Daily News Coverage — August 13, 2026: CoreWeave's $104B Backlog Confirms the AI Infrastructure Boom Is Still Accelerating
Thursday, August 13, 2026 — Today's newsroom judgment call: the clearest proof yet that AI infrastructure demand isn't cooling came from a neocloud's earnings call, not a hyperscaler's. Here's what else moved.
THE LEAD
CoreWeave's Backlog Hits $104 Billion as CEO Says Capacity Is "Sold Out"
Opening: CoreWeave's Q2 results landed well ahead of estimates and included a detail that matters more than the beat itself: the company says it's out of capacity to sell.
What happened: CoreWeave reported Q2 revenue up 112% year-over-year to $2.58 billion, against a $2.56 billion estimate, and disclosed a revenue backlog of $104 billion — up from $25 billion just six quarters ago — with 1.5 gigawatts of active contracted power. CEO Michael Intrator described capacity as sold out. Shares jumped as much as 18% in early trading. The backlog figure doesn't even include more than $25 billion in new customer commitments signed in early Q3. Net losses widened to roughly $626 million as debt, now around $35 billion, drove up interest costs.
Why it matters: This is a cleaner read on real AI infrastructure demand than a hyperscaler's capex guidance, because CoreWeave's backlog is committed customer revenue, not a spending plan. A $104 billion backlog on a company that had $25 billion just eighteen months ago says demand is still outrunning supply, not leveling off as some investors have started to worry.
What to watch: Customer concentration remains a real risk — reporting has flagged that the backlog leans heavily on OpenAI and Anthropic, with Microsoft and Nvidia historically representing the bulk of revenue. Whether CoreWeave's 10-K disclosures this cycle show that concentration easing or deepening will matter more than the headline growth rate.
AI & Frontier Models
Nvidia Ships an Open 30B Agent Model and Is Reportedly Building a 1-Trillion-Parameter Successor
What happened: Nvidia released Nemotron 3.5 Lightning, an open 30-billion-parameter mixture-of-experts model with only 3 billion active parameters, plus NeMo Switchyard, an open-source routing library the company says cuts agent benchmark costs by roughly 60%. Separately, The Information reports Nvidia is developing a much larger successor, Nemotron 4, targeting over 1 trillion parameters — up sharply from Nemotron 3 Ultra's 550 billion, though still smaller than leading Chinese open models.
Why it matters: Nvidia is now shipping open-weight models the day after Meta's own open agent release, explicitly positioning free, efficient AI as a driver of chip demand rather than a threat to it — CEO Jensen Huang said as much directly. It's also a direct bid to keep the open-weight performance crown in the U.S. as Chinese labs have pulled ahead on model openness.
What to watch: Whether Nemotron 4 actually closes the gap with Chinese open models once it ships, and whether NeMo Switchyard's cost claims hold up under independent agent benchmarks rather than Nvidia's own reporting.
xAI Launches Grok Bot, Betting on "Always-On" Agents With Their Own Cloud Computers
What happened: xAI, working with Cursor as the two companies move toward merging, launched Grok Bot in beta across Mac, iOS, Windows, and Linux — persistent AI agents that run on dedicated cloud virtual machines, sign into a user's own tools, and complete multi-step work autonomously. Access starts at $120 a seat and is bundled free into SuperGrok Heavy and Cursor Ultra subscriptions.
Why it matters: Early tester reaction has been unusually strong for a launch-day product, with several describing it as the first agent interface that feels genuinely usable by non-technical staff rather than developers. That matters because consumer-grade usability, not raw capability, has been the actual bottleneck slowing agent adoption in most workplaces.
What to watch: Whether the product's origins inside Cursor's team (evident in its iOS publisher listing and hosting) become a bigger story as the xAI-Cursor merger progresses, and how quickly competitors respond with comparably polished always-on agent products.
Cybersecurity
Suspected Chinese Hackers Used Open-Source AI Agents to Autonomously Breach Taiwanese Government Sites
What happened: Researchers reported that suspected Chinese state-linked hackers built an autonomous hacking tool from open-source AI agents that ran simultaneous reconnaissance and intrusion against Taiwanese government websites in July, compromising multiple targets. Separate CNBC reporting the same day describes a broader rush among enterprises to increase cybersecurity spending specifically in response to AI agents' hacking capability.
Why it matters: Autonomous, AI-driven attacks running reconnaissance and exploitation in parallel represent a genuinely new operational tempo for state-linked cyberwarfare — not just a faster version of existing tactics, but a shift in how many targets a single operation can pursue simultaneously.
What to watch: Whether attribution firms confirm state sponsorship beyond the current "suspected" framing, and whether similar open-source-agent-based tooling shows up in attacks against non-government targets in the coming weeks.
Researchers Extract Hidden "Reasoning Traces" From OpenAI, Anthropic, and Google Models Through a Shared API Flaw
What happened: A research team published a 116-page paper showing they could extract encrypted internal reasoning traces from frontier models by feeding them to weaker models from the same provider and having those models decrypt and output the content in plaintext. The technique worked across OpenAI, Anthropic, and Google APIs, and the researchers say it also surfaced evidence relevant to ongoing questions about whether competing open-weight models were trained on distilled reasoning from these providers' outputs. All three companies have reportedly blocked the specific cross-model attack since disclosure.
Why it matters: Beyond the immediate security fix, this cuts against a narrative several labs and policymakers have pushed about distillation requiring deliberate large-scale effort — the researchers argue the underlying vulnerability was simple enough that, as one commentator put it, "every lab" may have had reason to suspect it was exploitable.
What to watch: Whether this affects ongoing policy debates around AI distillation and export controls, given at least one wire report ties the findings to scrutiny from U.S. officials.
Semiconductors & Markets
Foxconn's Cloud and Networking Business Passes 50% of Revenue for the First Time on AI Server Demand
What happened: Foxconn (Hon Hai) reported Q2 revenue up 41% year-over-year to roughly $78.4 billion and net profit up 35% to about $1.86 billion, with cloud and networking products — largely AI servers — contributing more than half of revenue for the first time ever, overtaking its traditional consumer electronics and smartphone assembly business.
Why it matters: The company that built its identity assembling iPhones now generates more revenue from AI infrastructure hardware than from consumer devices — a concrete milestone in how thoroughly AI server demand has reshaped even the most consumer-electronics-identified names in the supply chain.
What to watch: Whether this cloud/networking share keeps climbing next quarter or was boosted by a temporary surge tied to specific customer ramp schedules.
Tencent's AI Capex Jumps 176% Even as Revenue Beats Estimates
What happened: Tencent reported Q2 revenue up 11% year-over-year to roughly $30.4 billion, above the roughly $30 billion estimate, driven by WeChat advertising growth and resilient gaming revenue. Capital expenditure tied to its AI push rose 176% year-over-year, and profit growth came in below revenue growth as a result.
Why it matters: Tencent beating on revenue while still showing profit compression from AI capex is the same pattern showing up across the industry this earnings cycle — strong top-line demand isn't yet translating cleanly to margin expansion anywhere AI infrastructure spending is ramping this fast.
What to watch: Whether Tencent's next quarter shows capex growth starting to moderate, or whether it continues accelerating alongside its AI-enhanced advertising gains.
Startups & Funding
AI Coding Startups Cognition and Lovable Both Chase Valuations Above $13 Billion Within Days of Each Other
What happened: Cognition, maker of the Devin AI coding agent, is in early talks to raise $1 billion or more at a valuation above $40 billion — up from $26 billion just three months ago in May. Separately, Stockholm-based Lovable closed a $400 million round at a $13.3 billion valuation, roughly double its $6.6 billion valuation from December, with backers including Menlo Ventures and EQT's Scaleup Europe Fund; the company says it has nearly tripled ARR since its last round and now sees 900 million monthly visits to apps built on its platform.
Why it matters: Two AI coding platforms raising at a combined $50 billion-plus in valuation within the same news cycle, with one more than doubling in eight months, shows investor appetite for AI-native developer tools hasn't cooled even as broader "AI bubble" concerns circulate elsewhere in the market.
What to watch: Whether Cognition's round actually closes at the reported target — early talks frequently shift before signing — and whether Lovable's revenue multiples hold up to scrutiny as a larger, more diverse set of investors examines the deal.
AI Code-Testing Startup Blacksmith Raises at a 10x Valuation Jump
What happened: Blacksmith raised a $45 million Series B led by Peak XV Partners at a $550 million valuation — up roughly tenfold from the $60 million valuation it held after a $10 million Series A in 2025.
Why it matters: As AI coding agents make writing code dramatically faster, testing and validating that code is emerging as the next bottleneck worth funding — a sign investors are starting to price the second-order problems AI coding tools create, not just the tools themselves.
What to watch: Whether other code-validation and testing startups see similar valuation jumps as this thesis spreads across the venture market.
THE CODEW TAKE
The most important trend emerging from today's news is that AI's growth story keeps getting confirmed from unexpected angles rather than the usual hyperscaler earnings calls: a neocloud's sold-out backlog, a contract-manufacturer's revenue mix flipping toward servers, and a state-linked cyberattack demonstrating AI agents' offensive capability in the wild. None of today's biggest stories were a new chatbot or a benchmark score — they were infrastructure, security, and capital all independently pointing to the same conclusion, that AI's material impact is now showing up in balance sheets and attack logs, not just product demos. For business leaders, that's the signal worth tracking more closely than any single model release: watch where AI shows up in numbers you didn't expect it to touch yet.
Sources
- Techmeme — story discovery and aggregation
- CNBC — CoreWeave surges 18% in premarket after a "cleaner quarter"
- Reuters — CoreWeave edges past quarterly revenue estimates
- The Stack — Life of a neocloud: CoreWeave doubles both revenue and net loss in Q2, as debt hits $35B
- NVIDIA — Nemotron 3.5 Lightning and NeMo Switchyard announcement
- The Information — Nvidia is developing a Nemotron 4 model with 1T+ parameters
- xAI — Grok Bot: A new kind of colleague
- Bloomberg — SpaceXAI Unveils Grok Bot to Work Like a Team of AI Agents
- Financial Times — Suspected Chinese hackers used AI agents to compromise Taiwanese government websites
- CNBC — AI agents' "alarming" hacking skills creates rush to spend on cybersecurity
- arXiv — Stealing Reasoning Traces from Proprietary LLM APIs
- Wired — A New Trick Reveals AI Models' Inner Thoughts
- Wall Street Journal — Foxconn's AI Hardware Sales Drive Profit Beat
- Bloomberg — Tencent Sales Top Estimates on WeChat Ad Surge, Resilient Games
- South China Morning Post — Tencent capex jumps 176% on AI push as revenue beats estimates
- Bloomberg — AI Startup Cognition in New Funding Talks at $40 Billion Value
- Wall Street Journal — Vibe-Coding Startup Lovable Hits $13 Billion Valuation
- TechCrunch — Blacksmith's Valuation Jumps 10x to $550M as AI Coding Fuels Software Validation
Reviewed by Erwin Castro
on
Thursday, August 13, 2026
Rating: