每日信息简报｜2026 年 08 月 14 日

DAILY BRIEFING · 2026.08.14

把技术与商业叙事拉回证据、成本与边界

今天从 AI 基建、网络安全到资本市场与公共议题，关注的不是单个标题，而是主张的证据、适用范围与后续影响。

今日要点

The builder’s guide to GPT‑5.6

报道摘要：Learn how startups use GPT-5.6 to build faster, more cost-efficient AI agents with smarter model selection and new Responses API capabilities.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：OpenAI，发布于 2026-08-13。

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed

报道摘要：Preview Ultrafast, a new OpenAI API service tier that runs GPT-5.6 Sol up to 14× faster. Powered by Cerebras, it delivers up to 750 output tokens per second.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：OpenAI，发布于 2026-08-13。

OpenAI appoints Dali Rajic as Chief Revenue Officer

报道摘要：OpenAI appoints Dali Rajic as Chief Revenue Officer to lead its global revenue organization and help businesses realize the full value of AI.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：OpenAI，发布于 2026-08-13。

AI 与科技

Writer introduces new AI model and upgraded harness to contain token costs

报道摘要：Built as a post-training variation on Z.ai’s open source model GLM-5.2, Writer says the new system should provide deployment-ready capabilities at a much lower price.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：TechCrunch AI，发布于 2026-08-13。

Databricks wanted to raise $1B, investors wanted $15B. It settled on $5B at a $190B valuation.

报道摘要：AI is expensive, Ali Ghodsi tells TechCrunch. With so many investors wanting into his latest round, he said yes to more than planned.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：TechCrunch AI，发布于 2026-08-13。

OpenAI introduces ‘Ultrafast,’ a new mode that makes GPT-5.6 Sol work at 14x the speed

报道摘要：OpenAI is launching a preview of a sped up version of its latest, most powerful model, in an effort to court enterprise users.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：TechCrunch AI，发布于 2026-08-13。

值得深读

Microsoft’s Clippy-like Mico character is no longer the face of Copilot

报道摘要：Microsoft Copilot will no longer show its emotive yellow blob, Mico, when you use the chatbot’s voice mode. In a support page, Microsoft says it’s going to move Mico to its Learn Live platform, where the avatar will have "more to react to," as reported earlier。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：The Verge AI，发布于 2026-08-13。

Microsoft is combining its Copilot apps ahead of a ‘super app’

报道摘要：Microsoft is finally beginning to combine its consumer and commercial Copilot AI assistants into a single "super app" interface, starting with the Copilot and Microsoft 365 Copilot apps. Both personal and work accounts will be moved to the new unified app, whi。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：The Verge AI，发布于 2026-08-13。

OpenAI is losing its second executive this week

报道摘要：Another OpenAI executive is departing. Denise Dresser, who joined OpenAI as its chief revenue officer in December after serving as CEO of Slack, will be leaving in the "coming weeks" to "pursue other opportunities," she said in a team note posted to LinkedIn.。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：The Verge AI，发布于 2026-08-13。

Only Global Cooperation Can Keep the World Safe From A.I.

报道摘要：A conversation with Robert Wright, the author of “The God Test.”。 这条线索的价值在于把标题中的主张与其证据、适用范围和后续影响分开看。来源：纽约时报 AI，发布于 2026-08-13。

快速浏览

纽约时报 AI · 2026-08-13Another OpenAI Executive, Denise Dresser, Departs the A.I. Start-Up：Denise Dresser, who was previously the C.E.O. of Slack, is the latest in a string of executives to leave the artificial intelligence start-up.↗ 纽约时报 AI · 2026-08-13Why A.I.’s Promise Isn’t Resonating：On “The Opinions,” the Yale law professor and economist Natasha Sarin argues that A.I. could transform the U.S. economy for the better. But with the public hearing about layoffs and job losses, she says, it’s no wonder the technology’s economic promise isn’t r↗ Hacker News 精选 · 2026-08-13Accelerating GPT-5.6 Sol Ultrafast：Article URL: Comments URL: Points: 398 # Comments: 165↗ Hacker News 精选 · 2026-08-13Choose Boring Technology (2015)：Article URL: Comments URL: Points: 230 # Comments: 121↗ Hacker News 精选 · 2026-08-13Gemini 3.7 Flash：Comments URL: Points: 584 # Comments: 330↗ BBC · 2026-08-13Conditions on US aircraft carrier at sea for more than 250 days raise alarms：Thousands of sailors on the USS Abraham Lincoln have reportedly faced food shortages and broken plumbing, with some considering jumping overboard.↗ BBC · 2026-08-13Engine parts smashed Ryanair window that man's head was sucked out of, report says：A preliminary report by US investigators says an engine fan blade broke on last month's flight from Greece to Germany.↗ BBC · 2026-08-13Palestinians told to leave homes in village where Israeli settlers besieged houses：The mayor of Qusra says Israeli troops have been telling families to leave their houses, using some as barracks.↗

@jukan05 · 请求账号 @jukan05 · 2026-08-14 00:34:37 UTC · 原创原帖：Taiwan's DigiTimes reports that Nvidia's Feynman platform, the generation after next, will move straight to A16, TSMC's upgraded version of its 2nm process, and will also raise the share of SoIC 3D stacking, with custom HBM and CPO expected to be combined on top of that. Accordingly, TSMC's monthly SoIC capacity is said to reach 50,000 wafers a month by the end of 2027. What is also interesting is that AP7 P2, which was planned for SoIC, may be partly allocated to CPO and CoWoS lines. This could mean CPO is being pulled forward earlier than expected. ASE's subsidiary SPIL is also said to be establishing itself as a key vendor winning Nvidia orders. This stems from the fact that even if TSMC expands capacity at full force, it will be difficult to meet all customer demand.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 22:53:00 UTC · 原创原帖：That’s truly incredible guidance...↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-13 21:47:45 UTC · 引用原帖：Sometimes you have have to go /ultrafast.↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-13 21:39:07 UTC · 引用原帖：Computer history is here.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-13 20:21:49 UTC · 原创原帖：Implication of all this, of course, is that the big fall in the bilateral deficit that the Trump Administration trumpets as one of the successes of tariffs isn't real -- as the Chinese goods just enter after taking a one stop flight through other jurisdictions↗ @demishassabis · 请求账号 @demishassabis · 2026-08-13 20:13:33 UTC · 引用原帖：Oh yes, and Flash 3.7 is also lightning fast! ⚡️↗ @Claudia_Sahm · 请求账号 @Claudia_Sahm · 2026-08-13 19:04:57 UTC · 原创原帖：move over burrito ... pork belly sandwich (with collard greens and peppers) and fried pickles. (under $18 but not by much.) 😋↗ @demishassabis · 请求账号 @demishassabis · 2026-08-13 17:11:11 UTC · 引用原帖：Gemini 3.7 Flash brings major upgrades for software engineering, web dev & knowledge work. And introductory price is half the original 3.6 Flash cost. Happy building - enjoy!↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 17:01:48 UTC · 原创原帖：.@AtlantaFed Wage Growth Tracker ticked up to +3.8% in July, with job switchers continuing to see stronger wage gains than job stayers↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 13:59:53 UTC · 原创原帖：Aletheia dropped some juicy information today: - NVIDIA’s Rubin Ultra is unlikely to use HBM4 and will instead use 8-high and 12-high HBM4E. - Kyber has not been canceled, and the four-die Rubin Ultra remains on the roadmap. - The four-die Rubin Ultra could be used in Kyber. It could also be implemented through an OSAT by stitching together two dual-die packages on an ABF substrate. - The four-die Rubin Ultra could feature up to 768GB of HBM4E. Looks like a good time to buy memory again.↗ @nft_hu · 请求账号 @nft_hu · 2026-08-13 13:46:29 UTC · 引用原帖：DELL 在黑色7月的股价表现也是很强。↗ @nft_hu · 请求账号 @nft_hu · 2026-08-13 13:43:09 UTC · 原创原帖：$SMCI 财报后的一些理解： 1️⃣堆积订单很多，但是卡在了推进上。 客户数据中心的供电、液冷和网络就绪度延后；Q4 新订单超过 600 亿美元、期末 backlog 创纪录。 2️⃣毛利率大幅改善，但可能只限于当季。 环比 750bp 的毛利率提升中，约 75%来自客户与产品组合，以及部分合约自 Q4 递延至 FY27 Q1；约 25%来自关税成本下降和库存准备降低。 3️⃣更分散的客户和更高的附加值。 公司正尝试将价值链从服务器整机推向 DCBBS（整机、储存、液冷、CDU、网络、机房管理软件与服务）的整合方案。企业／通路收入在 Q4 达 56 亿美元、占比 50%，较前季 28%显著提高。 4️⃣政策风险并未消失。 2026 年 3 月美国检方对三名与公司有关人士提出的出口管制违规指控。其中包括公司高管、台湾销售人员及承包商；SMCI 本身并未被列为被告。 这个事件的还在等待处理中，这也是股价较同类型企业PE更低的原因之一。↗ @nft_hu · 请求账号 @nft_hu · 2026-08-13 13:30:12 UTC · 原创原帖：财报里相互映证的故事： $SMCI 一直有一个观察仓位，看了一下这次的财报很好。毛利率大幅上修，backlog订单量也是创纪录。唯一的问题可能是营收略不及预期。 管理层对于这一问题的解释是“主因是大型客户资料中心的供电、液冷与网路建置尚未准备完成。” 这反映 AI 基建的瓶颈正从 GPU 供给，延伸为机房层面的 power delivery、冷却、网通与 commissioning。 其实这个观点在去年下半年以来已经多次被提到。拿不到的PPA、漫长的审批、建造流程、设备洁净室短缺等等， 那这恰恰利好另一个主要持仓 $NBIS ——作为另一个可运行 AI capacity 的供给者。“online capacity”是现在的稀缺资源。若客户自己的 GPU 已到、但场地无法运作，已能交付 GPU cloud 的业者价值提高。这有利 NBIS 的长约谈判、预付款与产能消化。↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:26:27 UTC · 引用原帖：Very unsettling. Thanks for sharing, Eric. Here’s a piece from us earlier this year on the subject:↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:24:50 UTC · 引用原帖：I couldn’t love this more↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:09:23 UTC · 原创原帖：Initial jobless claims up 209k vs. 202k est. & 200k prior; continuing claims at 1.777M vs. 1.794M est. & 1.799 prior … greatest increases in MI (+1.9k), NY (+1.6k), & TX (+1.3k); greatest decreases in OH (-0.3k), LA (-0.2k), & IA (-0.1k)↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:08:51 UTC · 原创原帖：July PPI #inflation +4.7% y/y vs. +4.9% est. & +5.5% prior … core +4.2% y/y vs. +4.1% est. & +4.7%↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:08:27 UTC · 原创原帖：July PPI components that feed into PCE (Fed’s preferred measure of #inflation)↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 13:08:01 UTC · 原创原帖：July PPI #inflation 0.0% m/m vs. +0.2% est. & -0.1% prior (revised up from -0.3%) … core +0.2% m/m vs +0.3% est. & +0.4% prior (revised up from +0.2%)↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 13:03:01 UTC · 原创原帖：Anthropic’s ARR could reach $120 billion by year-end....↗ @demishassabis · 请求账号 @demishassabis · 2026-08-13 12:53:42 UTC · 引用原帖：SL2T is our amazing sign-language-to-text model allows users to sign directly to their phones for the first time. Built in close collaboration with the Deaf community, it’s a great example of the good that can be done with AI. Congrats to the team for the launch!↗ @nft_hu · 请求账号 @nft_hu · 2026-08-13 12:33:57 UTC · 原创原帖：美国7月PPI年率 4.7%，预期4.9%，前值5.50%。3月以来新低。↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 12:29:22 UTC · 原创原帖：Interesting. In Q2 2026, Intel lost server CPU unit share but gained revenue share. $INTC↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:49:15 UTC · 原创原帖：Misery Index (calculated as sum of y/y CPI and unemployment rate) rolled over again in July and now stands at +7.5%, supported by moderating inflation and slight downtick in unemployment↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-13 11:43:50 UTC · 原创原帖：In IPO news… Vantage Data Centers is exploring going public at an enormous $100B valuation. If you’re not familiar, $ORCL is their anchor tenant for Stargate (OpenAI end user), and Vantage develops/operates hyperscaler DCs. Anthropic IPO is also expected, with investors expecting $2T valuation, eclipsing $SPCX per FT. Yeah… idk about Anthropic valuations. Jim Cramer defending that valuation is never a good sign. On a side note, Opus 5 is probably the worst consumer LLM I’ve used to date, it’s like a 5 IQ LLM operating off terrible data retrieval. With a Fable-like God complex on top. Enterprise is probably having a better time though. Unitree IPO is also expected this month and priced it at a $9B valuation. Which is a positive tailwind for Unitree for the humanoid sectors. Not really an AI name but SHEIN is going public as soon as the 19th this month. Go ask any girl you know and they’re probably familiar with this platform.↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:41:07 UTC · 原创原帖：@MBAmortgage mortgage applications for week ended Aug. 7 rebounded to +3.6% after last week’s decline of -2.9% … 30-year mortgage at 6.77% vs. 6.81% prior (first decline in six weeks)↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:40:40 UTC · 原创原帖：July supercore #inflation eased to +2.84% y/y and continued its sharp decline from the recent May peak but remains above pre-pandemic average↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:39:47 UTC · 原创原帖：Share of core CPI categories with annualized m/m increases >4% rose to 41% in July while those with price declines (<0%) dropped to 36% Bloomberg @economics↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:39:01 UTC · 原创原帖：Prescription drug prices -0.8% m/m in July and are now falling at fastest pace on record, -3.1% year/year↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:38:16 UTC · 原创原帖：Sectors/indexes yesterday and MTD/YTD↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:37:48 UTC · 原创原帖：Baskets’ performance (March column represents correction month)↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:37:20 UTC · 原创原帖：Index drawdowns and % outperforming↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:37:02 UTC · 原创原帖：Neural9 chart/table↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:36:48 UTC · 原创原帖：Sectors’ % of stocks trading at 4w and 52w highs↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-13 11:36:22 UTC · 原创原帖：MA breadth charts↗ @Claudia_Sahm · 请求账号 @Claudia_Sahm · 2026-08-13 11:17:30 UTC · 原创原帖：Alphabet soup of the economy? My take on C vs K, and @MichaelRStrain makes great points, too.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 10:23:39 UTC · 原创原帖：Lenovo: AI overall is far from a bubble. Actually, it is just the first stage of AI democratization. As it evolves and becomes more accessible, it will only generate more demand towards AI democratization. With that, we believe demand for memory will continue to rise for a considerable period ahead, at least by the end of next year. That means that the supply will be still constrained.↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-13 10:17:49 UTC · 引用原帖：Okay here's my take on $POET, since apparently people are celebrating on the subreddit that I took small positions. - I have 0 idea. ZERO. How they convinced shareholders to allow them dilute SO MUCH to the point: They have an absurd ~$830m cash on hand now. And my guess is that they're close to done, since I would be in disbelief if they needed more. So now... they have ~$830m to grow their business like M&A, and EV actually dropped really low recently (eg. $1.25B MC vs. $830m cash on hand), that I thought it was compelling to take a risk. - Annualized production capacity target is 12M optical engines/year for 2027, which is pretty absurd ramp. Need to do more research into ASP for Poet's products, but maybe $125-$200? So at ~$150 ASP, $1.8B revenue ceiling. But they're like inverse AAOI right now: no visible demand to fill all that capacity. Aside from Lumilens, which is now a highly valued hyperscaler supplier (probably Amazon or Microsoft), with billions in customer agreements. The interesting thing is if Poet signed prepayment EML agreements with players like Mitsubishi, or CW agreements with Sivers behind the scenes with that balance sheet. And given the current bottleneck, players might use Poet as a workaround for supply procurement. But basically, you have a super cash-rich company with a ton of capacity coming online. And they're not really disappearing anytime soon with their financials. I don't quite think it's as technically defensible compared to upstream laser chokepoints with something like $SIVE around the same valuation in terms of technical moat for CW lasers vs. OE packaging side. As you've seen recently with Celestial and Poet. But maybe... there's a chance some hyperscaler or downstream company announces a volume agreement sometime in the future. and $POET takes off. And they do have the supply chain setup for ramp with that. That's the risk I ended up taking, but it's not exactly a clear as day long for me like $AAOI.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-13 10:14:47 UTC · 原创原帖：Michael Froman makes an important point. The longer China's existing growth model continues – powered, as it is, by surging debt – the more disruptive its adjustment is likely to be, not just for China but also for its trade partners. via @ForeignAffairs↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-13 09:46:26 UTC · 引用原帖：Feels like my $IQE thesis Feb this year got validated alongside $AXTI and others. (I still long on both) So I stopped covering them as much. But TLDR on updates: Early this year, IQE was a distressed but critical chokepoint in the optical supply chain with links to $LITE and others when I published my thesis on it. Since then: - $MTSI signed long term epiwafer supply agreements, took a stake in the company, now IQE is bank debt free (Macom has convertibles, but that's positive) - $TSEM signed agreements for InP epiwafers. - Management explicitly said its IQE agreement was signed to secure the "strategic III-V epiwafer supply required for that roadmap" - IQE signed another ~$14M mysterious AI DC contract. The only part that's still playing out with my earlier thesis is IQE converting their latent/underutilized capacity toward AI DC segment, to match peers like Landmark in the space (which takes time, like slightly over a year if I remember correctly). There's some national security implications that I'll avoid talking about that makes them more important. But basically: FY2025 revenue: £97.3m FY2026 guidance: over £126.5m They're rapidly growing again in the optical sector, out of bank debt/distressed liquidation state, and remains a critical chokepoint in Western optical supply chains. Now... with new agreements with your leading photonics companies for revenue growth, and players like Macom serving as a soft backstop. After a thesis gets validated, typically I just sit back and enjoy growth of a business.↗ @nft_hu · 请求账号 @nft_hu · 2026-08-13 09:45:11 UTC · 原创原帖：一张把scale up/scale out/scale across解释的非常清楚的图。 图源 ： @FundaAI↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-13 07:44:59 UTC · 原创原帖：$SIVE announces a $3.4M development program with SemiNex! For CPO, DFB laser arrays, and optical amplifiers. Early production is targeted H2 2027 (in line with early scale up CPO timelines) Seemed weird to me at first given product overlap, since SemiNex is a much smaller Series B company... But then I got reminded of a less explicit version of $SPCX + Cursor relationship (derisking by working together first). There's probably something interesting Sivers found in this company.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-13 05:09:21 UTC · 原创原帖：1/5 Valor: "Brazil's industrial production fell 1.8% in June after a 0.9% decline in May. June’s result showed broad-based weakness, with declines in 16 of 25 industrial sectors, as well as across all four major economic categories."↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-13 04:19:28 UTC · 原创原帖：According to Yicai, students with high enough gaokao scores to get into university are increasingly picking vocational schools over university, even though the former have lower status, because of less uncertainty over employment after graduation.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 03:12:45 UTC · 原创原帖：It seems Intel is bullish on CXL. Interesting. "Moving workloads or the KV cache outside the GPU does not automatically remove the bottleneck." "Effective KV cache offloading requires faster chip to chip connections or CXL connections rather than depending on PCIe alone."↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-13 03:01:43 UTC · 原创原帖：1/4 In his Zhu Rongji obituary, Keith Bradsher refers to the "grand bargain" Zhu negotiated with Chinese provinces, i.e. the 1994 "Tax-Sharing Reform". I think this, and not WTO admission, is perhaps his most important legacy.↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-13 02:52:11 UTC · 原创原帖：Is it just me, or is the list of 3 letter optical jargon never ending? We have LPO, LRO, NPO, XPO, CPO, OBO, and others for approaches. Then InP, EML, ELS, OCS, SOI, and others to add to the random terms. Even the industry events with OCP or OFC get their three letter names. People who have followed me on my optical bottleneck scavenger hunt with $SOI or $IQE might know a lot by now. But wouldn’t new followers might be super confused…↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 02:47:59 UTC · 原创原帖：GF Securities Overseas Electronics and Communications: CoreWeave 2Q26 Earnings Comment Maintaining Buy rating and $172 target price: CoreWeave posted 2Q26 revenue of $2.6 billion and adjusted EBITDA of $1.5 billion, both above consensus. The company currently operates 1.5GW of compute power, and contracted capacity stands at 4.2GW, keeping it on track toward its 8GW target for 2030. Management guided to midpoints of $230 million for 3Q26 adjusted operating income and $1,055 million for full year 2026. That implies 4Q26 operating income of roughly $770 million at the top end of guidance, pointing to a very strong fourth quarter. We remain positive on CoreWeave for the following reasons. 1) With AI infrastructure demand strong and cash flow pressure building at the major cloud operators, the market for emerging cloud service providers continues to expand. 2) Pricing power keeps strengthening, and the company raised prices across all products by 25% starting in July. 3) The backlog stands at $129 billion, giving very high earnings visibility. 4) The launch of the DDTL 5.5 financing structure has made financing available even for contracts as short as three years. Shorter contracts generally carry higher prices. The company raised its 2026 capex guidance to $35 billion to $39 billion, but the 2Q26 EBITDA improvement and continued power capacity expansion have eased market concerns to some degree. We forecast revenue of $12.7 billion, $27.3 billion and $41.8 billion for 2026, 2027 and 2028, and adjusted EBITDA of $7.5 billion, $15.6 billion and $21.2 billion. Our $172 target price is unchanged and is derived by applying 10x 2028 estimated EV/EBITDA. A direct beneficiary of the expanding emerging cloud market: The market is concerned that the entry of new players such as Meta and SpaceX will intensify competition and disperse demand. We judge, however, that the current buildout plans of these players are not enough to resolve the structural shortage of compute resources. More importantly, the long term take-or-pay contracts CoreWeave signs offer greater revenue stability than on demand rentals or short term contracts. The company's total operating costs are also up to 47% lower than the average hyperscaler, giving it a clear edge in cost and operating efficiency. The path to margin improvement is on track as well. 1. As we noted in our initiation report, new bookings are capturing the full benefit of higher GPU rental rates. According to management, margins on 2Q26 new contracts are 5 to 10 percentage points higher than on existing contracts. On top of that, with the 25% across the board price increase in July, the $25 billion of new contracts to be signed in 3Q26 will carry even higher pricing. 2. The model of continuing to monetize assets after contracts expire is delivering. Some A100 GPUs deployed in 2020 are already contracted out through 2029. 3. The push into the Asia Pacific market is also accelerating. The company has secured 360MW of compute capacity in Indonesia, where low local data center operating costs should further improve profitability. $CRWV↗ @jukan05 · 请求账号 @jukan05 · 2026-08-13 01:21:18 UTC · 引用原帖：HBM will actually be in shorter supply in 2028 than in 2027—meaning that the HBM configurations of 2028 AI accelerators will also be scaled back.↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-13 01:01:37 UTC · 引用原帖：Old news actually from a bunch of days ago, but crossed that 15M. Enjoy a nice reset everyone. Landing in the next hour or so, go /fast.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-12 23:56:22 UTC · 原创原帖：SK Securities: Memory Revaluation Stemming from a Shift in the Perception of Compute Perception of compute: from consumable to asset AI compute is not a consumable. Rather than a technology good whose value erodes quickly, it could become an infrastructure asset that is redeployed across generations and generates recurring cash flows. From this perspective, AI CapEx is not simply a recurring cost but a process of accumulating a compute stock that can be monetized multiple times going forward. CoreWeave's recent earnings release shows this possibility. Even A100 based resources, six years past launch, are being signed into long term contracts running through 2029, and prices for older GPUs are also holding at meaningful levels. This does not have to be read purely as a supply shortage. Customers choose the cost relative to the performance their workload requires, not the fastest GPU. The newest GPUs handle frontier training and high performance inference, while existing H100/A100 can shift to general inference, fine tuning, batch workloads and the like. This is the concept of compute tiering, analogous to memory tiering. Once the perception of compute changes, the lens through which the expansion of AI infrastructure finance is viewed also changes. If AI infrastructure such as GPUs can generate revenue repeatedly through recontracting after the initial contract ends, it becomes a financeable infrastructure asset rather than simply a piece of electronic equipment. BEP within three years, and the period after that may be the real profit zone The same logic can apply to hyperscaler FCF. Amazon stated that its investments in servers and networking equipment reach BEP on average in less than three years. This means that if the economic life of compute assets including GPUs extends beyond three years, the subsequent period of use may be not simple principal recovery but a zone that generates additional profit. Revenue from new GPUs and second life revenue from the existing compute stock would occur at the same time. If, as with the A100, assets remain applicable into their ninth year, it means a considerable, highly profitable excess return zone opens up between the payback period the market worries about and the actual economic life. The change in memory's status is a structural phenomenon. A change that has to be accepted Once the perception of AI CapEx changes, it is entirely natural that memory's status changes. AI is an asset heavy industry, and physical supply capacity constraints in the real world are not something that can be changed. The reason five year LTAs with high binding force have been signed also becomes clear. Memory has a material impact on the customer's "true" profit zone. If memory does not make money, meeting demand will remain out of reach, volumes will not be secured while memory prices alone rise, and the future revenue curve will be revised downward. Memory making a lot of money is structural, and it is a change that has to be accepted. Inflection point in the change of memory valuation metrics: shareholder returns strengthen in earnest through greater earnings durability and visibility SK Securities maintains its Overweight view on memory. We focus on memory's structurally enhanced earnings power and on the shareholder returns that will begin in earnest, driven by greater durability based on a long cycle. Combined estimated operating profit for Samsung Electronics and SK Hynix is KRW 641tn in 2026 and KRW 977tn in 2027, among the highest globally, and aggressive shareholder returns are expected to be the key foundation for building confidence in the long cycle and for memory's revaluation. We are wary of peak out logic based on the inevitable slowdown in the rate of memory price increases. This is now an era in which the strength of the cycle is absorbed not only through "price" but also through "durability and visibility."↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-12 22:59:29 UTC · 引用原帖：My main takeaway was that your merchant CW laser suppliers seem to be EXTREMELY IMPORTANT now. But TLDR of $COHR earnings: "given the demand that we see in our DC business with transceivers, I don’t see any time in the near future where we would be selling indium phosphide lasers externally" - Ton more value placed on companies that sell InP lasers externally if $AAOI can't doesn't enough capacity for first gen-CPO deployments, $COHR not selling externally... I guess that's why $MTSI said a ton of customers are approaching them with urgency for InP CW DFB lasers (and they dont even have meaningful capacity coming online until EOY 2027) $AAOI -> Internal $COHR -> Internal Buyers -> fewer places to look for capacity. "Our backlog now extends out, fiscal 2027 is basically completely booked out. We are booked really through the end of calendar 2027" [ Massive massive demand visibility] - Coherent also mentioned LTAs spanning end of decade. So this kinda reinforces what $POET CEO said at their AGM that the "big 3" are fully sold out next 2 years. "we’re not constrained in the assembly and test capacity right now. We’re really just constrained by the ramp of the indium phosphide production" - Just for the bottleneck bros in terms of assembly/test not a bottleneck. "absolutely no push-out of CPO demand. In fact, it has been the opposite. We have seen demand increase and request from customers’ demand getting pulled in" "we continue to expect revenue from CPO for scale-up applications to start to flow in the second half of calendar 2027" - Big read through on other CPO players for scale up if $LITE, $COHR reaffirm revenue from CPO scale up coming h2027. - "quarterly revenue exceeding $3 billion by the end of fiscal 2027" - $COHR revenue go brrr So for regular numbers it's: - $2.046B revenue (+13.3% Q/Q, +33.8% Y/Y), main revenue ramp looks to be happening entering Q4 into 2027 - Non-GAAP gross margins: 40.2% Next quarter guide is: - Revenue: $2.2–$2.4B revenue - Non-GAAP gross margins: 39.5-41.5% Basically: $1.69B -> $1.81B -> $2.05B -> ~$2.30B -> $3.0B (end of FY 2027) ~39% -> 39.6% -> 40.2% -> ~40.5% (gross margins) For operating margins: 18.0% -> 19.5% -> 19.9% -> 20.3% -> 21.8% Little less impressed compared to $LITE margin growth in terms of profitability, but revenue scale for $COHR growing rapidly.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-12 22:39:07 UTC · 原创原帖：Korean media reported that Samsung Electronics is expected to unveil concrete shareholder return measures, including a special dividend, along with its next three year policy, as early as this month. SK Hynix is also expected to unveil its own plan within the third quarter.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-12 20:07:28 UTC · 原创原帖：Hard to overstate the extent to which the global customs trade surplus is concentrated in a single region of the global economy. East Asia's surplus in the last few months,annualized, is ~1.7 trillion.* * would be even higher if China's gold imports were netted out↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-12 18:57:50 UTC · 原创原帖：Typical conversation with @ajambrosino A: core alignment has been reached T: core alignment you say A: i say T: say more A: more And yet I’ve never been more excited about what him and the team are building.↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-12 18:50:05 UTC · 原创原帖：July U.S. federal budget deficit widened to -$432B vs. -$346B & -$291B prior, marking highest monthly total in more than five years … Medicare costs, Social Security and net interest were the largest expenditures↗ @Claudia_Sahm · 请求账号 @Claudia_Sahm · 2026-08-12 16:19:13 UTC · 原创原帖：The July CPI continued to build the case for disinflation. Core CPI rose 0.2% in July (or annualized 2.6%), firmer than flat prices in June, but still helping to move the trend down. Even so, two months is not enough, and the PPI will also shape whether it's a good PCE print.↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-12 15:22:19 UTC · 原创原帖：Looking forward to joining @PowerLunch on @CNBC around 2:20pm ET today↗ @jukan05 · 请求账号 @jukan05 · 2026-08-12 15:10:44 UTC · 原创原帖：【Tianfeng Securities Overseas Tech】 After reviewing all the various AI related earnings, we believe the AI bull market is returning on the back of strong fundamentals One very clear trend is emerging from the recent earnings of companies across the AI industry chain. Growth in cloud businesses is accelerating again, and compute demand continues to exceed supply. Customers are showing a willingness to accept price increases and even to prepay, and the unit economics of AI infrastructure are improving. Market attention is shifting away from concerns about excessive capital expenditure (CapEx) and back toward the belief that demand and profitability will actually materialize. Large cloud: The slope of cloud business growth turned upward this quarter, and the scale of new bookings in the quarter was overwhelmingly larger than in any of the past 20 quarters. Google has effectively no choice but to go all in. The mindset is that failure means the end. Even with negative free cash flow (FCF), every $1 of incremental cloud revenue converts into roughly $0.54 of new operating profit. On a simple calculation, the payback period is about 2 years. AWS's margin of 39.4% is very strong. This provides an answer to the questions about return on investment (ROI). AWS will ultimately become a business with $1 trillion in annual revenue, with roughly 5x of growth headroom over the long term. The payback period for the cloud business is also under 3 years. Microsoft: Azure's growth reaccelerated, providing an answer to concerns about the durability of that growth. Neocloud: Earnings, demand, pricing power, and financing capability have all been validated across the board. CoreWeave's (CRWV) recent compute resources are effectively all sold, with multiple customers competing to secure GPU volume. It added about 500MW of active power in a single quarter, and its year end target has been raised to more than 1.85GW. Its customer base is also broadening from large AI model companies and cloud giants to industrial, financial, life sciences, government, and traditional enterprises. AI is moving from a training centered frenzy toward a phase of adoption across every industry. CoreWeave raised prices across all SKUs by about 25% in July. The profit contribution rate on new contracts rose by 5 to 10 percentage points. This shows that customers judge the returns they earn from AI to be sufficient to absorb higher compute prices. Even the A100, launched in 2020, is being signed into high priced contracts running through 2029. This also refutes the core bear case that GPUs are rapidly rendered obsolete within 3 to 4 years and that their residual value goes to zero. Nebius (NBIS) saw total new contract value in the second quarter grow roughly 4x versus the prior quarter, with new customer contracts up more than 9x. Price tolerance on the demand side is so strong that, under current terms, it could sell all of its 2027 capacity today. Annual revenue on new contracts reaches $20 million to $25 million per MW. Pricing on prior generation GPUs also rose more than 30% versus the prior quarter, and short term total bookings reach as much as $40 million to $50 million per MW. Customers have begun prepaying the cost of building AI infrastructure. About 70% of new contracts include prepayment provisions, which can cover 50% to 60% of the related CapEx. Expected project payback periods have shortened from 2 to 3 years previously to 1 year and 10 months. This means AI infrastructure is not simply a business that scales while burning cash, but one that has already secured strong unit economics. Optical communications: The FCC related disruption is limited and the practical impact should be minimal. Lumentum (LITE) reported results above market expectations and raised guidance. EML demand exceeds supply by about 30%. The shortage of ultra high power lasers also continues to widen. Innolight's laser products are effectively sold out, and it plans to expand capacity roughly 4x over the coming quarters. The bottleneck remains delivery capability. Quarterly OCS revenue clearly exceeded $100 million for the first time. NPO is not a technology that replaces CPO but a newly added intermediate architecture. Key customers' CPO plans are unchanged and demand signals are getting stronger. AI architecture is becoming increasingly optically integrated. Memory and storage: It is now nearly certain that a bottom is forming. The core logic is to take time to secure room for recovery. The absence of price increases after the share price decline is already reflected in market expectations, and visibility on long term agreements (LTA) and shareholder returns is steadily improving. Compute: The market narrative is shifting back toward AI training and open source. We are positive that Nvidia (NV) will challenge its prior high, and a narrative around RSI is also gradually forming.↗ @LizAnnSonders · 请求账号 @LizAnnSonders · 2026-08-12 14:58:32 UTC · 原创原帖：It has been an honor over the past several years to be Vice Chair of Nantucket’s Boys & Girls Club (NBGC), while more recently also serving on the Board of Trustees of Boys & Girls Clubs of America (BGCA), for which Charles Schwab has been a corporate sponsor for many years. We have NBGC’s Tim Russert Summer Groove annual gala on Nantucket this coming weekend, so I thought I’d share a link to our silent auction, which has some amazing items! Regardless of your association with Nantucket, these are for art, music, golf, and experience lovers everywhere; all for a wonderful cause.↗ @nft_hu · 请求账号 @nft_hu · 2026-08-12 14:34:58 UTC · 原创原帖：从洗盘的干净程度和情绪准备，我觉得像样的行情真的快开始了。↗ @nft_hu · 请求账号 @nft_hu · 2026-08-12 13:59:42 UTC · 引用原帖：虽然Coreweave 也不错，但我更相信NBIS的纪律性。 拖家带口大搬迁，孤注一掷输不起。↗ @nft_hu · 请求账号 @nft_hu · 2026-08-12 13:58:00 UTC · 引用原帖：最有说服力就是自己出钱买↗

YouTube 更新

今日最值得看

Money & Macro：Why the US isn’t broke yet

本期由 Money & Macro 围绕“Why the US isn’t broke yet”展开。频道简介提到：🤖 Let an AI agent handle your SEO and marketing work with Letaido, powered by Ahrefs data! 👉🏻 Try it using my link: If you appreciate the research, consider buying me a ‘coffee’ at or supporting long-term for membership benefits via: WANT AN IN-DEPTH INDUST。 它适合作为线索入口；其中的解释、预测或因果判断仍应结合一手资料和反例核对。

Matt Wolfe：Making An Actually Fun 3D Game with AI

本期由 Matt Wolfe 围绕“Making An Actually Fun 3D Game with AI”展开。频道简介提到：I vibe coded a full game sequel using Claude Code and Codex, and it proved something important about AI and game development. A year ago I built The Librarian, a playable vibe coded game you can still find on GitHub. This time I wanted to see how far AI coding。 它适合作为线索入口；其中的解释、预测或因果判断仍应结合一手资料和反例核对。

其他频道更新