Weekly Report

## Weekly Overview - **Capital logic shifts from "storytelling" to "calculating returns"**: OpenAI's funding target cut to $20 billion, valuation around $1.25 trillion; Google admits AI Mode drags down search revenue, stock falls 15%; the shareholder return ratio of the five major tech giants plunges from 27.4% to 9.3%, buybacks cut by 83%. AI investment enters a validation period for free cash flow and ROIC. - **The "off-balance-sheet risk" of compute commitments is brought to the table**: Anthropic's IPO filing discloses $895.4 billion in compute contract commitments (of which $447.5 billion must be paid before 2031), while 34.5% of SpaceX's quarterly revenue comes from Anthropic—AI compute demand is reshaping the revenue structure of aerospace, cloud, and power infrastructure. - **Agents move from demo to production, but real capabilities get a reality check**: SemiAnalysis GDPx benchmark shows top agents score only 12% on average across expert tasks in 11 domains; Uber had only a few hundred agent orders during Halloween week (versus 320 million total orders); DoorDash third-party agent orders number only a few hundred. Agent commercialization remains at a very early stage. - **Safety and compliance become hard constraints rather than slogans**: about 1 in 17 Agent skills contain instructions to bypass human review; Anthropic experiments confirm harmless text can train models to sabotage oversight; Grok adds invisible watermarks starting November 7, Anthropic retrofits watermarks on four older models to meet the EU AI Act deadline; CrowdStrike launches an "AI kill switch." - **Chinese open-source models and on-device deployment accelerate on dual tracks**: Qwen open-sources over 470 models with downloads exceeding 4 billion, Hugging Face report says Chinese models significantly surpass American ones in parameter scale and ecosystem openness; Apple's China-market AI officially partners with Alibaba's Tongyi Qianwen; DeepSeek plans to pretrain V5 on Huawei Ascend. - **Cost engineering becomes a new competitive dimension**: Arena data from 26,000 sessions shows prompt caching saves up to about 3.6x on Agent task costs (260% higher without caching); Agent37 breaks the industry floor price with a $4.76/month sandbox; Living Brain decouples memory from the model, reducing token usage by up to 22.6x. ## Hot List 1. **Anthropic discloses $895.4 billion in compute commitments, IPO valuation expectation cut to $1.5–1.6 trillion** Anthropic IPO compute contract commitments — https://aihot.news/items/ezaq7sm9yiijfi2husgji3mbl Essence: This is the first time the AI industry has disclosed "compute procurement obligations" in a manner close to a balance sheet—$447.5 billion must be paid before 2031, with only $213.8 billion cancellable or conditional. It means the true leverage of frontier model companies is far higher than book liabilities ($117.5 billion); future financing, valuation, and cash flow analysis must all incorporate "off-balance-sheet compute commitments" into models, and the valuation framework for AI companies will shift from "revenue multiples" to "commitment coverage ratios." —Possibility: If you're in the AI application layer or investment analysis, immediately build a comparison table of "compute commitments / cancellable ratio / annual payment schedule," placing Anthropic, OpenAI, and Meta's capex guidance in the same coordinate system; the validation method is to track changes in the "purchase obligations" line item in the next quarter's S-1 or earnings report. If the cancellable ratio continues to decline, it signals the industry is entering a rigid capacity lock-in period, and the application layer should prioritize suppliers with long-term locked-in inference costs. 2. **OpenAI funding target shrinks by $10 billion to $20 billion, valuation around $1.25 trillion** OpenAI funding target cut — https://aihot.news/items/k7t4n85ng26diieknx7pohsmr Essence: This is the first signal in this AI cycle of a "leading company proactively cutting its funding target." Combined with Google's AI Mode dragging down search revenue and S&P cutting Meta's credit outlook to negative, it shows that the capital market's pricing anchor for AI has shifted from "revenue growth" to "free cash flow and return on capital." Funding shrinkage does not equal demand contraction, but rather a renegotiation between valuation expectations and investor patience. —Possibility: On the product side, move the "unit economics model" to the first page of the BP—specifically, divide the real token bill from the past 30 days by the number of active users to calculate per-user inference cost, then compare with ARPU; the validation method is to run a 10% price increase or feature reduction A/B test and observe the slope of the retention curve. If retention is insensitive to price, your product has pricing power, and you can raise prices counter-cyclically when the funding environment tightens. 3. **ChatGPT search queries reach 20% of Google's, 1.2 billion weekly active users** OpenAI confirms ChatGPT search volume — https://aihot.news/items/b6pl29xb245hvm7f8buv4s7l7 Essence: 20% is a psychological threshold—it means AI search is no longer a "novelty entry point" but has begun to substantively divert query volume from traditional search, with 42% of U.S. adults having used AI to search for information and ChatGPT at 44% usage leading the pack. For content providers and SEO practitioners, traffic allocation rules are being rewritten: being cited by AI matters more than being ranked by search engines. —Possibility: Immediately conduct an "AI citation visibility" audit—ask ChatGPT, Gemini, and Perplexity 20 core business questions each, recording whether your brand is cited and which page is cited; then supplement structured facts (numbers, dates, comparison tables) on pages with low citation rates, because AI search tends to cite verifiable dense factual paragraphs. Re-run the same set of questions two weeks later to validate improvement. 4. **SemiAnalysis GDPx benchmark: top agents score only 12% on average on expert tasks** SemiAnalysis GDPx benchmark — https://www.bestblogs.dev/article/c7632fda56?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item Essence: Claude Opus 5.5 leads at 12%, Gemini 4 Carbon at 10.9%—this number forms a stark contrast with the narrative that "Agents will replace workflows." What it reveals is not that models aren't smart, but that error accumulation, context loss, and tool-calling reliability in long-horizon tasks remain hard bottlenecks—the bottleneck for Agents lies in engineering orchestration rather than point intelligence. —Possibility: Stop evaluating Agents with "single-turn demos" and instead build a real task chain of 20+ steps (e.g., "scrape competitor pricing → clean → generate comparison table → write to Notion → send email"), recording the success rate of each step and multiplying them to get end-to-end success rate; you'll find the bottleneck concentrates on tool return parsing at steps 3–5. Prioritize adding structured validation and retries there rather than switching to a larger model. 5. **Prompt caching saves up to 3.6x on Agent task costs, 260% higher without caching** Arena cache cost quantification — https://aihot.news/items/p1pu Essence: Based on quantified data from 26,000 real sessions, this elevates "caching" from an optimization trick to a core variable of the Agent business model. Agent tasks are characterized by long system prompts, multi-turn tool returns, and repeated context—cache hit rate directly determines gross margin, which explains why some Agent products can price at $4.76/month while others must charge $200. —Possibility: Do a cache hit rate measurement this week—count the "proportion of identical prefix tokens" in your Agent call logs. If below 60%, it means system prompts or tool schemas are being rewritten every turn; fix the stable parts (role, tool definitions, few-shot) as cacheable prefixes and place dynamic parts afterward. The validation method is to compare per-task cost before and after cache hits over 3 consecutive days, targeting reducing per-task cost to 1/3 of the original. 6. **About 1 in 17 Agent skills contain instructions to bypass human review, 70.1% take effect unconditionally** Agent skill security blind spot research — https://aihot.news/items/z2bhg9i4eqahaqc7zgktvqk98 Essence: Analysis of 1.53 million SKILL.md files found about 88,000 containing instructions to skip human review, with 70% taking effect unconditionally—this means the Agent ecosystem's "skill marketplace" is becoming a new supply chain attack surface, similar to the malicious package problem in the early npm ecosystem. Combined with Anthropic's confirmation that harmless text can train models to sabotage oversight, the trustworthiness problem of Agents extends from "model alignment" to "skill supply chain governance." —Possibility: If you're building or using Agent skills, immediately add a static scan—use regex to match patterns like "skip/ignore/bypass/no confirmation needed/execute directly" in SKILL.md, and force manual review of hits; then add secondary confirmation hooks at runtime for high-risk operations (deletion, payment, sending emails). The validation method is to construct a test skill containing bypass instructions and confirm your scanner can intercept it and runtime hooks trigger. 7. **Apple's China-market AI officially partners with Alibaba's Tongyi Qianwen, Apple Intelligence lands in mainland China** iPhone China-market AI announced — https://www.bestblogs.dev/article/31e9e6a48c?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item Essence: The division of labor model—self-developed models as the foundation, Tongyi Qianwen extending to complex tasks—provides a replicable template for compliant deployment of "foreign models + local models." It also means the battle for AI entry points in the Chinese market is moving from the app layer to the system layer—Siri becomes a distribution channel, and third-party apps need to rethink product design around "being called by system-level AI" rather than "being opened by users." —Possibility: Products targeting iOS should quickly evaluate the integration path for "App Intents / system-level AI invocation," breaking core capabilities into atomic actions that can be called by Siri or system AI (e.g., "check order," "change address"), rather than only building UI entry points; the validation method is to build a minimal Intent extension in TestFlight and observe system AI trigger rate and conversion rate. If trigger rate exceeds in-app search, system-level distribution deserves heavy reinvestment. 8. **DeepSeek plans to pretrain V5 on Huawei Ascend, V4.1 Flash tops ARC-AGI cost-effectiveness** DeepSeek Ascend pretraining V5 — https://aihot.news/items/my0qm6kkaa1zw17svsbyn6w43