Price:

AI & TECH

China's Kimmy K3 undercuts US AI at one-fiftieth cost

Thursday, July 23, 2026 ·

Summary

  • Chinese model Kimmy K3 beats US counterparts on performance and costs 50 times less.
  • US chip bans backfired, pushing China to innovate around compute limits.
  • American firms now route Chinese AI into US products despite policy bans.

China didn't close the AI gap - it vaulted over it. Moonshot AI’s Kimmy K3, a 2.8 trillion parameter model, now leads on coding and data analytics benchmarks. It costs about a dollar per million tokens. US models like Opus 4.7 charge fifty times that.

The shift is economic as much as technical. On July 19, Emad Mostaque confirmed Kimmy K3 was trained on H-800s and domestic Huawei chips, bypassing US export controls. This isn’t brute force. It’s surgical engineering - mixture-of-experts, linearized attention, and lean data pipelines. The model will be open-weight by late July, accessible to any developer with a local server.

US labs look increasingly bloated in comparison. Dario Amodei, speaking at the G7 summit in France, pushed for international cooperation on AI risks. But the US denied UK Prime Minister Keir Starmer a carve-out from export bans the same week. Allies noticed. Emmanuel Macron called it a 'kill switch' - access to American intelligence now depends on Washington’s goodwill.

The irony deepens inside Microsoft. According to The AI Daily Brief, the company is fine-tuning DeepSeek V4 to power parts of Copilot. A US firm is integrating a Chinese model into its flagship productivity suite while the US government blocks access to that same model abroad. The economic logic is undeniable: a task costing 49 cents on Opus drops to 6 cents on DeepSeek.

"They use the West's existence proofs to cut their own R&D costs by 90%."

- Emad Mostaque, Moonshots with Peter Diamandis

Mostaque’s point cuts deep. American labs spend billions chasing marginal gains. China replicates the results at pennies by focusing on efficiency - a direct result of US chip bans forcing innovation. Nathaniel Whittemore noted this pivot: enterprises now use 'model councils,' pairing cheap Chinese workers with expensive US advisors. The hybrid outperforms either alone.

Regulation is becoming a self-inflicted wound. Conner Brown on TFTC warned that Anthropic and EA-aligned lobbyists are pushing KYC on compute and licensing for frontier models. They cite 'existential risk' - the same playbook used to strangle nuclear innovation. If open-source AI is banned, only governments and giants will control intelligence.

China isn’t waiting. It formed a UN-backed AI coalition with 29 countries, including Russia and Brazil. While the US debates kill switches, Beijing builds influence through economic gravity. The frontier is no longer a race. It’s a reversal.

Source Intelligence

- Deep dive into what was said in the episodes

#773: Open Source AI Is Non-Negotiable with Conner BrownJul 22

Also from this episode: (19)

Other (19)

  • Marty Bent argues Bitcoin will prevail as central banks globally continue to devalue their national currencies, reinforcing Bitcoin's role as a safe haven asset.
  • Marty Bent believes Washington D.C. will become a crucial city for technological and policy developments over the next five to ten years, surpassing current hubs like Austin and Miami.
  • Conner Brown notes that BPI's Q2 report highlights significant accomplishments in three months, demonstrating how AI tools amplify human productivity and enable deeper research and new international work.
  • Conner Brown explains BPI developed a research paper on Taiwan diversifying its cash reserves into Bitcoin, noting Taiwan's history of using gold to fund an exiled government after fleeing mainland China.
  • BPI's paper gained traction in Taiwan, leading to an invitation to brief the Legislative Yuan and Central Bank, who showed surprising understanding and detailed questions regarding Bitcoin's practicalities.
  • Conner Brown states the US administration's Executive Order for the Strategic Bitcoin Reserve (SBR) is actively influencing global policy, though its implementation faces bureaucratic hurdles, including a past scandal involving mishandled funds.
  • Conner Brown indicates the UK aims to be a 'fast follow' to US digital asset policy, having recently rescinded a stablecoin balance cap and showing unexpected openness to Bitcoin policy discussions in the House of Lords.
  • BPI research, spearheaded by Sam Lyman, uncovered Chinese foreign influence, specifically from the Party for Socialism and Liberation, aggressively opposing US data center and Bitcoin mining projects with unfounded claims.
  • Conner Brown warns that promoting pessimistic AI narratives while simultaneously building the technology, as some big tech executives do, risks repeating past mistakes like the nuclear energy industry's over-regulation.
  • Marty Bent notes that Bitcoiners are early adopters of AI tools, developing complex workflows, unlike traditional think tanks or legacy institutions that demonstrate significant institutional inertia in AI integration.
  • Conner Brown observes a coalition of diverse groups, including foreign adversaries, radical environmentalists, populists, national security hawks, and effective altruists, are strangely aligned in advocating for AI restrictions.
  • Marty Bent describes TFTC's three-layered 'company brain' for AI agents, integrating flat markdown files, a QMD semantic search database, and an Obsidian vault (Cogni) to provide comprehensive, temporal context from all past content.
  • Marty Bent leverages voice dictation via Super Whisper to interact with his AI agent 'Martin,' enabling him to articulate thoughts faster and more comprehensively for tasks like newsletter writing, similar to Napoleon's multi-scribe dictation.
  • Conner Brown envisions an optimistic AI future where free and open-source models proliferate, amplifying human agency, creativity, and education to unlock unprecedented individual capabilities, contrasting with a dystopian consumption-focused path.
  • Conner Brown emphasizes the critical need for American leadership in open-source AI to prevent strategic adversaries like China from dominating the field and embedding biases that could serve their interests.
  • Conner Brown is increasingly convinced that an abundant future enabled by AI will strengthen the need for Bitcoin, as open-source AI agents, if trained neutrally, naturally prefer Bitcoin and stablecoins for digital commerce.
  • Aven offers a Bitcoin Visa card that provides a line of credit up to $1 million, backed by Bitcoin custodied by BitGo, featuring fixed rates starting at 7.99% APR and 2% unlimited cash back.
  • Unchained provides collaborative multisig Bitcoin custody, securing over $12 billion in Bitcoin for more than 12,000 clients, where users hold two keys and Unchained holds one, ensuring institutional-grade security and user control.
  • CrowdHealth offers crowdfunded healthcare as an alternative to health insurance, significantly reducing monthly costs and negotiating prices down by 50-80%, with members paying a small amount ($500-$3000) for health events.

7/21/26: Trump Threatens Iran Payback, Trump Rages Over Bibi Arrest Push, China Overtakes US On AI, Biological War PlotJul 21

Also from this episode: (55)

Other (55)

  • Saagar reports President Trump has issued a directive to military leaders, stating Iran will pay "many times over" for every American soldier killed, signaling continued escalation.
  • Saagar notes multiple vessels in the Strait of Hormuz, including a Saudi ship, were attacked or abandoned recently, escalating pressure on global oil markets.
  • Saagar indicates President Trump faces a critical decision: pursue a ten-day ceasefire to reopen the Strait of Hormuz, or launch a "massive joint military campaign" with Israel.
  • Krystal views Trita Parsi's pessimism about diplomacy with Iran as a significant warning, especially since he typically advocates for negotiations.
  • Krystal cites US intelligence, which assesses that new military strikes against Iran are unlikely to significantly impact its negotiating position or weaken it.
  • Saagar refers to Professor Pape's work, stating that air campaigns alone historically fail to cause regime change or strengthen negotiating positions against adversaries like Iran.
  • Saagar highlights Israeli intelligence claims that Iran moved thousands of uranium enrichment centrifuges into "Pickaxe Mountain" last fall, with President Trump threatening to attack the site.
  • Saagar suggests neoconservative radio host Hugh Hewitt may have influenced President Trump's focus on attacking "Pickaxe Mountain" during a presidential interview.
  • Saagar reports Professor Morandi advised Gulf nations like Kuwait, Bahrain, Qatar, Saudi Arabia, and Oman to evacuate due to impending strikes on critical infrastructure if US escalation continues.
  • Saagar identifies Private Isabella Gonzales (19) and First Lieutenant Tyler James Fien (killed July 18, 2026) as US service members killed, alongside two Iranian students, highlighting human costs.
  • Saagar highlights satellite imagery showing a direct strike on US barracks in Jordan, where soldiers were killed, indicating Iran's ability to evade US defenses with three hits in one day.
  • Krystal expresses concern that conventional weapons may be insufficient to destroy Iran's hardened "Pickaxe Mountain," fearing President Trump might consider using nuclear weapons for a "win."
  • Krystal cites "Posn," who argues the US lacks military means to defeat Iran or control Hormuz, suggesting the US empire is no longer viable due to modern precision strike regimes.
  • Krystal draws parallels between the US situation with Iran and Israel's struggle against Hamas or Russia's bogged-down invasion of Ukraine, attributing these outcomes to modern warfare.
  • Saagar references a reported CIA estimate that a Russian soldier's average lifespan on the front is "thirty seconds," illustrating the extreme human cost of the conflict.
  • Krystal asserts Iran is a more effective industrial power than the US, demonstrating superior manufacturing of drones and a better understanding of its strengths and US weaknesses.
  • Saagar reports Israeli Finance Minister Smotrich stated Israel has no interest in joining the current US-Iran confrontation, deeming the situation "the best one for us."
  • Krystal mentions Israel rationed missile interceptors during a recent 40-day war, indicating stretched defense capabilities and potentially influencing its decision to avoid current US-Iran escalation.
  • New York Mayor Zoran (presumably Zorn) indicated he is consulting his legal department regarding the possibility of arresting Benjamin Netanyahu, citing an ICC arrest warrant for alleged war crimes.
  • Krystal views Mayor Zoran's comments on arresting Netanyahu as a partial walk-back, noting the US is not a signatory to the International Criminal Court, making such an arrest legally difficult.
  • Trump publicly stated Benjamin Netanyahu "will not be arrested" in the US, blaming Iran for regional conflict, ensuring Netanyahu's safety during his September visit.
  • Mayor Zoran declined to meet with the Israeli consulate, asserting that an ICC arrest warrant for Netanyahu regarding alleged war crimes and genocide against Palestinians should be honored.
  • Krystal observes a radical shift in sentiment, noting an anti-Zionist, BDS supporter like Mayor Zoran could be New York City mayor, partly due to Jewish New Yorker support.
  • Saagar clarifies a local mayor lacks statutory authority to arrest a foreign dignitary on US soil, as federal law and the US State Department hold exclusive jurisdiction.
  • Saagar reports Benjamin Netanyahu struggled to secure a meeting with President Trump, despite multiple previous Oval Office visits, with sources suggesting Trump's team (Vance's team) might be preventing it.
  • Krystal suggests Trump's reluctance to meet Netanyahu could stem from personal irritation, potentially due to Netanyahu congratulating Joe Biden on his election victory.
  • Axios reports China "just erased America's AI lead," with Moonshot AI's Kimmy K3 model threatening the foundations of America's AI boom.
  • Saagar reports the AI Evaluator Arena initially ranked Kimmy K3 as the number one model with 1,679 points, surpassing Claude Fable Five.
  • Saagar details Kimmy K3 ranked first in six of seven domains, including brand/marketing and data/analytics, placing second only in gaming behind Fable Five.
  • Moonshot AI temporarily paused new Kimmy K3 subscriptions due to overwhelming demand, prioritizing existing users and planning two focused membership plans to manage compute.
  • Saagar highlights a perceived difference where Chinese AI companies like Moonshot AI prioritize user experience by managing capacity, unlike some US counterparts who keep taking money.
  • Saagar explains an internal White House debate over banning Chinese AI models, with OpenAI and Anthropic advocating for bans to protect their market dominance against cheaper alternatives.
  • Saagar references a tweet criticizing "cyberguardrails" in US models like Codex and Fable for hindering functionality, while Kimmy K3 "just gets the job done" without such restrictions.
  • Chamath Palihapitiya states leading AI models cost $26-56/million tokens for US closed-source, versus $0.50-$1/million tokens for open-weight Chinese models.
  • Chamath Palihapitiya argues that forcing American companies to pay significantly more for AI will make them financially impaired and lead to market collapse.
  • Krystal notes US policy banning top-tier chips for China inadvertently drove Chinese AI companies to develop more efficient, open-source models, now undercutting US closed labs.
  • Krystal suggests that if US AI frontier models are not profitable, the larger AI ecosystem (data centers, applications) could still thrive, with models themselves becoming a utility.
  • Saagar asserts the US economy excels at chip design but outsourced manufacturing, leading to expensive American AI models struggling against subsidized, efficient Chinese alternatives.
  • Saagar describes China's strategy of subsidizing and incentivizing its AI ecosystem, similar to its approach with electric vehicles, creating intense internal competition before global expansion.
  • Saagar mentions current private market valuations for frontier AI models are around $1 trillion each, indicating a potential $2 trillion total value at risk if devalued.
  • Krystal draws a parallel to the US auto industry, where protection from Chinese EV competition may offer a bailout but risks broader failure as the world adopts advanced Chinese EVs.
  • Annie Jacobson's new book, "Biological War: A Scenario," warns of a potential biological apocalypse, deemed an "extraordinary threat" by her counterproliferation sources.
  • Annie Jacobson states biological weapons pose the "greatest overall threat," contrasting with nuclear weapons as "most destructive," due to their low barrier to entry.
  • Annie Jacobson explains that, with AI, biological weapons can be created in a basement by individuals with minimal biological knowledge, unlike heavily controlled nuclear weapons.
  • Annie Jacobson highlights 3,500 BSL-3 labs (90% lacking safety) and 110 BSL-4 labs globally (17 in US), posing significant lab leak risks.
  • Annie Jacobson discusses "gain of function" research, where pathogens are modified to increase virulence or transmissibility, which some experts like Andy Weber call "biological weapons engineering."
  • Annie Jacobson identifies two COVID-19 lessons: US "woefully unprepared" despite billions spent, and public trust in science/health officials fractured significantly.
  • Annie Jacobson notes COVID-19 had a 1% case fatality rate, but pathogens with 100% rates exist, underscoring the urgency of improved preparedness.
  • Krystal expresses concern that Trump administration cuts, including a 25% CDC workforce reduction and WHO withdrawal, further degrade US readiness for biological threats.
  • Annie Jacobson reveals a classified US government "devolution" program to transfer power to a hidden, pre-selected leadership group during a "near extinction-level event" or anarchy.
  • Annie Jacobson explains the devolution program is triggered by "COGCON One" (Continuity of Government Readiness Condition One) in scenarios like a biological attack or lab leak causing widespread anarchy.
  • Annie Jacobson reports DARPA's Biology, Science and Technology Study group discussed the combined danger of AI and biology, and the "fascinating concept" of pulling the plug on both.
  • Annie Jacobson notes six private AI CEOs now hold "more power" than government organizations like DARPA, posing a significant challenge to national security.
  • Annie Jacobson criticizes the WHO and CDC for repeatedly tweeting COVID-19 was "not airborne" for three months, despite millions dying, demonstrating a failure to correct misinformation.
  • Annie Jacobson states Pentagon wargames indicate nuclear war, regardless of its initiation, inevitably leads to "total apocalyptic annihilation."

Urgent Update- AI Sputnik Moment: Kimi K3 Released w/ Emad Mostaque | Ep. 272Jul 19

  • China has prioritized humanoid robot development, with 150 companies, and stages public robot combat events that test demanding engineering aspects like balance, impact resistance, and locomotion.
Also from this episode: (15)

Models (11)

  • Moonshot AI, a Chinese lab, released Kimi K3, a multimodal, open-weight model with 2.8 trillion parameters, which rapidly achieved number one ranking in the Code Arena and six other domains.
  • Alex Weiser Gross notes Kimi models held state-of-the-art among open-weight models for 9 of the past 12 months, despite building on a recognizable transformer architecture without novel breakthroughs.
  • Imad Mostaque compares Kimi K3's engineering to Chinese EVs, achieving high performance and usability through manufacturing efficiency, despite using older hardware like H-800s under US export controls.
  • Dave Blundon describes Kimi K3's release as a "Sputnik times infinity" moment, enabling any entity to achieve near-frontier AI capabilities by using open-source weights and fine-tuning, independent of US models.
  • Salim Ismail argues frontier intelligence is a perishable asset with a shelf life of weeks, suggesting future value will reside in architectures that can rapidly swap and adapt different AI models.
  • Imad Mostaque reports Kimi K3 used the same compute as Inkling but achieved 2.5 times better data-to-intelligence conversion, optimized for Chinese silicon like Huawei 910 Ascend chips.
  • Since mid-April, 13 new frontier models have launched, averaging one every 10 days, a significant acceleration from previous years, with Alex Weiser Gross predicting daily releases by January.
  • PrismML's Bonside 27B is the first 27 billion parameter class model to run entirely on a smartphone, achieving ternary quantization that significantly reduces model size and increases speed fivefold.
  • Alex Weiser Gross predicts sub-1-bit quantization will go mainstream within the next year, and Imad Mostaque forecasts Kimi K3-level capability on a 16GB RAM MacBook by late next year due to distillation and new chipsets.
  • AI models are now statistically indistinguishable from human "superforecasters" in predicting novel events, implying cheap, tireless superhuman advisors for decisions in insurance, investing, and policy.
  • Salim Ismail argues that AI forecasting will make most senior management expertise obsolete, as AI systems can reproduce judgment without human biases, shifting focus to purpose and objectives.

Startups (1)

  • Alex Weiser Gross clarifies that Moonshot AI founder Yang Jilin started his Chinese startup, Recurrent AI, during his CMU PhD program, not after graduation due to visa issues.

Immigration (1)

  • Salim Ismail notes 70% of elite AI researchers are not U.S. citizens, with Chinese and Indian talent being prominent; he argues that the U.S. immigration system fails to retain this crucial talent.

Health (1)

  • Dr. Don Musilam reports that 3.3% of Fountain Life members, presumed healthy, have undetected cancer, emphasizing the critical role of early detection via tools like full body MRI for better cure rates.

AI Infrastructure (1)

  • Salim Ismail highlights that U.S. data centers consume 17 billion gallons of water annually, significantly less than golf courses (531 billion gallons) or California almond farming (1 trillion gallons).

RABBIT HOLE RECAP #418: BRAVE NEW WORLDJul 17

  • China formed a UN-backed AI cooperation organization with 29 countries, including Russia, Belarus, Serbia, Cuba, Brazil, Venezuela, and others across Africa and Asia.
  • Stephen Miller announced a National Security Presidential Memorandum (NSPM 7) directing agencies to identify, debank, arrest, and prosecute left-wing political terrorists, framing the DSA's use of Democratic primaries as a Trojan horse for revolution.
  • The Human Rights Foundation reported Burma's military junta froze over 100 bank accounts of companies and summoned individuals for questioning.
  • Truth Social plans to sell faster API access to Trump's market-moving posts, creating a monetization strategy where paid users can front-run public announcements.
  • Gabriel Perez, Trump's teleprompter operator, settled allegations he used inside knowledge of speech content to win over $100,000 on Kalshi's prediction markets.
  • The Federal Reserve formed task forces including Mark Andreessen and Asha Sharma to assess AI's economic impact, announced the same day Microsoft laid off 10% of its Xbox workforce.
  • The .me registry suspended Telegram's t.me domain because a sanctioned VPN provider had a channel there, restoring it only after Telegram removed the account.
Also from this episode: (9)

Protocol (6)

  • Connor Brown explains the Bitcoin Policy Institute intervened as a defendant in a lawsuit claiming dormant Bitcoin addresses as abandoned property, using BPI's long-term treasury as an example of non-dormant institutional holdings.
  • Sam Lyman's research for BPI on CCP influence in US anti-data center campaigns gained significant attention, being featured on the front page of the New York Times.
  • BPI made its quarterly report public for the first time, highlighting activities from April to June 2026.
  • Anchor Watch launched multi-institution custody with Coin Corner and Lloyd's insurance, offering users a custodial solution where three separate institutions hold keys to mitigate single-point failure.
  • Bitcoin's current price is $63,410 with a $1.27 trillion market cap, and the network is 1,181 blocks from a difficulty retarget estimated at a -0.7% adjustment.
  • Mempool transaction fees are low, with 2 sats/vbyte for high priority and 0.2 sats/vbyte for no priority currently clearing next blocks.

Lightning (2)

  • A Zeus and Axiom report on Lightning economics shows Olympus achieved a 4.65% return on invested capital over the trailing twelve months, comparable to treasury yields.
  • The report also details using Validating Lightning Signer (VLS) to separate node operation from key custody, allowing professional routing management without surrendering fund control.

Stablecoins (1)

  • Nunchuck Android 2.7.1 added self-custodial USDT on Liquid, noted as the most private Tether option due to confidential transactions hiding amounts and asset types.

Is Kimi K3 Really Fable Class?Jul 17

  • Nathaniel Whittemore notes the G7 meeting featured unprecedented AI industry representation with leaders like Sam Altman, Demis Hassabis, Arthur Mensch, and Dario Amodei attending alongside heads of state.
  • Dario Amodei argued for international cooperation including structured access to frontier models, chip trade deals excluding China, and a unified approach to AI risks like cyberattacks and bioterrorism at the G7.
  • European leaders, like Emmanuel Macron, expressed concern that the US holds an AI 'kill switch' and pleaded for shared access to frontier models, fearing reliance on the US.
  • Sam Altman asserted AI regulation must be shaped by democratic institutions and society, not just corporations, and called for an international forum to establish global testing standards and risk analysis.
  • Whittemore reports the US did not concede on Fable access at the G7, and the UK's request for an export control carve-out was denied, souring the EU mood.
  • Noam Shazeer, co-author of the Transformer paper, left Google for OpenAI after Google spent $2.7 billion licensing his Character AI technology to retain him less than two years earlier.
  • OpenAI is sunsetting the Pulse daily briefing feature and expanding scheduled tasks to all paid subscribers, signaling a shift towards prioritizing users who need automated task workflows.
  • The Fable 5 ban has accelerated enterprise interest in open-source models for cost predictability and to avoid government kill switches, with media consensus pointing to open source as the biggest winner.
  • Chinese lab Moonshot AI released Kimi K 2.7 Code, claiming a 22% improvement on its code bench and 30% lower reasoning token usage compared to 2.6, though early users report it underwhelms in practice.
Also from this episode: (6)

AI Infrastructure (3)

  • Europe's AI sovereignty plan commits only 20 billion euros to build five gigafactories deploying around 100,000 GPUs, while US hyperscalers spend three times that monthly on AI data centers.
  • Open Router's Fusion API uses a panel of models routed by a judge and synthesizer to achieve frontier-level performance at half the price, validating a future where tasks are routed to specialized, cheaper models.
  • Harvey's experiment combining an open-weight GLM 5.1 worker with a closed frontier Opus 4.7 advisor increased performance and lowered costs, demonstrating that smart model routing beats using the most expensive model for every task.

Models (1)

  • The open-source model GLM 5.2 from ZAI beats GPT-5.5 and Opus 4.8 on some benchmarks at one-tenth the cost, fueling speculation about its distillation from Anthropic models and its viability as a Fable alternative.

Enterprise (1)

  • Microsoft is reportedly considering a locally hosted fine-tune of DeepSeek V4 for Copilot Co-work to offer cheaper AI access to enterprise customers, potentially normalizing Chinese models in US enterprise stacks.

Coding (1)

  • Cursor's Composer 2.5, built on a Kimi foundation, scores near Opus 4.7 and GPT-5.5 on benchmarks at a fraction of the cost, but user reports and updated agent-focused benchmarks show mixed real-world performance.