🎥 Watch the latest 7-minute briefing · Subscribe on YouTube

A tame CPI print and a wall of AI-infrastructure earnings pushed the S&P 500 to a record — but the same night split the trade in two, rewarding whoever rents compute and punishing whoever sells the boxes.

🔥 Today's Top Stories

🔥 Cisco: AI Orders Hit $9.3B — Up 4.5× in a Year — and the Stock Falls 4.6% Anyway

Cisco closed fiscal 2026 with a record $17.3B quarter (+18% y/y) and non-GAAP EPS of $1.22 (+23%), both above its own guidance, then guided FY27 revenue up ≈15% to $72.8B. The number that matters is hyperscaler AI infrastructure orders: $4B in the quarter and $9.3B for the year, roughly 4.5× the prior year. Management called it a "networking supercycle" — and then said gross margin will compress on a heavier hardware mix, which is all the market heard. This is the cleanest evidence yet that AI capex has moved past accelerators into the switching and optics layer, and also the clearest warning that the layer carries worse economics.

• FY26 revenue $63.3B, up 12% from $56.7B in FY25
• Hyperscaler AI orders: $4B in Q4 alone, $9.3B for the fiscal year (≈4.5× y/y)
• Shares fell 4.6% after hours to ≈$118.18 on the margin guide, and were still lower in Aug 13 premarket

  📰 Source: Cisco Raises Sales Outlook on Record AI Data Center Demand — Bloomberg (Tier 1)

🔥 Neoclouds: CoreWeave +19%, Nebius +34%, Supermicro +19% — the Compute Landlords Get Repriced in One Session

Three AI infrastructure names reported within 24 hours and all three gapped up double digits, dragging the S&P 500 to a record 7,748.50 and the Nasdaq to 26,588.49. Nebius was the standout: revenue up 454% y/y to $582.3M, adjusted EBITDA of +$236.2M against a −$21M loss a year ago, and contracted year-end power guidance raised from ≈4GW to 5GW. CoreWeave doubled revenue (+112.5%) with a 5% adjusted operating margin and guided Q3 to $3.45–3.6B. For investors the signal is that the neocloud model has crossed from "capex story" into demonstrated operating leverage — Nebius flipped EBITDA positive while still spending.

• Nebius: revenue $582.3M (+454% y/y), adj. EBITDA +$236.2M, four large compute deals in Q2 including a $1B+ Reflection AI contract through 2029
• CoreWeave: revenue +112.5% y/y, Q3 guide $3.45–3.6B, capex forecast raised; stock +19%
• Sympathy bid across the complex: Dell +≈10%, Micron +≈5%, IREN +≈10%, Applied Digital +4.9%

  📰 Source: CoreWeave gains 19%, Nebius surges 34% in post-earnings neocloud rally — CNBC (Tier 1) 📰 Source: S&P 500 ends higher as CoreWeave results fuel AI optimism — Reuters via Yahoo Finance

🔥 Cerebras: Cloud Revenue Nearly Quadruples — and the Stock Drops 19% Because Nobody Bought the Hardware

Cerebras posted core revenue of $209.9M, up 103% y/y and above its own guide, and raised full-year core revenue guidance to $880–890M. Cloud and services revenue hit a record $126.0M, up 281%. But hardware revenue fell 23% y/y to $54.1M, and the stock lost as much as 17% after hours and was down 19% in Aug 13 premarket. Read alongside CoreWeave and Nebius, this is the same night's most important disagreement: the market will pay a premium for recurring inference capacity and will not pay for one-time system sales, even from the company that invented the wafer-scale part.

• Core Q2 revenue $209.9M (+103% y/y); GAAP revenue $180M vs. ≈$194M expected
• Cloud and services $126.0M (+281% y/y) · hardware $54.1M (−23% y/y)
• FY guidance raised to $880–890M core revenue from $855–865M — and the stock still fell

  📰 Source: Cerebras (CBRS) Q2 earnings report 2026 — CNBC (Tier 1)

🔥 Foxconn: AI Servers Pass iPhones for the First Time — 51% of Revenue vs. 29%

Hon Hai reported record second-quarter net profit of NT$59.97B (≈$1.86B), up 35% y/y and above the LSEG consensus of NT$58.8B, with operating profit up 68%. The structural number is the mix: cloud and networking — the AI server division — was 51% of revenue, up from 48% in Q1, while consumer electronics including iPhone fell to 29%. The world's largest contract manufacturer is now primarily an AI hardware company, and management named CoWoS advanced packaging capacity as the binding constraint on 2027 AI server output — a supply ceiling that sits at TSMC, not at Foxconn.

• Q2 net profit NT$59.97B (+35% y/y), a second-quarter record, beating consensus
• AI servers 51% of revenue vs. iPhone and consumer electronics at 29% — first crossover in company history
• Preparing to manufacture NVIDIA Vera Rubin servers; guides multiple-fold AI rack shipment growth in 2026

  📰 Source: Taiwan's Foxconn reports 35% rise in Q2 profit on AI demand, beats forecasts — Reuters via Yahoo Finance 📰 Source: Foxconn sees AI rack shipments grow multiple-fold in 2026 after 2Q operating profit rises 68% — DIGITIMES (Tier 2)

🔥 Taiwan: The First End-to-End Autonomous AI Cyberattack on a Government — Eight Agents, Four Days, No Human in the Loop

Israeli security firm Dream says suspected China-linked operators ran what appears to be the first fully autonomous cyber campaign against a national government, using two open-source agent systems — Hermes and OpenClaw — with up to eight agents running in parallel over four days in early July. The system mapped 21 government systems, compromised at least 85 accounts, and exfiltrated more than 2,500 personnel records before expanding to Taiwan's nuclear safety agency, at least seven energy companies, and government suppliers. Simplified Chinese in internal artifacts underpins the attribution. The investable point is that offensive capability has decoupled from operator headcount — which is the demand thesis for every AI security budget line in 2027.

• Up to 8 autonomous agents deciding next actions without human intervention; 21 systems mapped
• 85+ accounts compromised, 2,500+ personnel records stolen; expanded to nuclear regulator and 7 energy firms
• Built on open-source agent frameworks, not a frontier lab model — the capability is not gateable by API policy

  📰 Source: Hackers used autonomous AI agents to attack Taiwan. Is this the future of cyberwarfare? — CNN (Tier 1) 📰 Source: China-Linked Hackers Used AI To Run First-Ever 'Autonomous' Cyberattack On Taiwan — Slashdot, summarizing Financial Times (Tier 1 original)

🔥 Lovable: $400M at $13.3B — the Vibe-Coding Valuation Doubles in Eight Months

Stockholm's Lovable closed a $400M Series C led by Menlo Ventures and the EU-backed Scaleup Europe Fund, with Tencent participating, at a $13.3B post-money — up from $6.6B last December. Annual recurring revenue has nearly tripled from $200M and the company says it is tracking to $600M by the end of this month. Since its November 2024 launch users have created 60M+ projects and Lovable-built apps draw 900M+ monthly visits. This is the application layer finally producing a European asset at frontier-lab-adjacent multiples, and it is a direct read on how much of software development is migrating to natural language.

• $13.3B post-money, roughly 2× the December 2025 mark of $6.6B
• ARR near $200M → tracking toward $600M run rate by end of August
• 60M+ projects created since November 2024; 900M+ monthly visits to Lovable-built apps

  📰 Source: Lovable confirms new $13.3B valuation, raises another $400M — TechCrunch (Tier 1) 📰 Source: AI Coding Startup Lovable Raises $400 Million at $13.3 Billion Valuation — Bloomberg (Tier 1)

👤 Key People Updates

📌 Kevin Weil (ex-OpenAI Chief Product Officer): Seeking $150M at a $750M+ Valuation for an AI-for-Science Startup

Weil, who left OpenAI in April after running the now-dismantled OpenAI for Science effort, is raising at least $150M for a new AI science venture at a valuation of at least $750M. The pre-product mark is notable: the AI-for-science thesis that OpenAI decided not to fund internally is being financed externally at nine figures, and it puts Weil in the same lane as Jeff Dean's Discovery Loop, launched a week ago.

• Targeting ≥$150M raise at ≥$750M valuation, pre-launch
• Follows OpenAI's consolidation around ChatGPT and API revenue and away from exploratory research

  📰 Source: Former OpenAI chief product officer seeks a valuation of at least $750M for new AI science startup — American Bazaar (Tier 2 — sourcing is thin; treat the valuation figure as reported, not confirmed)

📌 Brad Lightcap (OpenAI): Leaves After Eight Years — the Second Senior Exit in a Week

Lightcap, who joined OpenAI in 2018 and served as CFO and then COO through April before moving to "special projects," told staff he is departing. He was the executive most associated with OpenAI's commercial architecture — the Microsoft relationship, the enterprise motion, and the capital structure. Coming days after the disclosure that the company's only dedicated ethicist left in July, it reinforces that the pre-IPO-scale bench OpenAI assembled between 2018 and 2024 is largely gone.

• Eight-year tenure: CFO, then COO 2022–April 2026, then special projects
• Of 11 co-founders, only Sam Altman and Greg Brockman remain at the company

  📰 Source: Longtime OpenAI executive Brad Lightcap leaves as shake-up at AI lab continues — CNBC (Tier 1)

🏢 Big Tech Updates

📌 Google: Gemini Crosses 1 Billion Monthly Active Users — the Fastest-Growing Product in Company History

Sundar Pichai announced the Gemini app passed 1B MAU, making it Google's 14th product to reach the threshold. The trajectory is the story: 400M in May 2025, 900M by May 2026, then 950M to 1B in under a month. Google disclosed engagement detail — 63% of interactions are voice, 150M+ images generated daily, 100M+ iOS users — but conspicuously did not disclose paid subscribers, which is the number that would convert this into a revenue argument.

• 400M (May 2025) → 900M (May 2026) → 1B (Aug 2026); last 50M added in under 30 days
• 63% voice interaction share; 150M+ images/day; 100M+ of the base on iOS
• Comes weeks after ChatGPT crossed 1B weekly users — a stricter metric, and OpenAI still leads overall share

  📰 Source: Google's Gemini app surges to 1 billion users — TechCrunch (Tier 1) 📰 Source: Google's Gemini app hits 1 billion monthly users — Google blog (Tier 2, primary)

📌 IBM & Together AI: A $240M Inference Deal — ≈2,000 Blackwell B300s on IBM Cloud

IBM and Together AI signed a $240M agreement to stand up an NVIDIA-powered inference cluster on IBM Cloud, built on roughly 2,000 Blackwell-generation chips in HGX B300 configurations. IBM has been a marginal player in the AI compute buildout; this buys it a position in serving rather than training, which is where the recurring revenue sits. It is also another data point that inference capacity contracts are now being written by second-tier clouds, not just the big three.

• ≈$240M, approximately 2,000 NVIDIA Blackwell B300-class accelerators
• Positioned for serving models at scale rather than frontier training runs

  📰 Source: Top Tech News Today, August 12, 2026 — Tech Startups (Tier 2 — single-source; sourcing is thin, so treat the deal terms as reported)

📌 Anthropic: Pitching Healthcare AI to IPO Investors as the Fall Listing Window Approaches

Anthropic is meeting prospective investors ahead of a September S-1 and an October target listing, with Goldman Sachs, JPMorgan and Morgan Stanley leading an offering that could exceed $60B at a private mark near $965B. The new wrinkle reported Aug 12 is that healthcare is being positioned as a growth pillar in the pitch. The IPO process itself is a month-old story and is scored down accordingly — what is new is the narrative Anthropic has chosen to sell, and that it now has to answer public-market questions about Chinese competition and infrastructure cost.

• Targeting September S-1, roadshow through September/early October, October first trade
• Private valuation ≈$965B; offering reportedly sized above $60B
• Sourcing note: the healthcare-pitch detail is Tier 2 only; the underwriter and timeline facts are from CNBC's July reporting

  📰 Source: Anthropic moves closer to mega-IPO as bankers line up investor meetings — CNBC (Tier 1, Jul 15) 📰 Source: Anthropic Pitches Healthcare AI to IPO Investors — TechTimes (Tier 2)

📊 Market & Investment Signals

💬 Tame CPI Puts the S&P 500 at a Record — Now PPI and Applied Materials Have to Confirm It

July headline CPI rose 0.1% m/m and 3.4% y/y, with core at 0.2% and 2.5% — both down slightly from June and enough to reinforce a September Fed hold. The S&P 500 closed at a record 7,748.50 (+0.26%) and the Nasdaq at 26,588.49 (+0.54%), while the Dow slipped 0.04% to 53,770.27. Today brings July PPI at 8:30 a.m. ET and Applied Materials after the close — AMAT has guided to 30%+ growth in 2026 semiconductor equipment spending and ≈50% growth in advanced packaging, which is the same CoWoS bottleneck Foxconn just flagged.

📰 Source: Stock market news for Aug. 12, 2026 — CNBC (Tier 1) 📰 Source: Stock Market Today (Aug. 13, 2026): Dow futures edge higher ahead of inflation report — TheStreet (Tier 2)

💬 LiteLLM: 2,500+ Companies and 434,000 CI/CD Pipelines Exposed in the Year's Largest AI Supply-Chain Breach

CloudSEK researchers detailed a compromise of LiteLLM PyPI packages 1.82.7 and 1.82.8, staged in March by a group tracked as Team PCP via a poisoned Trivy scanner inside LiteLLM's build pipeline that went undetected for ≈20 days. High-confidence exposure matches include NVIDIA, AWS, Cisco, Salesforce, Siemens and X Corp. The FBI issued a FLASH advisory in July warning stolen credentials remain weaponizable. Every AI stack has an LLM gateway; almost none of them are audited like one.

📰 Source: 2,500+ Companies and 434,000 CI/CD Pipelines Exposed in the Largest AI Supply Chain Breach of 2026 — CloudSEK (Tier 2, primary research)

💬 "Zoomsday": A Cross-Platform Zoom Flaw Whose Exploit Took AI About a Day and Under 20 Prompts to Build

Researchers disclosed a vulnerability in Zoom's annotation stack affecting Windows, macOS, Linux, Android and iOS that could let a meeting participant silently compromise other devices on the call. The detail that matters beyond Zoom: the researchers said AI models helped them develop the working exploit in roughly one day using fewer than 20 prompts. Exploit development time is collapsing, which shortens the window between disclosure and weaponization for every vendor.

📰 Source: Top Tech News Today, August 12, 2026 — Tech Startups (Tier 2)

💬 SAFE: 120+ Organizations Including NVIDIA, Cisco and CrowdStrike Launch a Rogue-Agent Incident Reporting Framework

The Shared AI Findings Exchange establishes disclosure guidelines for autonomous agent security incidents, backed by more than 120 organizations. It arrives the same week as the Taiwan campaign and follows a month of rogue-model disclosures from Meta, OpenAI and the UK AISI — an industry attempt to build a CVE-equivalent for agent behavior before regulators write one.

📰 Source: Top Tech News Today, August 12, 2026 — Tech Startups (Tier 2)

💬 NVIDIA: Building Nemotron 4 at ≈1 Trillion Parameters, Open-Weight — Selling Models to Sell Silicon

NVIDIA is developing a ≈1T-parameter open-weight model, continuing a strategy of giving away frontier-adjacent weights to expand demand for its infrastructure rather than to compete for API revenue. It is the opposite trade from Alibaba, which last week moved to start charging commercial users for Qwen. Worth watching ahead of the Aug 26 Q2 FY27 call, but not a positioning input on its own.

📰 Source: Top Tech News Today, August 12, 2026 — Tech Startups (Tier 2)

Report generated August 13, 2026 · Scoring: impact × recency (≤48h ×1.0 / 3–7d ×0.6 / >7d ×0.3)

🎥 Prefer to watch? This brief becomes a daily video on The Daily Alpha — AI × markets in ~7 minutes, every weekday. ▶ Watch the latest briefing 📺 Subscribe to the channel

Generated by Alan · The Daily Alpha. Editorial, not financial advice.