Big AI is burning cash and political goodwill at the same time: OpenAI is deeply unprofitable, Uber’s AI budget is already gone, and the vendors gaining enterprise share are locking in with the Pentagon and NSA. Public and legal patience for concentrated, surveillance-adjacent AI power is thinning, while alternative hardware, infra, and search ecosystems quietly get more credible.
The real tension is how long capital stays parked in today’s dominant stacks before weak unit economics and legitimacy risk force a rebalance.
Key Events
/Uber exhausted its entire 2026 AI budget in four months as tokenmaxxing costs surged past expectations.
/Anthropic is reportedly valued at about $850B on $30B in revenue, closing a funding round exceeding $30B while finalizing a classified NSA surveillance-tools contract as OpenAI partners with the Pentagon.
/OpenAI reported a Q1 2026 operating margin of -122%, highlighting deep losses at frontier model scale.
/Norway deployed 2 petabytes of Huawei flash storage for large language model training and Huawei announced a Tau Scaling Law targeting 1.4nm-equivalent density without EUV by 2031.
/SpaceX’s Starship V3 flight was mostly successful, Starlink gen2 satellites reached 100MW of solar capacity, and the company filed for an IPO with dual-class founder control.
Report
At current architectures, frontier AI is a cash bonfire: OpenAI posted a Q1 2026 operating margin of -122%. In parallel, Uber has already exhausted its entire 2026 AI budget in four months on tokenmaxxing with little user‑perceived improvement in service quality.
a i unit economics crack
Uber’s COO says AI investments are increasingly hard to justify as tokenmaxxing expenses outstrip benefits, with the company burning through its entire 2026 AI budget in four months.
Both Uber and Microsoft report that current AI deployments are more expensive than employing human workers, directly contradicting earlier cost‑savings narratives.
OpenAI is operating at a -122% margin in Q1 2026, showing that even the leading model vendor is deeply unprofitable at present price points.
On the hardware side, Nvidia holds a large GPU stockpile as DRAM and VRAM prices rise and support for older 10‑series GPUs is dropped from the latest Torch release, adding volatility and obsolescence risk to the cost base.
Commenters explicitly warn that AI technology costs may end up higher than human labor costs, questioning the sustainability of current large‑model usage patterns.
anthropic, openai, and the state
Anthropic is reportedly valued at about $850B on $30B in revenue and is closing a funding round exceeding $30B, putting it on venture‑style multiples at national scale.
Anthropic is said to have surpassed OpenAI in business adoption, becoming the perceived ethical default even as observers doubt the financial viability of such rapid valuation growth.
The company is finalizing a classified NSA contract for surveillance tools and moving closer to supplying technology for U.S. spy agencies, while OpenAI has secured a Pentagon partnership, embedding both vendors inside the security state.
These deepening ties coincide with Trump canceling a planned AI safety executive order after consultations with Elon Musk and Mark Zuckerberg, highlighting how major platform leaders are directly shaping the regulatory baseline.
legitimacy and backlash: pope, palantir, section 230
Pope Leo XIV’s 42,300‑word encyclical 'Magnifica Humanitas' calls for robust AI regulation, warns against control of AI by a powerful few, and attacks the ‘idolatry of profit’ and ‘culture of power’ driving the AI boom.
Commenters say they are surprised to find themselves aligned with the Pope on tech and see him as offering more credible ethical leadership on AI than many politicians, even as others question his expertise and doubt the practical impact of his warnings.
In London, the mayor blocked a £50M contract between the Met Police and Palantir over procurement rule violations, amid public criticism that the company prioritizes profit over public safety and threatens privacy and civil liberties.
Parallel debates around Section 230 in the DOE vs Meta case and on Y Combinator forums question how much liability platforms should bear for harmful content and moderation decisions, with rising demands for accountability over pure corporate profit protection.
hardware and infra forks: huawei, nvidia, spacex, cloudflare
Norway has deployed 2 petabytes of Huawei flash storage for large language model training, meaning Chinese hardware is already underpinning Western AI workloads.
Huawei is promoting a Tau (τ) Scaling Law with a roadmap to reach transistor densities equivalent to 1.4nm processes without EUV lithography by 2031, explicitly aiming to close the gap with TSMC and alter the chip hierarchy.
At the same time, Nvidia’s large GPU stockpile and rising DRAM/VRAM prices are contributing to a more volatile and inflationary AI hardware market.
SpaceX is scaling Starlink’s solar power to 100MW across 7,000 gen2 satellites with a 1,000MW target for gen3, has flown a mostly successful Starship V3, and is exploring space‑based data centers while pursuing an IPO with dual‑class founder control.
Closer to the edge, developers are consolidating onto Cloudflare for DNS, tunnels, Workers, and security because it is cheap and easy, even as they voice concerns about centralization and potential monopolistic behavior by a single infra provider.
google’s search moat frays
Google AI Studio has already been used to create over 250,000 native Android apps without coding, pushing AI deeper into the Android ecosystem.
Critics argue that Google is cannibalizing the open web to power its AI initiatives and describe its relationship with websites as predatory, with AI‑heavy search interfaces that emphasize ads over genuine content.
Google’s AI systems have spread misinformation, including a false claim about a YouTube family, adding to unease about AI‑mediated search. High‑value users report deteriorating search quality and are moving to alternatives like DuckDuckGo and Kagi or viewing services like Ecosia as mere ad frontends, even though Google remains the default for local searches and news.
What This Means
AI is simultaneously consuming large amounts of capital, entangling itself with state security apparatus, and eroding user and public trust, while alternative infra and discovery channels quietly gain traction. The spread between where money is concentrated in today’s AI platforms and where sustainable economics and legitimacy may sit in a few years is widening fast.
On Watch
/Developer reliance on Cloudflare for DNS, tunnels, Workers, and security continues to grow alongside explicit worries about centralization and monopolistic behavior, setting up a potential edge-infra chokepoint story.
/Autonomous ride-hailing via Waymo is delighting some users, including blind riders, but persistent skepticism about safety in emergencies and edge cases could slow broader adoption and keep gig-work disruption uneven.
/Microsoft canceling a 244-acre data center and cutting about 4% of its workforce in response to AI spending adjustments may signal an early pullback in hyperscale AI capex growth curves.
Interesting
/The Grok 0.5T model's open-sourcing is part of xAI's strategy to address concerns about AI transparency and accountability.
/Microsoft's Gemini AI is reportedly outperforming Google's Copilot, indicating a shift in AI capabilities.
/The No. 1 deep research system, Onyx, operates without web search access, contrasting with OpenAI's tool-reliant approach.
/There is a growing sentiment that AI technologies controlled by a few companies pose risks similar to nuclear weapons, prompting calls for nationalization or open-sourcing.
/Microsoft has banned its own engineers from using AI tools, indicating internal concerns about the reliability and security of these technologies.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/Uber exhausted its entire 2026 AI budget in four months as tokenmaxxing costs surged past expectations.
/Anthropic is reportedly valued at about $850B on $30B in revenue, closing a funding round exceeding $30B while finalizing a classified NSA surveillance-tools contract as OpenAI partners with the Pentagon.
/OpenAI reported a Q1 2026 operating margin of -122%, highlighting deep losses at frontier model scale.
/Norway deployed 2 petabytes of Huawei flash storage for large language model training and Huawei announced a Tau Scaling Law targeting 1.4nm-equivalent density without EUV by 2031.
/SpaceX’s Starship V3 flight was mostly successful, Starlink gen2 satellites reached 100MW of solar capacity, and the company filed for an IPO with dual-class founder control.
On Watch
/Developer reliance on Cloudflare for DNS, tunnels, Workers, and security continues to grow alongside explicit worries about centralization and monopolistic behavior, setting up a potential edge-infra chokepoint story.
/Autonomous ride-hailing via Waymo is delighting some users, including blind riders, but persistent skepticism about safety in emergencies and edge cases could slow broader adoption and keep gig-work disruption uneven.
/Microsoft canceling a 244-acre data center and cutting about 4% of its workforce in response to AI spending adjustments may signal an early pullback in hyperscale AI capex growth curves.
Interesting
/The Grok 0.5T model's open-sourcing is part of xAI's strategy to address concerns about AI transparency and accountability.
/Microsoft's Gemini AI is reportedly outperforming Google's Copilot, indicating a shift in AI capabilities.
/The No. 1 deep research system, Onyx, operates without web search access, contrasting with OpenAI's tool-reliant approach.
/There is a growing sentiment that AI technologies controlled by a few companies pose risks similar to nuclear weapons, prompting calls for nationalization or open-sourcing.
/Microsoft has banned its own engineers from using AI tools, indicating internal concerns about the reliability and security of these technologies.