系统性脱节：狂热与失控的隐喻
在科技界充满非理性繁荣的当下，整个行业正陷入一种“重演未来、忽视当下”的系统性脱节。正如一位创作者在毫无规划、未作草图与色彩研究的情况下，便急于将一幅拙劣的丙烯画付诸画布（这无异于一场视觉灾难），当今估值达数十亿美元的数据中心项目和 AI 战略也呈现出极其相似的盲目性。这种现象的背后是科技行业对未来图景的过度贩卖，以至于忽视了现实世界正在其身后分崩离析。近期行业内发生了三起表面上毫无关联的事件——一位科技公司首席执行官在电视直播中长达二十分钟的情绪失控、一位亿万富翁指控教宗是“中国代理人”、以及美国多位政治家因支持数据中心建设而在选举中落败。这些事件并非孤立存在，而是科技产业在追求非线性技术跨越的同时，由于缺乏审慎规划和现实反馈，而在心理防线、商业模式及社会政治层面引发的集体性焦虑与反弹。
Original English Source
Before I begin, I think it's only fair that I disclose a conflict of interest. This week, I made the worst painting I have ever produced. I looked at the blank canvas with the confidence of a Silicon Valley tech bro announcing a revolutionary AI product, skipped every sensible step in the creative process, no sketch, no color study, no planning of any kind, and deployed directly to production. This was the outcome, or shall we call it a visual war crime that has not permanently lowered the property value of this room, and in that sense, it has something important in common with several billion-dollar data center projects. Both appear to have been produced under the working title lol 2026. I have deliberately left it hanging behind me for the duration of this video because if I'm going to spend the next 20 minutes criticizing billion-dollar strategic failures, it seems only fair that I am judged under the watchful gaze of one of my own. The people closest to me were not sufficiently upset about this monstrosity, so I'm going to need the perfect strangers of this audience to endorse my retirement from visual art. I trust you'll be honest. The comments are open. Please be brutal. As evidenced by this painting, my ego is currently as inflated as a tech valuation, and it definitely needs a pin. In this video, I'm going to cover three stories that on the surface have nothing to do with each other. A tech CEO's 20-minute meltdown on live television, a billionaire accusing the Pope of working for the Chinese Communist Party, and a wave of US politicians losing their elections because of data centers. But, all three are symptoms of the same thing. An industry that has spent so much time selling the future that it's forgotten to notice the present falling apart behind it. And if that sounds dramatic, well, so is this painting.商业倒挂：Karp的失控与算力主权
在 AI 商业化落地的博弈中，企业客户与前沿模型供应商之间的利益冲突已近乎白热化。Palantir 首席执行官 Alex Karp 在 CNBC 节目中的失控表现，实质上是代理了美国企业界对现有 AI 计费模式的强烈控诉。Karp 直言不讳地将 OpenAI 和 Anthropic 等前沿模型提供商所采用的按 Token 计费的定价模型斥为“极度疯狂”，并视其为对企业征收的“财富税”。这一商业摩擦暴露出深层的系统性矛盾：企业和政府客户强烈渴望拥有算力所有权、模型权重和自主的数据栈，以确保其数据主权与成本控制，而非长期向硅谷寡头租用 API 接口。然而，算力（Compute: 用于运行 AI 模型的计算资源）的稀缺性与前沿模型极高的运行成本，使得绝大多数企业无法独自构建或维护数据中心。这导致整个产业在“开源模型的自主化部署”与“闭源前沿模型的租赁”之间陷入了痛苦的商业拉锯。
Original English Source
On July 1st, Palantir CEO Alex Karp appeared on CNBC Squawk Box to discuss his company's new partnership with Nvidia to build AI infrastructure for the US government. That was the plan anyway. What followed was a nearly 20-minute televised meltdown complete with stuttering, nervous backtracking, flailing arms, and a steady supply of digressions so abstruse that the host appeared genuinely concerned for his well-being. It had the distinct energy of a man trying to explain quantum physics while actively being hunted by bees. Karp was supposed to explain Palantir's deal, but instead he launched into a sprawling broadside against the entire AI industry. He called the token-based pricing models used by OpenAI and Anthropic effing insane. He said something has gone completely wrong with how AI is sold to enterprises. He called the pricing structure a wealth tax on businesses. He demanded that viewers pick up the phone and call any CEO in private, say "Madman Karp is on TV saying we're livid." And see they're twice as livid as me. When the CNBC anchor interjected and said, "You sound pretty angry." Karp snapped back and said, "No, this is the voice of American business that is being channeled through me." Which is a terrifying thought because if American business sounds like that, it desperately needs a kiss on the forehead and a bit of a nap. At multiple points he veered into whether elite universities would accept him as a professor, which is apparently an aspiration his parents still hold for him. My parents hold a similar hope for me, but I feel I'm just not old enough for that yet considering the greatest professor of all time other than my dad is Professor Dumbledore, and I simply haven't mastered the facial hair required for the tenure track yet. Karp asked the host if they were off camera, they were asking him to just wrap it up, but he just kept talking. Legend say he is still talking to this day whispering about enterprise architecture into the cold empty void of the studio after the lights go out. Now I have to be fair here. Buried underneath this chaos, Karp was making an argument that isn't entirely without merit. His core claim is that enterprise and government customers want to own their compute, their model weights, and their data stack rather than renting access through API token pricing from frontier model providers. He asked, "Are we really going to outsource the battlefield of this country to the consensus view in Silicon Valley?" He argued that companies are frustrated in private, but won't speak publicly because they fear upsetting their AI vendors. There is a kernel of something real there. Owning your own infrastructure gives you sovereignty over your data and your costs, but the practical reality is that compute is just not easy to come by. The reason companies like OpenAI and Anthropic are still operating at a loss is precisely because the infrastructure required to run frontier models is extraordinarily expensive. Not every company has the capacity to build or maintain its own data center. What Karp is essentially advocating for is either widespread adoption of open-source models, which are improving but not yet at parity with closed frontier models for many use cases, or a fundamental shift in how compute is provisioned and priced. Both of those are valid long-term directions. Neither of them is helped by screaming about them on live television while the host try to cut to a shot of Air Force One. This painting and that interview have one thing in common. Both began with irrational confidence and absolutely no pilot study. The difference is that I accept responsibility for my failure. Karp called himself the voice of American business. I'm just the voice of someone who should never have been allowed near acrylic paint. But Karp's performance didn't happen in isolation. It happened 2 days before his co-founder said something significantly worse.地缘偏执：极端的地缘政治指控
在技术商业博弈之外，科技精英的深层意识形态偏执正将公共讨论推向荒谬的极端。Palantir 联合创始人、亿万富翁投资者 Peter Thiel 在阿斯彭论坛上，公开指控天主教会历史上首位美籍教宗 Leo XIV 为“中国共产党的特工”。Thiel 这一惊人逻辑链条的起点，是教宗于 2026 年 5 月发布的名为《人类圣母赞歌》（Magnificat Humanitatis: 关于 AI 监管的教宗通谕）的四万字通谕。该通谕呼吁对人工智能实施国际监管，警告其可能带来的去人性化风险与失业潮，并强烈要求解除自主武器系统的武装。Thiel 的地缘政治逻辑认为，这一具有广泛道德感召力的通谕能切实约束西方（特别是美国）的 AI 研发进程，却根本无法影响中国的研发决策。因此，他得出结论：教宗的道德警示客观上延缓了美方的 AI 军备竞赛，等同于为中国服务。这一指控暴露出极端的冷战式对抗思维，也映射出硅谷精英在面对社会伦理和地缘博弈双重压力时，所展现出的心理脆弱性与偏执倾向。
Original English Source
On June 30th, Palantir co-founder and billionaire tech investor Peter Thiel sat down alongside political scientist Francis Fukuyama for a panel titled Humanity at the End of History. During the discussion, Thiel accused Pope Leo the 14th, the first American-born pope in the history of the Catholic Church, of being a Chinese Communist agent, which is a logical leap so spectacular it makes you wonder if Thiel's data analytics software is currently using the same decision-making framework that led me to believe this color palette was salvageable. His reasoning was that the Pope's first encyclical, Magnificat Humanitatis, a 42,000-word document released in May, called for international regulation of artificial intelligence, warning that AI heightens the risk of dehumanization, and that the pursuit of profits must not lead to choices that systematically sacrifice jobs. The encyclical also warned against autonomous weapons systems and declared that AI must be disarmed. Because the Pope's message could influence Americans, but it's unlikely to be taken headed off in China, Thiel argued the encyclical would slow down only one side of what he framed as a race between the US and China to advance AI. Therefore, Thiel concluded, the Pope is working for the Chinese Communist because nothing really screams Marxist revolutionary quite like the Bishop of Rome. The Aspen audience laughed, the Vatican did not respond, and why should they really? This is not Thiel's first confrontation with the Holy See. In March, he delivered an invitation-only lecture series on the Antichrist in Rome, just blocks from the Vatican, which reportedly unnerved Vatican officials and prompted two Catholic universities to publicly distance themselves. Imagine being so exhausting that an institution that has spent 2,000 years perfecting the art of theatrical drama and flawless tailoring looks at you and basically says, "Sweetie, you're trying too hard." Thiel has previously argued that the Antichrist could manifest not as an individual, but as a world government that seizes power by promising to protect humanity against existential threats, which is an interesting position for the co-founder over surveillance company that provides AI-powered intelligence tool to governments and militaries worldwide. Talk about conflict of interest, right? At the same event, Thiel also claimed that Anthropic, which he described as a woke liberal company, is winning the AI race and would rig the 2028 elections using its AI models to completely outwit Elon Musk's efforts through X. Anthropic declined to comment pointing to a blog post on election integrity and political bias, and Thiel, presumably embarrassed by how little he appears to know about AI, has reportedly created a foothold in Argentina as a backup country just in case everyone in the United States found out that he still thinks the Matrix was a documentary and that LLM actually stands for a large lunch menu. This, by the way, is the sort of contingency planning that has historically done an extraordinary amount of heavy lifting in expressions of patriotic confidence. Now, as somebody who was born in the Vatican, let me just say it very clearly, leave the Pope alone. The man is successfully running a 2,000-year-old global enterprise spanning billions of souls. He does not have time to debug Peter Thiel's late-night paranoia about of vampire blood and AI apocalypse cults. I read Magnificat Humanitatis in full. I watched the livestream when it was released because I'm probably one of the Vatican's biggest groupies, and the document is not anti-AI. It is not calling for a ban on artificial intelligence. It is calling for collective responsibility, for measured development, for the protection of human dignity, for regulation that prevents the worst excesses while preserving the potential. It warns against autonomous weapons, it warns against mass surveillance, it warns against displacing workers without care for their livelihoods. It's something we've discussed many times on this channel. It was also in the letter that I sent to Pope Leo last October. And the Holy See has been engaging with the AI industry directly, including notably collaborating with Anthropic on questions of AI ethics. This is not a reactionary institution shouting from the sidelines. This is an institution trying to participate in the conversation by asking the experts themselves. Calling the Pope a Chinese agent because he thinks AI should be regulated is, and I'm trying to find the most measured way to say this, a bit of a leap from an encyclical about human dignity to an accusation of espionage. I slashed my painting when I realized it was beyond saving. Thiel's approach to public discourse appears to be very similar, except he seems to be slashing the Pope.物理阻击：数据中心的能耗与民意
在数字世界的技术畅想之外，AI 产业正遭遇物理世界与基层民主的沉重阻击。2026 年 7 月初，美国犹他州参议院议长 J. Stuart Adams 及 Box Elder 县委员 Lee Perry 在党内初选中惨败。这两人落选的直接导火索，是他们极力推动的“Stratus 数据中心项目”——这是一个规划于大盐湖附近、能耗需求高达 9吉瓦（GW: 1吉瓦等于100万千瓦）的庞然大物。9吉瓦的电量甚至超过了犹他州全州的用电总量，这种为了极少数人生成虚拟娱乐图像而牺牲全州电网稳定性的做法，彻底激怒了选民。这并非孤立事件，数据中心监测（Data Center Watch: 专注于分析数据中心对资源及环境影响的独立监测机构）的数据表明，全美已有价值约 640亿美元 的数据中心项目因地方社区的强烈反对而被搁置或延迟。这场物理层面的抵制呈现出鲜明的跨党派特征：民主党选民主要担忧其对水资源（数据中心平均每千瓦时 compute 消耗达 1.8 至 4.3 升水）和生态环境的破坏；共和党选民则强烈反感企业获得的巨额税收减免与政府补贴。这种来自基层的民主反弹表明，当 AI 的发展成本以更高的电费账单和干涸的水源形式转嫁给普通民众时，公众将用选票捍卫其生存资源的完整性。
Original English Source
While the tech CEOs were melting down on television and accusing religious leaders of foreign intelligence work, something quieter, but arguably more significant, was happening across the United States. On July 2nd, Utah State Senate President J. Stuart Adams, one of the most senior politicians in the state, lost his primary elections. Former Box Elder County Commissioner Lee Perry conceded his primary race the same day. Both had backed the Stratus data center project described as one of the largest in the world, backed by Shark Tank investor Kevin O'Leary, planned near the Great Salt Lake. The project would have required up to 9 gigawatts of power. Just to put that in context, that's more electricity than the entire state of Utah currently uses. Just so a few thousand people can generate hyperrealistic images of Elmo fighting Darth Vader in 4K. Priorities, everyone, please. And this is not an isolated incident, of course not. According to Data Center Watch, $64 billion worth of data center projects across the United States have been blocked or delayed by local opposition. In Cascade Locks, Oregon, voters recalled two port commissioners who had supported negotiations over a proposed data center. The project was subsequently canceled. In Warrenton, Virginia, all four council members who voted to approve an Amazon data center were voted out over two election cycles. In El Paso, Texas, a recall petition was filed against a city representative who voted to keep a Meta data center incentive deal. In King George County, Virginia, the board chair accused Amazon of hiring a company to dig up dirt on people and declared that the country was ready to go to war and honestly, nothing unites local American politics quite like the shared bipartisan dream of turning an Amazon data center into a very expensive, very warm community tomato farm. The opposition is bipartisan, 55% Republican, 45% Democrat. On the left, the concerns center on environmental impact. On the right, they center on tax abatements and corporate subsidies. Both sides agree on the basic grievance. These facilities consume enormous amounts of electricity and water, drive up utility costs for residents, and are often approved without adequate public consultation. A Pew Research survey of more than 8,500 US adults found that Americans are far more likely to view data centers negatively when it comes to their impact on energy bills, environmental strain, and nearby living conditions. And data centers consume an average of 1.8 L of water per kilowatt hour of compute energy, rising to 4.3 L in air-cooled hot regions. For communities in water-stressed areas, this is an existential concern, not just some policy abstraction. This is what the AI backlash looks like when it stops being abstract. It's no longer AI might take jobs someday. It's why is my electricity bill higher now? It's why is my water table dropping? It's why did nobody ask us before approving a facility that uses more power than our entire state? The backlash has gone physical, local, and democratic, and it's costing politicians their entire careers.效能飞跃：非线性技术突破与责任
在商业狂热与物理阻击的夹缝中，技术本身正以超越预期的非线性速度实现自我突破。2026 年初，苏黎世联邦理工学院（ETH Zurich）及法国国家信息与自动化研究所（Inria）记录显示，LLM 已在 2025 年普特南数学竞赛（Putnam Competition: 北美最具声望的大学生数学竞赛）中取得了研究员级的顶尖成绩，比行业预测提前了整整七年。在底层效能方面，产业界也开始打破“算力翻倍、能耗翻倍”的传统线性扩张模式。塔夫茨大学（Tufts University）研发的神经符号 AI（Neurosymbolic AI: 结合神经网络与结构化规则推理的混合系统）在提升准确率的同时，将能耗降低了百倍；同时，混合专家模型（Mixture of Experts: 仅激活模型中与当前查询最相关的专业子模块的架构技术）与键值缓存优化（Key-Value Cache Optimization: 减少对话期间模型需要调取的上下文记忆量以降低计算开销的技术）的普及，使得单位查询能耗不再随模型能力的提升而呈线性增长。这些科学突破表明，算力约束是可以通过技术创新得以缓解的，但这绝不能成为管理层进行轻率、越权和无视社会成本进行部署的借口。行业亟需建立起包含多方利益相关者在内的共同责任（Collective Responsibility）机制，确保颠覆性技术的迭代不以牺牲人类尊严和地方社会福祉为代价。