enterprise API share from 12% in 2023 to 55% this year, as OpenAI recovered to 31% and Google fell to 9%. Whittemore links Anthropic’s lead to Claude Opus models becoming a default for long-running agentic coding tasks.
• OpenAI’s Images 2.6 added short animated GIF generation, frame editing, speed controls, and reference-image input. OpenAI said users create 4 billion images per week, up from 3 billion a couple of months earlier, and the model is 50% faster, generating images in as little as six seconds.

Open Source (1)

• Open models accounted for 4% of enterprise AI spending but 19% of tokens, up from 11% of tokens the prior year. Kimi captured 61% of open-source spending; Whittemore argues token share may better indicate where adoption is heading than revenue share.

Coding (1)

• Revenue for Anthropic’s Claude Code and OpenAI’s Codex rose from $1 billion at the end of 2025 to $24 billion at the end of 2026. Other coding platforms, including Cursor, Cognition, Factory, and Lovable, also grew from $2.6 billion to $7.8 billion.

Agents (3)

• Agentic coding applications grew 7.7x, just behind financial-services AI applications at 8x, though coding spending was far larger. Menlo estimated $34 billion in coding-app spending against $800 million for financial services; customer success, sales, legal, and marketing also expanded.
• Claude’s personal-finance feature can connect to bank accounts through read-only Plaid access and analyze spending for budgeting, but it will not provide investment advice. Anthropic says users should rely on licensed professionals and regulated firms for major financial decisions.
• Meta’s Muse agent added voice and video conversations with animated avatars, selectable voices, and subtle expressions, using Meta’s live-voice models and real-time avatar technology. The feature lacks visual input for reading nonverbal cues and is available on mobile, web, and Mac.

Markets (2)

• AI stocks suffered their worst drawdown since 2020, with the Magnificent Seven losing a combined $1 trillion in market value. The S&P 500 fell 4.4% and the Nasdaq 6.4%, while semiconductor suppliers Micron and SanDisk dropped more sharply than major platform companies.
• Whittemore says the broad selloff looked like forced deleveraging after semiconductor losses, rather than a sudden change in company fundamentals. Reports said Walleye returned capital, Lone Pine had collapsed, and Millennium withdrew $1.5 billion in leveraged capital from two outside managers.

Models (4)

• Sundar Pichai said AI is already increasing Alphabet’s training and launch velocity, and expects that acceleration to continue over the next 12 months as models help improve AI development.
• An MIT study found GPT 6 Astra could improve Qwen 3.5 0.8B by editing its weights directly, without gradients or training. Astra identified reasoning loops and removed them, sharply raising math and instruction-following scores.
• The MIT researchers also reported that frontier models could infer a network’s purpose from its weights and write small game-playing networks without environment feedback. Alex Krensel replicated the results but stressed that edited models remain tiny compared with frontier systems.
• Wei Aizhou argued that the sixfold math gain came from fixing a looping bug, not broad mathematical improvement. The host says the result still suggests a possible mechanism for recursive self-improvement, while runtime self-editing remains beyond current models.

Agents (6)

• Elon Musk said GrokBot’s serving costs were unsustainable relative to revenue after users reported new usage limits. The host says the post could reflect product economics or market maneuvering, but the evidence does not establish which.
• GrokBot users objected that limits had changed without warning, while Musk’s earlier claims of rapid growth and revenue-generating demand clashed with his cost warning. Dots, Duet, and Hermes promoted themselves as alternatives for users considering a switch.
• Dax from OpenCode argues personal agents may be hard to monetize directly: consumers may not pay enough, leaving agents as top-of-funnel products whose business model remains unclear. The subsidy debate predates GrokBot and resurfaced during the 2025 AI bubble debate.
• Olivia Moore argues Meta Muse’s strongest opportunity may be small businesses: Meta can distribute agents through its apps, connect business tools, and turn existing advertising relationships into a broader role in SMB operations.
• SemiAnalysis’s GDPX benchmark ranks Opus 5.5 highest on economically valuable tasks, but Gemini 4 Carbon costs far less per task for a small performance gap. The results make model efficiency central to whether personal agents can become affordable.
• OpenAI’s LifeVal measures agents on everyday tasks such as finding childcare, budgeting, and ordering groceries. OpenAI reports improving success from GPT 5.2 to GPT 6 Astra and is targeting practical tasks relevant to free users.

Markets (2)

• Hedgy Markets says AI plans priced below token costs depend on investor subsidies; one analysis estimated a month of ChatGPT Pro use would cost about $14,000 at regular API prices. Nick Huber argued that subsidy withdrawal could raise prices, reduce usage, and crush valuations.
• The host expects API pricing to sort users into those satisfied by cheaper, less capable plans and power users willing to pay for valuable work. The host says API spending can create net-new value when it avoids human labor’s cost and management time.

Enterprise (1)

• The host argues enterprise adoption depends on more than model capability: companies still use a small fraction of available AI because corporate systems move slowly and adoption is difficult. That gap supports both a bearish case about weak demand and a bullish case about future growth.

Safety (1)

• The host says AI debate remains focused on whether RSI has arrived, while safety advocates are the main group describing what an RSI world might mean and some argue it should be illegal. The host calls for more attention to consequences than to defining the acronym.

Elections (1)

• The host says the midterms put AI infrastructure into election debates at an unusual scale. Goldman Sachs estimates restrictions could reach 45% of planned US data-center construction, while candidates backing pauses won or led in 12 gubernatorial races.

AI Infrastructure (3)

• The host argues that data-center resistance is largely a state and local issue, not a uniform rejection of AI. Alex says communities should bargain for concrete benefits such as school lunches, medical coverage, or tax relief rather than ban projects outright.
• The host says political uncertainty could shift data-center investment toward states that welcome it, including Tennessee and Virginia, and toward Canada’s abundant power. Alex expects orbital compute to gain momentum as terrestrial projects face election-driven permitting risk.
• The host says major buyers have reserved much of the country’s compute and chip capacity, leaving smaller firms unable to obtain cloud access. Speakers warn that compute access could become a basic requirement for business, while Alex says scarcity should raise prices and create incentives to build elsewhere.

Models (4)

• Alex says frontier labs need consumer-facing benefits that make AI tangible, proposing that they use superintelligence to tackle the top 5,000 diseases. The host adds that a personal health tool or widely loved AI product could build public support faster than abstract benchmark gains.
• Alex expects frontier-model improvement to move from scheduled releases toward continuous updates around early 2027. He predicts AI systems will explain new architectures, while the host argues rapid releases will force companies to rethink how people absorb new capabilities.
• The speakers expect recursive self-improvement to redirect compute toward science, engineering, medicine, and products such as drugs. The host predicts many people’s first vivid AI encounter could be a cheap, personalized talking avatar; Alex also predicts AI could enable relativistic probes and force renewed attention to the Fermi paradox.
• A paper measuring AI economic value per joule tested professional tasks across occupations: GLM 5.2 led in average task value, while Gemma 4’s 31-billion-parameter version delivered more value per joule. Alex favors measuring economic output against economic input, since watts may not remain the right denominator.

Labor (1)

• Alex frames AI’s social bargain as a trade between capital and labor: superintelligence could bring disease treatment, universal basic benefits, and broader human uplift. He warns municipalities that banning data centers may forfeit leverage before compute shifts to space.

Markets (1)

• The host says markets priced in slower power generation and delayed data-center approvals, while approved construction is less likely to stop because of sunk costs and jobs. The selloff also reflected fears that China could gain ground as US infrastructure slows.

Energy (1)

• The host says power constraints were already pushing developers toward efficiency, and political backlash adds urgency. The host also urges European energy executives to build generation, arguing that Europe’s focus on gas supply and local politics obscures rising AI power demand.

Reasoning (1)

• Alex defines recursive self-improvement partly by whether new compute is assigned to AI researchers rather than human researchers; he says the balance may already have crossed 50%. The speakers cite AI-generated research ideas, large agent runs, and scientific problem-solving as evidence that AI can improve its own research process.

Regulation (1)

• The White House’s NSTM-6 policy requires federally funded researchers to use approved DNA and RNA synthesis providers, with customer verification and record retention. Alex favors controls on physical synthesis over government rules on what people or AI systems may think.

Robotics (1)

• Alex expects similar physical-world rules for general-purpose robots, potentially requiring approved manufacturers and logs of commands. He argues that regulation should focus on actions and safety controls, rather than restricting access to compute or AI research.

Space (1)

• A New Zealand photographer combined long exposures to show thousands of satellite trails, which Alex calls evidence of an emerging “baby Dyson swarm.” He says orbital compute could reduce reliance on politically exposed terrestrial data centers, with sun-synchronous orbit offering room before expansion elsewhere in the solar system.

Politics (1)

• Alex sees political swings as cyclical and superintelligence as a continuing secular trend; he cites Ray Kurzweil’s law of accelerating returns to argue that wars and depressions have had little effect on compute’s long-term price. The host warns that institutions must adapt actively rather than let technological change unfold unchecked.

Enterprise (9)

• Ara Kurazzian says Ramp's payment data covers more than 80,000 businesses and over $275 billion in annual business spend, but skews toward tech-forward companies likely to adopt AI early.
• Ara Kurazzian says AI has been Ramp's fastest-growing tracked spending category over the past two years. Anthropic has become the most popular AI model among businesses, while competition now includes OpenAI, xAI, and Meta.
• Ara Kurazzian says AI spending is highly concentrated: the top 1% of businesses account for 80% of it. Ramp's median AI spender pays about $15 per employee monthly, up from roughly $12 a month earlier.
• Ara Kurazzian says the top 10% of firms still increase AI spend, while the top 1% have slowed from earlier hypergrowth. Top 1% firms spend roughly $7,000 to $10,000 per employee monthly, even as token use continues to rise.
• Ara Kurazzian attributes slower spending growth partly to falling prices and better efficiency: AI prices fell nearly half over the observed period, allowing firms to use more tokens without equivalent spending growth.
• Ara Kurazzian says advanced adopters are starting to control costs with usage limits on frontier models, better defaults, and model routers. Even profitable Ramp is a high AI spender, but those controls can reduce bills without reducing usage.
• Brett sees a private-equity opportunity in bringing AI adoption at ordinary firms closer to the top 10% of adopters. Firms that deploy AI effectively could outperform less capable competitors in the same industry.
• Ara Kurazzian expects AI use to become table stakes for the median firm, while arguing that meaningful returns may not require top-tier spending. He says definitions of AI spend will shift as AI features become embedded in ordinary software.
• Asked for a five-year spending estimate, Ara Kurazzian said median-firm AI spend would probably remain below $1,000 per employee monthly, then questioned whether that figure should be annual instead and declined to offer a firm forecast.

Models (2)

• Ara Kurazzian says model competition can shift enterprise adoption quickly: Anthropic rose to leading business usage within about a year. He links enterprise demand partly to consumer visibility, coding tools, and product launches such as Grokbot.
• Brett argues that demanding business tasks still require substantial human guidance and stronger, more expensive models; he expects firms to spend on frontier models for exploration and competitive advantage. Ara Kurazzian counters that non-tech businesses can get useful returns without spending heavily.

VC (1)

• Ara Kurazzian says Ramp's highest-spending AI businesses are largely venture-backed technology companies. That concentration also creates correlated risk: if the AI trade or fundraising slows, those firms may cut spending together and affect the wider sector.

Coding (1)

• Ara Kurazzian calls coding agents AI's most commercially advanced use and says they improve productivity for software engineers, especially in tech. He doubts the median business can extract comparable value when it has few or no developers.

Safety (1)

• Brett argues that AI-enabled attacks will push cybersecurity spending across industries, since each company has its own technology stack to protect. He specifically raises open-weight models as a potential source of attack capability.