# AI capex — X 热门讨论 (2026-09-21 04:13 UTC)
## @gherkinit (Pi-Fi) · 09-19 18:41 · ♥188 ↻42 💬9 Automated Bailouts https://x.com/gherkinit/status/2101381245947322391
## @qinbafrank (qinbafrank) · 09-20 10:44 · ♥65 ↻14 💬9 油价是决定加息路径的关键变量，能源冲击是核心扰动。这是高盛GBM系列播客 《The Breaks of the Game》的最新一集核心观点。这期是高盛对冲基金业务全球主管Tony Pasquariello主持，高盛研究部高级顾问Dominic Wilson与全球风险主管Josh Schiffrin主要分享。他们认为市场目前定价的约四次累计加息"明显偏多"，两人均认为本轮加息周期实际落地次数更可能为两次。 两人均指出，油价走势是决定最终加息路径的关键变量，一旦油价出现实质性回落，将为全球多类资产提供同步走强的条件。跟我之前很多观点一致性很类似，详细梳理下。
这期播客核心内容要点， 1、加息路径：市场定价“四次”明显偏多，两人更看好两次 市场当时已定价本轮累计约四次加息（一个较完整的紧缩周期）。两人认为这明显过头。基准情形是10月再加一次，然后进入较长时间暂停。实际风险更可能是各央行加息力度不及市场预期；
2、油价是决定加息路径的关键变量 油价大涨被两人视为今年最大的市场意外，持续推高通胀预期，也让市场对加息路径定价越来越激进。一旦油价出现实质性回落，通胀压力会缓解，加息必要性下降，并为全球多类资产同步走强创造条件。油价回落也被视为债市最可能的催化剂；
3、债市：长端实际利率已处多年罕见高位，利空消化较多，但还需催化剂 10年期美债约5%、30年期实际利率超过3%，是相当长时间未见的水平。财政赤字、通胀粘性等利空已被市场消化不少，估值吸引力上升。但在AI资本开支融资需求、财政赤字和增长韧性没有实质性变化前，收益率大幅下行空间有限。需要油价下跌这类偏多头催化剂，买家才会更积极回归；
4、股市：韧性超预期，当前是近期较清晰的做多窗口之一 尽管油价飙升、收益率上行、AI不确定性存在，股市仍展现超预期韧性。原因主要是大规模资本开支周期和实体经济增长依然强劲。当前点位已计入较悲观情景（四次加息、高收益率、油价超100美元），因此“看到缓解路径变得容易得多”。Wilson 将此时点描述为近期多头侧较清晰的做多窗口（虽不如7月底那么好，但感觉类似），可用低成本期权做保护或博取上行。
5、宏观与美联储会议后环境 9月FOMC后，短期政策不确定性下降，被视作配置窗口。增长仍有韧性（约2%），劳动力市场好于部分预期。通胀仍高于目标，但能源冲击是核心扰动。AI、财政赤字和资本竞争推高了实际利率，若没有AI capex转向或增长意外下行，鸽派转向空间有限。
6、高盛年底交易思路 1）美元：受美联储相对鹰派、美国经济领先支撑，倾向走强。 2）“全资产上涨”窗口：若中期选举后油价下行，可能出现股债商品等同步走强的交易机会，但需要油价下跌作为催化剂。 3）超长端高实际收益率：中长期仍有配置价值。 4）利率定价：对年底前加息定价向下修正越来越有信心，倾向用低成本期权结构参与。
这跟个人之前观点很类似，是油价再决定市场近期的走势，油价回落叠加基本面还是很强势，支撑本周后半周市场上行。后续还是要看油价走势： https://t.co/R8noHp3Gme > 引用 @qinbafrank: 把这三张图放在一起看，就能清晰看到是谁在影响市场走势，核心是油价。只要油价在下行，即使沃什放鹰导致US10y跳涨纳指暂短下跌，但很快就回到了受油价驱动的轨道上来。所以并不是什么加息利空还是利好之类的论点，因为都站不住脚。核心要去看为什么加息？因为通胀。为何通胀反弹或者下降慢？至少今年内都是油价导致的。长债收益率走高有相当一部分原因也是油价。这也是8月下旬在“收益率高中枢时代的结构性问题”的推文https://t.co/1Zk9fGfdda里聊到的“长债走高的三个驱动因素：油价、财政赤字发债、AI基建科技公司发债潮，最后就要下油价上下功夫”。
而昨天中国再次入局斡旋https://t.co/rMGS4XEUYg也给了市场一个很强的预期，然后昨晚又传出来美国与胡塞武装在阿曼的会晤、今天传出来川普下周会在纽约会见海湾国家领导人商议战后问题，这些都是可能让中东局势降温的潜在因素。油价回落-》通胀预期回落-》长债收益率回落-》市场情绪提振，自然沃什的加息和鹰派观点就往后站了。
看今天伊朗和胡塞的动态，貌似也很平静。看起来通过昨天向中国传递了信息和态度，也是在等下周川普与中东各国领导人的会晤和中美会晤的结果，自然是要等川普的实质性让步。未来一周估计会比较平静，局势平静自然油价回落，利好资产上行。
期待下周的两场会晤有一个好的结论，这个结论自然是川普做出实质性让步，这次可能沙特也要让步，伊朗和胡塞才能收手。关于让步的程度之前这里https://t.co/L533aRqZCZ有聊过。
当然还要看明天日本央行的动作，加息是确定的，核心还是对后续节奏的表述，日本央行的表态则可能决定市场会出现估值波动，还是叠加更强的杠杆平仓。当谈油价走势还是当下市场首要因素。 https://x.com/qinbafrank/status/2101623460615102871
## @TheVictorBuilds (TheVictorBuilds) · 09-20 09:08 · ♥42 ↻7 💬33 OPENAI IS SPENDING $856 BILLION ON COMPUTE. THE CAPEX WAR HAS ESCALATED
OpenAI just leaked its internal financial projections, and the numbers are completely redefining the scale of the artificial intelligence arms race. They are preparing to spend nearly $900 billion on data centers and computing infrastructure by the end of the decade.
As a data driven analyst, the sheer velocity of this capital deployment is historic. Based on the latest internal presentations:
+ OpenAI projects a cumulative compute and infrastructure spend of $856 billion by the end of 2030.
+ To fund this, they are forecasting a negative free cash flow of $278 billion between 2026 and 2030. Despite projected revenue climbing to $350 billion annually by 2030, the physical infrastructure costs will entirely dwarf their income.
+ To bridge this staggering financing gap, OpenAI is already in early talks for a new funding round that could value the company at over $1.2 trillion, a massive premium over its recent $852 billion valuation.
=> The traditional venture capital model is dead at this frontier. You cannot fund an $850 billion infrastructure buildout with standard equity rounds. OpenAI is transitioning from a software research lab into a sovereign scale infrastructure developer. By committing this much capital to physical data centers and energy grids, they are attempting to build an insurmountable hardware moat. My analytical take is that the frontier AI models of the 2030s will not be constrained by algorithmic breakthroughs, they will be constrained by physical gigawatts. The fact that OpenAI expects $278 billion in negative cash flow proves that they view short term profitability as a distraction. They are playing a winner take all game for the cognitive infrastructure of the internet, and they are willing to burn hundreds of billions to lock out the competition.
Do not get distracted by the consumer chatbots. The real alpha is in the physical layer. Stop focusing on software margins. If the leading AI company on earth is preparing for five years of massive negative cash flow, the software application layer will be heavily commoditized.
Position in the physical supply chain. That $856 billion has to flow somewhere. It is going to commercial real estate, cooling systems, fiber optics, and energy generation. Align your capital with the legacy industrials and utility providers that will physically build these data centers. Track the energy bottleneck. OpenAI's data center projects are targeting tens of gigawatts of capacity. Companies that can provide sustainable, baseload power to these hyper-scale facilities will capture a massive premium over the next five years.
The artificial intelligence war has moved from the cloud to the concrete. Follow the capital expenditure, understand the scale of the hardware buildout, and position your portfolio for the infrastructure supercycle. > 引用 @TheVictorBuilds: FEDERAL COURT REINSTATES THE 7 BILLION DOLLAR SOLAR FOR ALL PROGRAM.
A federal judge just blocked the current administration's attempt to claw back billions in clean energy infrastructure funding. The 7 billion dollar Solar For All program is officially back online.
I track the legal bottlenecks that control capital flow. The timeline of this ruling is critical: On September 18, 2026, U.S. District Judge Mary McElroy in Rhode Island ruled that the EPA illegally terminated the 7 billion dollar program.
The program, established by the 2022 climate law, was designed to make solar power accessible to more than 900,000 lower income Americans.
In August 2025, EPA Administrator Lee Zeldin rescinded the already obligated funds, labeling the initiative a boondoggle.
Judge McElroy vacated this termination, stating the EPA acted contrary to congressional intent and without statutory authority when it froze the grants.
The EPA is currently reviewing the decision and considering options for an appeal.
This is a massive legal precedent for the energy sector. For the last year, clean energy developers, nonprofits, and state level green banks have been entirely paralyzed by the political weaponization of authorized budgets.
By legally forcing the EPA to unblock these funds, the court is injecting 7 billion dollars of guaranteed, congressionally authorized capital directly back into the distributed solar market. This is a legally mandated capital deployment. My analytical take is that the residential solar sector, which has been starving for subsidized demand amidst high interest rates, is about to experience a massive localized boom as these state coalitions rapidly ramp up their delayed deployments. Capital is about to flow heavily into the clean energy pipeline. Here is your strategic framework:
+ Position in distributed solar operators. The companies that specialize in residential and community solar installations are the direct beneficiaries here. A federally backed mandate to equip 900,000 households will flood these regional installers with guaranteed contracts.
+ Track the state level coalitions. The funding flows directly through state agencies and nonprofits. Follow the localized contract approvals from state energy boards to see exactly which public and private contractors are winning the bids.
+ Monitor the appeal timeline. While the capital is legally unfrozen now, the EPA is exploring an appeal. Do not over leverage your clean energy exposure, but respect the immediate liquidity injection this ruling provides to the sector.
The regulatory freeze on clean energy has been broken. Follow the legal mandate, track the state level deployments, and position your capital for the infrastructure rollout. https://x.com/TheVictorBuilds/status/2101599409511584045
## @cryptob28811588 (Mr. Hunter) · 09-20 23:32 · ♥33 ↻0 💬27 .. @vangrid_io Technology
Smartphone → Sensor Node: Turns smartphones into a distributed sensor network for collecting real-world spatial data.
Multi-View Ingestion: Captures multiple views of the same area to enable 3D spatial reconstruction.
Edge Privacy: Processes sensitive data on-device to protect faces and license plates.
3D Reconstruction: Converts accepted captures into 3D spatial models that can be exported as GLB/ZIP.
Cryptographic Provenance: Uses blockchain to verify the origin and authenticity of collected data.
Base + USDC Escrow: Bounties are locked in USDC on Base before contributors complete tasks.
Watermarked Preview: Buyers can view watermarked previews before acceptance, while raw video remains protected.
Enterprise Spatial API: Provides spatial data for robotics, autonomous logistics, defense, and Physical AI.
Zero-CapEx Sensor Swarm: Uses existing smartphones to scale the network without requiring specialized hardware.
Core Pipeline: Smartphone → Spatial Data → Privacy → 3D Reconstruction → Provenance → Enterprise AI/Robotics. https://x.com/cryptob28811588/status/2101816677101183259
## @FrankieIsLost (frankie) · 09-21 01:08 · ♥44 ↻0 💬4 simply unexplainable why all this AI capex isn’t translating into real productivity growth > 引用 @sweis: 636606729769440499166579950236036751749912014371509557713570027508971809534551913252252094954941974952859310861988904737359709200557919 is a factor of RSA-896 https://t.co/yGWyEC8RpZ https://x.com/FrankieIsLost/status/2101840999320539394