Price:

BUSINESS

Analysts warn AI circular economy risks systemic shock

Friday, July 10, 2026 · from 3 podcasts
All-InThis Week in StartupsBreaking Points
  • Hyperscaler capex and private credit fuel an AI circular economy that could break.
  • Frontier labs burn billions, their failure could trigger a wider credit collapse.
  • Treasury draft reports systemic risk deeper than the dot-com bubble.

Capital is pooling into a shortlist of six to twenty private AI companies, including OpenAI, SpaceX, and Anthropic. Secondary markets operate without disclosure, creating dangerous opacity. Michael Kim of Sindana Capital notes fund managers use Special Purpose Vehicles as bait to entice LPs into larger fund commitments. The market is defined by desperation and rumor.

'Capital is pooling into a hit list of just six to twenty companies.'

- Nikil Basu Trivedi, This Week in Startups

That private market frenzy is mirrored by public market fragility. Nikil Basu Trivedi expresses more concern over the massive burn rates at labs like OpenAI than the revenue predictability of Nvidia. The risk isn't growth; it's capital consumption.

On Breaking Points, Saagar Enjeti highlights a leaked Treasury Department draft report warning the AI sector is more entrenched in the economy than dot-com firms were. Any failure to hit productivity targets or finish data center build-outs could ripple through utilities, chip makers, and the broader credit system. Officials publicly dismissed the findings.

Kim warns the entire chain relies on hyperscaler capex, private credit for data centers, and retail demand for leveraged ETFs. The music stops if one link fails. Andrew Feldman of Cerebras, speaking on the All-In podcast, quantified the physical scale: data centers the size of football fields drawing more power than mid-sized cities, backed by a $25 billion backlog. This isn't speculative; demand exists.

'The market sentiment is so fragile that even record earnings from players like Samsung can result in a stock price drop if they don't exceed astronomical expectations.'

- Michael Kim, This Week in Startups

The story hasn't progressed beyond these warnings in three days. The risk remains theoretical but systemic. Analysts see the circular economy, but CEOs and hardware builders see relentless demand. The disconnect is the story.

#AI Infrastructure#Macro#Regulation
OpenAIAnthropicNvidia

Source Intelligence

- Deep dive into what was said in the episodes

All-In with Chamath, Jason, Sacks & Friedberg
All-In with Chamath, Jason, Sacks & Friedberg

Open Source Wins, AGI Is Here, and Scorsese's AI Toolkit with CEOs of Cerebras & Black Forest Labs • Jul 10

  • Cerebras’ wafer-scale chip architecture broke Moore's Law and Feldman expects performance to double within 18 months, far exceeding traditional GPU trajectories.
  • Andrew Feldman argues tech giants build their own AI chips not to compete but to avoid dependency, echoing hyperscalers’ previous experience with Intel.
  • Andrew Feldman sees a bifurcation in AI use: frontier models for hard problems and open-source models for routine tasks like data manipulation.
  • Andrew Feldman says sovereignty is a growing trend, with regulated industries opting for domestic open-source models to ensure data control.
  • Andrew Feldman considers government red-teaming of powerful AI models before release reasonable, analogous to pharmaceutical trials.
  • Andrew Feldman argues AGI has already arrived by any definition from 20 years ago, citing Turing Test and intent understanding milestones.
  • Robin Rombach explains latent diffusion algorithms compress natural data into efficient representations, forming the basis for current generative models.
  • Robin Rombach sees multi-modal models trained on images, video, and audio evolving into action prediction systems deployable on robots.
  • Robin Rombach says AI is a medium for exploration, citing Martin Scorsese using Black Forest Labs’ tools to visualize scenes from his imagination.
  • Robin Rombach sees generative AI used for production, citing a $30 million Bitcoin film where AI replaced green screen set construction.
Also from this episode: (7)

AI Infrastructure (3)

  • Andrew Feldman argues the AI infrastructure buildout dwarfs any previous technological mobilization, with nations globally racing to construct data centers that consume more power than entire cities.
  • Cerebras has a $25 billion backlog from hyperscale customers like OpenAI, Anthropic, Google, and Microsoft who are ordering capacity to meet demand that already exists.
  • Andrew Feldman likens early AI adoption to unmanaged AWS credit card spending, where enterprises now shift from unlimited token access to strategic cost management.

Models (2)

  • Recent AI models now understand intent and can critique user goals, moving beyond simple prompt execution to prompting users for clarity on requirements.
  • Andrew Feldman says unlimited compute tokens enable unlimited reasoning, citing his own experiment where an agent debated its own strategy for trend discovery.

Enterprise (2)

  • Robin Rombach says Black Forest Labs blocks generation of certain IP in public tools and works directly with IP holders to develop custom models.
  • Robin Rombach explains Black Forest Labs hires researchers for model training and engineers for custom AI solutions, with over 100 employees now.
#AI Infrastructure#Reasoning#Robotics#Chips#Models
This Week in Startups
This Week in Startups

Why Data Is the Next $1 Trillion Market • Jul 8

  • GPT0’s acquisition by Superhuman was driven by distribution access to tens of millions of users; GPT0 was lifetime profitable with more cash on its balance sheet than it ever raised.
  • Kim observes M&A is rising as overvalued private companies use their stock as currency for acquisitions, similar to Cisco in the 90s, while legacy SaaS firms trade at low multiples due to AI fear.
  • Michael Kim argues the ‘SaaS apocalypse’ is overblown; companies like Intercom with proprietary data, workflows, and customer relationships can leverage AI to create more economic value.
  • Nikquille’s bigger worry is companies with massive burn requiring endless capital raises, like OpenAI needing ‘another several hundred billion,’ rather than top-line misses at firms like NVIDIA.
Also from this episode: (11)

VC (4)

  • Michael Kim describes SPV stacking as a common tactic emerging managers use to entice LPs: they offer access to a coveted ‘hit list’ company through a special purpose vehicle to secure commitments for their main fund.
  • Nikquille Basu Trevetti says venture’s power-law obsession concentrates LP demand on a narrow ‘hit list’ of 6-20 companies, creating intense pressure for creative secondary access deals.
  • Michael Kim notes early-stage investors in hot companies like Cursor or 11 Labs are encouraged to sell secondaries, but late-stage lead investors face negative signal risk if they sell.
  • Kim reports fraud in SPV secondary markets is headline-grabbing but not prevalent in his network; his joint venture secondaries fund with Klein Hill rarely sees it.

Startups (2)

  • Nikquille says his early-stage fund’s strategy is to back companies that later become power-law winners; meaningful exits ($500M generating $75M for a $225M fund) matter more than for billion-dollar funds.
  • Michael Kim highlights Josh Browder’s ability to identify genuine young founders through the teal fellowship pipeline, contrasting them with ‘tourist founders’ seeking credentials rather than building enduring businesses.

AI Infrastructure (3)

  • Nikquille highlights that data is having a cyclical ‘moment’ in the AI stack, but unlike energy (NVIDIA) or compute, there are few huge public companies purely focused on data, creating more white space.
  • Kim warns the market’s unrealistic growth expectations pose risk: semiconductor stocks like Samsung dropped 7-8% after reporting a $58 billion EBITDA jump, showing sentiment demands constant outperformance.
  • Windborne collects proprietary atmospheric data via its own weather balloons, builds AI forecasting models, and sells to governments and companies; cuts to NOAA and global agencies created commercial demand.

Enterprise (1)

  • His portfolio company Protege facilitates data licensing deals between providers and model companies; in its second year, it generates hundreds of millions in revenue.

Models (1)

  • Nikquille says startups can manage inference costs by mixing frontier models for coding with cheaper open-source or older closed models for other tasks, mitigating the risk of Chinese open model export bans.
#VC#Macro#Markets#AI Infrastructure
Breaking Points with Krystal and Saagar
Breaking Points with Krystal and Saagar

7/7/26: AI CEOs Panic Over Job Losses, WH Press Sec Attacks GenZ As Lazy, McConnell Still Missing After Heart Attack • Jul 7

  • An internal Treasury Department draft report, whose existence has not been previously reported, warns that AI companies are more entrenched in the US economy than dot-com firms were and would send shockwaves through the ecosystem if they falter.
  • Saagar cites historical parallels where internal warnings were dismissed, comparing the Treasury AI report to the ignored August 2001 bin Laden briefing and the Glass-Steagall repeal.
  • Saagar points to current market euphoria, noting the S&P is up 10% year-to-date, 21% over the last year, and 72% over five years, calling it a 'full boom-bull market.'
  • Krystal notes AI CEOs like Sam Altman and Dario Amodei have shifted their public messaging from predicting mass job displacement to emphasizing AI as a job growth tool.
  • Krystal argues this rhetorical shift stems from fear of political backlash and potential wealth taxes, referencing Alex Karp's anger at 'prophets of doom' who risk triggering such policies.
  • Saagar argues housing affordability is the core issue for Gen Z grievances, citing median home prices nationally at $450k and noting wage-to-price ratios have worsened exponentially.
  • San Francisco exemplifies the AI-driven housing crisis with a median home price of $1.7 million and average apartment rent surpassing $3800 a month, creating a market where even $180k tech salaries are insufficient.
Also from this episode: (4)

Politics (4)

  • Polling from William Lawrence shows 74% unfavorable opinion of data centers versus 19% favorable, and Krystal cites a candidate opposing them leading a Democratic primary in a Michigan swing district.
  • White House Press Secretary Caroline Leavitt called Gen Z lazy with 'silver spoons' and blamed liberal indoctrination, later clarifying her comments targeted those falling for 'communist' promises.
  • Senator Mitch McConnell suffered a heart attack on June 14th, received CPR from paramedics, and has not provided substantive updates on his condition for weeks while his wife, Elaine Chao, traveled to China.
  • Saagar and Krystal criticize the gerontocracy clinging to power, citing McConnell's re-election at 78 despite health incidents and Feinstein's public deterioration, arguing it reflects narcissism over service.
#Big Tech#AI Infrastructure#Markets#Macro#Fed

Related Stories