Frontier AI is turning into a barbell: OpenAI is loading up on massive, long‑dated capex under a damaged governance story while Anthropic quietly outruns it on revenue with its own capacity crunch. At the same time, AI infrastructure is becoming a geopolitical target and a power bottleneck just as open models, agents, and regulated vertical apps start to look like the parts of the stack that actually work today.
The big question is how much risk you really want tied up in a few contested data centers and a handful of people versus the messier, smaller bets already throwing off real cash.
Key Events
/Iran threatened the ‘complete and utter annihilation’ of OpenAI’s planned $30B Stargate AI data center in Abu Dhabi.
/An 18‑month investigation found Sam Altman privately lobbied against AI regulations he publicly supported, deepening internal distrust at OpenAI.
/OpenAI forecasts about $121B in compute spending by 2028 with no profitability expected before at least 2030.
/Oracle plans to cut over 10,000 jobs, with the layoffs publicly linked to its commitments to OpenAI.
/Anthropic has reportedly surpassed OpenAI’s $25B ARR and is preparing for an IPO.
Report
OpenAI is trying to buy the future with money it doesn’t have yet while its CEO bleeds political capital in public. At the same time, the physical and political ground under AI infrastructure is starting to look a lot less solid.
frontier labs: capex mania meets governance risk
OpenAI is guiding to around $121B of compute spend by 2028 and tells investors not to expect profits before at least 2030. On top of that sits a reported $300B deal with Oracle that is already showing up in Oracle’s plan to cut more than 10,000 jobs tied to its OpenAI promises.
An 18‑month New Yorker investigation says Sam Altman lobbied against AI regulation he publicly backed, sought billions from Gulf autocracies, and tried to bury an internal post‑firing probe, which has OpenAI insiders openly questioning his trustworthiness.
In the same window, Anthropic is reported to have passed OpenAI’s $25B ARR and even exceeded a $30B run‑rate, while still throttling users because of a compute crunch and scrambling for TPUs via an expanded Google–Broadcom partnership.
ai infra as a wartime target and power bottleneck
Iran explicitly threatened the ‘complete and utter annihilation’ of OpenAI’s planned $30B Stargate data center in Abu Dhabi, putting a bullseye on flagship AI campuses in geopolitically exposed regions.
In parallel, the US and Israel are bombing Iran’s South Pars petrochemical hub, which supplies about 85% of its petrochemical output, feeding into higher fuel prices.
Amazon has already pushed a 3.5% surcharge onto third‑party sellers, explicitly tying it to war‑linked fuel costs. Roughly half of planned US data center projects are delayed by supply‑chain constraints, and there are open calls for moratoria on new data centers that critics say could jeopardize US tech dominance as rivals scale AI.
model commoditization: meta, gemma, and the collapse of the api premium
Meta plans to open source its next generation of AI models while its internal dashboards chew through more than 60 trillion tokens in a recent month, a scale that lets it treat models as cheap infrastructure rather than the product.
Open‑source models are already outperforming proprietary systems in some evaluations, undercutting the idea that closed automatically means better.
Google’s Gemma 4 now runs fully offline on phones, with an iPhone 17 Pro pushing about 40 tokens per second versus 13 on a Pixel 10 Pro, and the model is trending on Hugging Face within days.
NVIDIA has a quantized 31B‑parameter Gemma variant with roughly 4x weight reduction, and hobbyists are benchmarking 37‑model suites on MacBook Air‑class hardware, signaling that ‘good enough’ LLMs are rapidly becoming cheap, local commodities.
agents and vertical ai: real jobs, real regulators, right now
AI agents are already credited with displacing significant numbers of jobs over the past year, not in theory but in actual deployments. Klarna fired 700 people on the back of its AI rollout, then admitted it had over‑automated and started rehiring to fix customer experience, exposing how easily the AI narrative can overshoot operational reality.
At the same time, Orchids—now Bud—claims to be the first agent with a full computer and says it has already passed seven figures in ARR, and an EU payment SDK now lets AI agents move money directly.
Regulators aren’t asleep: Noah Labs’ voice‑based AI just got FDA breakthrough status for five‑second heart‑failure detection off 3M vocal samples, and a San Francisco startup secured approval for AI to prescribe psychiatric meds, putting autonomous or semi‑autonomous models directly into clinical decision loops.
Meanwhile, AI‑enabled cyberattacks are getting more capable with a doubling time of 9.8 months for models released since 2019 and 5.7 months for those since 2024, and DeepMind is publishing ‘agent traps’ that can hijack autonomous agents against their users.
europe’s gold and nukes: quiet prep for a multipolar world
France has pulled its remaining gold from the US, upgraded to higher‑purity bars, and booked about a €15B gain while holding roughly 2,437 tonnes, explicitly citing long‑running worries about keeping reserves abroad.
Germany already repatriated about half its gold from New York and Paris between 2013 and 2017, but domestic critics still question why the rest remains under foreign control.
Commentators tie these moves to France’s preparation for a post‑Pax Americana order, which includes expanding its nuclear arsenal for the first time in 30 years, authorizing forward‑based nukes, and building the 80,000‑ton ‘France Libre’ carrier alongside a new UK–France nuclear pact.
All this plays out as the US’s global reputation hits historic lows and allies openly doubt its reliability, intertwining monetary, military, and tech‑supply dependencies.
What This Means
Capital, compute, and credibility are concentrating in a few AI and geopolitical nodes at the exact moment those nodes are becoming more fragile and contested. The gap between how ‘inevitable’ the AI future is sold and how constrained and politicized the underlying machinery has become is widening fast.
On Watch
/AI‑generated singers recently occupied eleven spots on the iTunes Singles Chart, an early sign that synthetic performers can dominate mainstream music rankings.
/A San Francisco startup has received approval for AI systems to prescribe psychiatric medications, pushing automated decision‑making further into high‑liability clinical practice.
/Researchers have trained rat neurons for real‑time AI computations, hinting at future hybrid bio‑digital architectures beyond current silicon limits.
Interesting
/Researchers at Tufts University developed a hybrid neuro-symbolic AI approach that consumes up to 100 times less energy than current standard systems, indicating a shift towards more sustainable AI technologies.
/Concerns about foreign investments in US media are rising, especially after a $24 billion deal involving Saudi and Qatari funds.
/Discussions are ongoing about utilizing surplus resources from satellite networks for data center-like functions, challenging traditional definitions of data centers.
/The trend of companies laying off AI teams due to underperformance suggests a potential 'AI bubble' that may be bursting.
/A single individual in China can now produce full manga episodes in just two days using AI, a task that previously required full teams months to complete.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/Iran threatened the ‘complete and utter annihilation’ of OpenAI’s planned $30B Stargate AI data center in Abu Dhabi.
/An 18‑month investigation found Sam Altman privately lobbied against AI regulations he publicly supported, deepening internal distrust at OpenAI.
/OpenAI forecasts about $121B in compute spending by 2028 with no profitability expected before at least 2030.
/Oracle plans to cut over 10,000 jobs, with the layoffs publicly linked to its commitments to OpenAI.
/Anthropic has reportedly surpassed OpenAI’s $25B ARR and is preparing for an IPO.
On Watch
/AI‑generated singers recently occupied eleven spots on the iTunes Singles Chart, an early sign that synthetic performers can dominate mainstream music rankings.
/A San Francisco startup has received approval for AI systems to prescribe psychiatric medications, pushing automated decision‑making further into high‑liability clinical practice.
/Researchers have trained rat neurons for real‑time AI computations, hinting at future hybrid bio‑digital architectures beyond current silicon limits.
Interesting
/Researchers at Tufts University developed a hybrid neuro-symbolic AI approach that consumes up to 100 times less energy than current standard systems, indicating a shift towards more sustainable AI technologies.
/Concerns about foreign investments in US media are rising, especially after a $24 billion deal involving Saudi and Qatari funds.
/Discussions are ongoing about utilizing surplus resources from satellite networks for data center-like functions, challenging traditional definitions of data centers.
/The trend of companies laying off AI teams due to underperformance suggests a potential 'AI bubble' that may be bursting.
/A single individual in China can now produce full manga episodes in just two days using AI, a task that previously required full teams months to complete.