Home → archive → November 2026

The 24-hour window as it closed at midnight, with the final briefing of the day.

The Brief


final cycle 00:00 · window closed 00:00

Ukrainian deep-strike drones gut more than a third of Russia's refining capacity in thirteen days while SpaceX recovers a Starship upper stage from orbit, a Hoya EUV mask-blank shortage throttles Samsung's leading-edge roadmap, and the bond market's 5.15% ten-year yield strips the AI trade of its equity cushion.

Sort
Group

01Tech in AsiaSemiconductors

Nvidia fallout raises scrutiny of Southeast Asia’s AI sector

Southeast Asia's chip and data center industries face heightened scrutiny following reports linking a suspected Chinese spy to Singapore-based cloud firm Megaspeed, through which Chinese customers reportedly accessed Nvidia chips. US lawmakers are pushing tighter curbs on China's AI chip access, with current export rules not clearly covering remote access via overseas cloud and data center facilities. Possible new export rules and licensing requirements could target Singapore, Malaysia, Thailand and Indonesia, while Malaysia said on November 12 it is considering a law to monitor data center operations and high-performance chip use.

The takeThe disclosure that 22% of Nvidia's chip sales were booked to Singapore entities while only 1% physically arrived there exposes a remote-access loophole in export controls, and Malaysia's November 12 move toward a data center and high-performance chip monitoring law signals the region is bracing for new licensing regimes across Singapore, Malaysia, Thailand and Indonesia.

02The InformationSemiconductors

Everything That Led to Nvidia CEO Jensen Huang’s Fall

A curated timeline traces how Alice Huang, wife of private-equity investor Kang Wei, became managing director of Singapore-based Megaspeed and built a network that funneled Nvidia hardware to China, including banned A800/H800 chips acquired after the October 2023 export restrictions. U.S. intelligence concluded Alice was a Chinese intelligence asset targeting Jensen, and after the relationship surfaced publicly in late 2026, Nvidia's board launched an investigation led by Wachtell Lipton; a second alleged affair with a direct report triggered Jensen's resignation on Nov. 11. Nvidia named CFO Colette Kress interim CEO and Harvey Jones board chair, while the Senate banking committee summoned Huang and Commerce Secretary Howard Lutnick to testify over export-ban supervision.

The takeThe resignation of Jensen Huang as Nvidia CEO is a seismic event for the AI-chip industry: a Chinese intelligence operation allegedly run through Megaspeed and Alice Huang moved at least $7.5 billion of Nvidia chips—about 240,000 of them banned—into China, exposing a catastrophic failure of export-control enforcement and corporate governance at the world's most important AI chipmaker.

03Financial TimesSemiconductors

Nvidia’s cult of personality was bound to end badly

Column argues Nvidia's collapse from its peak, with Huang's resignation and nearly $2tn of value erased, followed the classic pattern of imperial-CEO hubris seen at Enron, Tyco, Fannie Mae and Valeant — hyperbole (a claimed inevitable $10tn valuation, the '$1 in, $100 back' circular-financing boast), contempt for partners like Nscale, and a cult of personality ('Jensanity', a $1mn leather jacket). It notes Nvidia's aggressive denial of reports that Megaspeed, linked to alleged Chinese spy Alice Huang, diverted Nvidia-powered equipment to China, despite FT reporting that multiple US agencies suspect the diversion. The core warning: if Nvidia knowingly sold chips to China in defiance of US law, the scandal becomes far bigger than the earlier corporate blowups.

The takeThe resignation of Jensen Huang and a ~$2tn market-cap wipeout at Nvidia is a defining event for the AI-compute complex, and the piece ties it to concrete allegations — suspected Megaspeed chip diversion to China that multiple US agencies are investigating — making it a governance and export-control story, not just a personality profile.

04ReutersGeopolitics & Defense

SpaceX Starship returns to landing pad from orbit, acing test for reusable rockets

SpaceX's Starship upper stage returned from orbit to its Texas launch tower on Saturday, a milestone toward making the vehicle suitable for human spaceflight. The 15th test flight was intended as a historic double-catch with both stages recovered intact, but the lower booster malfunctioned, triggered an automated abort and exploded over the ocean.

The takeStarship's upper stage survived orbital reentry and executed a return-to-tower landing on its 15th test flight, a concrete step toward full reusability — though the booster's malfunction and automated abort over the ocean shows the double-catch goal remains unmet.

05BloombergInfrastructure

Duffy Eyes Possible L3Harris ‘Incompetence’ in FAA Meltdown

Transportation Secretary Sean Duffy said officials are investigating whether L3Harris was at fault after a router at a remote FAA site flooded other routers with over 1,000 connection attempts every 4-5 seconds during a copper-to-fiber replacement, disrupting aircraft communications and radar feeds and grounding flights for more than four hours. Officials ruled out a cyberattack and said there was no evidence AI played a role; L3Harris says it has completed 70% of the fiber conversion and is cooperating. Duffy is seeking $50 billion in emergency funding to modernize the aging air traffic control system, with the request due Monday.

The takeA router flooding the FAA network with 1,000+ connection attempts every 4-5 seconds during L3Harris's copper-to-fiber conversion triggered the longest nationwide ground stop since 9/11, grounding ~12,000 flights Friday and forcing 4,000-5,000 cancellations Saturday — a concrete critical-infrastructure failure that exposes the fragility of the aging air traffic control network and the contractor dependency behind it.

06Financial TimesMarkets

China’s beaten-down tech stocks could offer AI with a little less drama

The FT argues that while US-China tech decoupling fears hammered Hong Kong-listed Chinese LLM developers Z.ai and MiniMax, the sell-off may create opportunity in domestically oriented AI supply-chain names. It cites Goldman Sachs estimates that two-fifths of China's $3.2tn AI universe sells mainly at home, and that China holds over half of global PCB production — components that could become as scarce as memory chips. It also flags China's $250bn of planned AI spending next year as barely a tenth of the rest of the world's, implying either efficiency or a coming spending catch-up.

The takeThe piece surfaces concrete, thesis-relevant data: China's $250bn planned AI capex is barely a tenth of the rest of the world's, the Star 50 is down 40% from its summer peak, and two-fifths of a $3.2tn Chinese AI universe is domestically focused — framing a contrarian case that China's AI value sits in supply-chain 'picks and shovels' like PCB materials and networking rather than frontier labs.

07ReutersGeopolitics & Defense

What is APEC and who will be there?

The White House confirmed Trump, Xi and Putin will hold their first three-way meeting at APEC in Shenzhen on November 18-19, with talks opening on nuclear proliferation — specifically a replacement for the lapsed New START, which Trump wants to expand to include China, though officials say they expect no agreement. Also on the agenda: the deadlocked US-Israeli war on Iran, with the Strait of Hormuz still closed after 100+ days and a third of China's pre-war crude supply disrupted, plus US-China trade and the chip/rare-earth rivalry now roiled by the Nvidia espionage allegation. Ukraine is unlikely to be raised, and analysts frame the summit as performative statecraft rather than conflict resolution.

The takeA first-ever Trump-Xi-Putin trilateral on the APEC sidelines puts nuclear arms control, the closed Strait of Hormuz and the US-China chip/rare-earth fight on one table — with Washington openly saying it expects no New START agreement, making the summit more a power-projection exercise than a dealmaking venue.

08ReutersMarkets

US Democratic senator wants assessment of whether Nvidia is 'too big to fail'

Senator Jack Reed, the top Democrat on the Senate Armed Services Committee, wants Congress to assess whether Nvidia is 'too big to fail' and how its potential failure would ripple through US banking and credit markets, citing the company's backstopping of the AI ecosystem via complex financing deals and Americans' retirement/pension exposure. The call comes as global AI-linked stocks fell following the departure of CEO Jensen Huang amid a personal crisis. Reed is demanding full transparency into Nvidia's financing transactions and an analysis of systemic exposure.

The takeA sitting senator calling for a systemic-risk ('too big to fail') review of Nvidia's role backstopping the AI ecosystem through complex financing transactions is a genuine macro-financial and tech-statecraft event, signaling that AI compute concentration is now being treated as a banking/credit-market stability question.

09ReutersGeopolitics & Defense

SpaceX's Starship nears end of 15th test flight and return-landing attempt

Starship's upper stage returned from a 22-hour orbital mission on Nov 14 and was caught by the launch tower's mechanical arms at Starbase, a first for the vehicle and a milestone toward a fully reusable rocket system. The flight orbited Earth 14 times and deployed 26 Starlink satellites, but the Super Heavy booster failed to return to the pad about 10 minutes after the Nov 13 lift-off, so SpaceX has yet to recover both stages in one mission. SpaceX has spent over $15 billion on Starship development, with NASA contributing $3.2 billion toward a $4 billion-plus Artemis lunar lander contract; heat-shield reuse and in-orbit fuel transfer remain unresolved.

The takeSpaceX caught Starship's upper stage with the launch tower's 'Mechazilla' arms for the first time after a 22-hour, 14-orbit flight that deployed 26 Starlink satellites — the first recovery of an orbital-class second stage, the piece that Falcon 9 still throws away. The Super Heavy booster missed its own pad landing, so full-stack reuse remains unproven, and heat-shield tile reuse and in-orbit propellant transfer are still open before the 2028 NASA Artemis crewed landing.

10BloombergCybersecurity

Russia Struggles to Restore Key Statistics Two Weeks After Hack

Two weeks after targeted destructive cyberattacks (Oct. 29–Nov. 1), Rosstat still cannot publish or collect key economic data: GDP Excel files return 404 errors, October CPI was pushed from Nov. 13 to Nov. 25, and weekly inflation data has slipped twice, while companies got a data-submission extension to Nov. 20. The Bank of Russia and finance/economy ministries depend on these figures for rate decisions and budget drafting. Cyber Partisans and Silent Crow claim to have exfiltrated ~23TB, including records indicating roughly 400,000 Russian military deaths in Ukraine.

The takeA destructive cyberattack on Rosstat has left Russia's statistical apparatus degraded for two weeks, delaying CPI and inflation releases and forcing deadline extensions — a real-world demonstration that cyber operations against state data infrastructure can disrupt central-bank rate-setting and budget planning, not just leak data.

11BloombergGeopolitics & Defense

SpaceX Catches Starship in Novel Landing at Texas Site

Starship returned from orbit and was caught midair by the launch tower's mechanical arms at Starbase in South Texas, a novel engineering feat toward a fully reusable orbital vehicle. The flight launched Friday, deployed a batch of Starlink satellites, and spent about a day in orbit before reentry; SpaceX failed to catch the Super Heavy booster this time, which Musk blamed on engine underperformance. SpaceX says it takes four weeks to build a Super Heavy and two weeks to build a Starship, and plans to use the vehicle for larger Starlink satellites, AI datacenter spacecraft, and eventually moon and Mars missions.

The takeSpaceX caught the Starship upper stage with the launch tower's 'Mechazilla' arms after a roughly day-long orbital flight that deployed Starlink satellites — the first time any operator has recovered an orbital-class upper stage, a key proof of concept for full reusability. Musk attributed the missed Super Heavy booster catch to engine underperformance but said it doesn't change the timeline to high-volume reusable flights.

12BloombergInfrastructure

FAA Grounds Flights Across US Until Further Notice

The FAA halted all US departing flights until further notice after a communications/telecommunications outage hit every air route traffic control center, the first such nationwide stop since 2001 and 2023. Officials said no threat or cyberattack is suspected, with one line of inquiry focused on a software update, while Transportation Secretary Duffy and Senator Mullin declined to rule out an AI link — prompting Air Space Intelligence, whose AI SMART system just went nationwide across all 22 centers, to publicly deny responsibility. Airlines diverted over 800 flights and canceled most Friday evening departures, stranding passengers ahead of Thanksgiving as the FAA seeks additional funding beyond its $12.5B overhaul.

The takeThe third nationwide US ground stop in history, triggered by a telecom/software failure across all 22 air route traffic control centers, exposes the fragility of critical aviation infrastructure and directly implicates the FAA's freshly nationwide AI-driven SMART airspace system, which contractor ASI rushed to disclaim. With 827 flights diverted, 11,000 flights and 1.5M passengers affected, and a $12.5B modernization program still underfunded, this is a high-signal critical-infrastructure and AI-in-the-loop failure story.

13SiliconANGLEInfrastructure

FAA grounds domestic flights after reported network outage

The FAA's Air Traffic Control System Command Center issued a nationwide ground stop affecting all 22 regional centers after a communication issue impacting all Air Route Traffic Control Centers, with CBS reporting a private communications network crashed following an update. The shutdown, starting 3:54 p.m. ET, has delayed or canceled over 13,000 flights and led Eurocontrol to suspend U.S.-bound flights at the FAA's request. The agency last imposed a national ground stop in January 2023 over a NOTAM system failure that lasted about two hours and grounded 9,500 flights.

The takeA nationwide FAA ground stop triggered by a communications-network failure—reportedly an update that caused a reboot and crash—halted all U.S. domestic flights, delaying or canceling 13,000+ flights and prompting Eurocontrol to suspend U.S.-bound traffic, exposing critical-infrastructure fragility in air traffic control.

14Wall Street JournalInfrastructure

CoreWeave Balances Capital Needs Against Expanding Data Center Footprint

CoreWeave beat on operating metrics — revenue $3.66B (+168% YoY), adjusted EBITDA $2.26B at 62% margin, guidance raised to $13.0–13.3B — but debt jumped $11.8B in one quarter to $46.9B, with Q4 interest guided at $1.05–1.15B, roughly 1.4x implied adjusted operating income. Capex hit $14.5B in the quarter (full-year guidance $37B) against ~$2B operating cash flow, leaving a ~$7.5B quarterly cash deficit, while only ~37% of contracted power (1.7GW of 4.6GW) is operational. The company is leaning on ~$9B of expected customer prepayments to bridge funding, with Street estimates of incremental financing needs ranging from $22.9B to $76B.

The takeCoreWeave's Q3 shows the AI buildout's defining tension: revenue up 168% to $3.66B and a $132.5B backlog against $46.9B debt, ~$1.1B quarterly interest and only 1.7GW of 4.6GW contracted capacity live — with 2031 bonds yielding >15%, the bond market, not demand, may set the pace of compute expansion.

15CNBC TechnologyGeopolitics & Defense

SpaceX catches upper stage of Starship rocket in historic first

On Test Flight 15, SpaceX's Starship upper stage completed 14 orbits and was caught by the launch tower's mechanical arms at Starbase, Texas — the first time the spacecraft portion was recovered from orbit, though the booster was diverted to a Gulf splashdown and exploded. The flight also deployed 26 Starlink v3 satellites, part of a constellation already over 11,000 strong. Full reusability is central to SpaceX's Starlink expansion, NASA lunar cargo and crew transport, and Musk's Mars ambitions.

The takeSpaceX caught Starship's upper stage from orbit with its tower 'chopsticks' for the first time, a genuine reusability milestone that underpins Starlink v3 deployment (up to 60 satellites per flight against a constellation already exceeding 11,000) and NASA's lunar lander obligations — even as the booster was lost to an offshore explosion.

16BloombergInfrastructure

The Nuclear Investment Boom in Eastern Europe Picks Up Pace

Russia's invasion of Ukraine has triggered a nuclear investment wave across Eastern Europe, with the Czechs, Slovaks and Bulgarians signing Westinghouse fuel deals, Poland selecting Westinghouse for its first plant, and Bulgaria building two AP1000 units, while only the Czech Republic (KHNP), Romania (NuScale/Candu) and Hungary (Rosatom) went elsewhere. Poland's $50 billion, three-reactor project still lacks a final EPC contract four years after selection, with negotiations targeted for early 2027 and first power in 2036. Supply-chain strain, skilled-labor shortages and chronic cost overruns threaten the timelines, and Russian entities still supplied over a fifth of EU enriched uranium last year.

The takeNine Eastern European nations plan up to 18 reactors at a combined cost of at least €185 billion, with Westinghouse holding the upper hand over EDF and KHNP as the region pivots away from Russian fuel and technology — a concrete reordering of nuclear supply chains and energy geopolitics. The piece also carries hard risk data: Oxford researchers find nuclear megaprojects run 117% over budget on average, 65% over schedule, with 40% abandoned, and the EU needs 180,000–250,000 extra nuclear professionals by 2050.

17Institute for the Study of WarGeopolitics & Defense

Iran Update, November 14, 2026

ISW-CTP's November 14 update reports Houthi ballistic missile strikes on Yemeni government staging areas and depots in Lahij and Aden, which failed to halt Southern Giants Brigades advances in Madaribah wa al Arah District, with Saudi close air support hitting Houthi frontline units and reinforcements from Taiz. The Houthis also claimed drone strikes on Riyadh's King Khalid International Airport and an East West Pipeline pumping station, which Saudi Arabia intercepted. Separately, Iranian naval forces claimed a gas carrier and multiple tankers struck mines on the US-escorted southern Strait of Hormuz route, an information operation to discredit US escort operations and push vessels into Iranian waters.

The takeIran's IRGC and Artesh navies are running a coordinated information operation claiming commercial vessels are hitting mines on the US-escorted southern Hormuz route — despite CENTCOM's August 27 demining and no observed re-mining — to erode confidence in US escorts and funnel traffic into Iranian territorial waters, a low-cost maritime coercion play against the US-backed transit corridor.

18DigiTimes AsiaSemiconductors

MetaX-UBTech robot chip venture Xilect raises CNY500 million to take on Nvidia Thor

Xilect, the robot-chip venture formed by GPU designer MetaX and humanoid robot maker UBTech, closed a CNY500 million (~US$75M) angel round co-led by GGV Capital and Matrix Partners China, with MetaX and UBTech each holding 35.01% of the Wuxi-registered entity. The company is building a domestic computing platform for embodied AI benchmarked against Nvidia's Thor, with first silicon tape-out planned for H2 2027 and mass production in 2028. Analysts flag the core challenge as designing a low-power edge architecture from scratch, given robot compute budgets of only a dozen to a few dozen watts.

The takeA CNY500M angel round for a MetaX-UBTech JV targeting Nvidia Thor is a concrete signal that China is building a domestic embodied-AI compute stack from scratch, with tape-out set for H2 2027 and mass production in 2028. The engineering constraint is stark: robot compute budgets of a dozen to a few dozen watts versus ~700W for a datacenter GPU like the H100, meaning success hinges on a purpose-built edge architecture rather than a shrunk GPU.

19DigiTimes AsiaSemiconductors

Foxconn targets automotive chiplets with 5μm fan-out RDL packaging

At Hon Hai Tech Day 2026, Foxconn showcased approximately 5μm-class fine-line redistribution layer (RDL) circuitry as its entry vehicle into the advanced packaging market. The company is targeting automotive chiplets, edge AI and consumer electronics, framing the technology around the trade-off between interconnect density, package size and manufacturing cost as AI workloads migrate from cloud training to end devices.

The takeFoxconn is staking a claim in fan-out RDL advanced packaging with ~5μm-class fine-line circuitry, positioning itself to compete in the automotive chiplet and edge-AI packaging space where interconnect density, package size and cost must be balanced — a direct move into territory dominated by TSMC's InFO/CoWoS and OSAT players like ASE.

20DigiTimes AsiaAI & Models

Synopsys sees AI accelerating EDA as chip-specific models divide work

Synopsys CPO Shankar Krishnamoorthy said AI will accelerate EDA as the company builds semiconductor-specific models with OpenAI while keeping the platform neutral across frontier and open-weight models. The OpenAI partnership aims to train a model that operates Synopsys' EDA tools, letting engineers explore far more design iterations and compress weeks-to-months from chip development. Commercially, a >$1B Amazon deal licenses Synopsys IP across multiple Graviton, Trainium and Nitro generations with customization fees and per-chip royalties, offsetting Intel Foundry weakness and underpinning upgraded FY2030 targets.

The takeSynopsys is converting AI from an existential threat to EDA into a growth vector: a >$1B Amazon IP license spanning Graviton/Trainium/Nitro generations under a new royalty-bearing AoIP model, plus a GPT-Synopsys model co-developed with OpenAI to autonomously drive EDA tooling. Guidance was raised to mid-teens revenue CAGR and a 50% adjusted operating margin by FY2030 (from 41.5% in FY2026), with $1B AoIP revenue targeted on current contracts alone.

21DigiTimes AsiaInfrastructure

Lite-On sees AI server orders through 2028 as 800V DC reshapes power market

Lite-On expects AI-related revenue to reach 40–50% of total revenue by 2027 on strong demand for high-end AI power systems, with US cloud customers extending order visibility through 2028. However, major revisions to 800VDC specifications will delay volume production of those power supplies until late 2027, shifting the ramp timeline for next-generation datacenter power delivery.

The takeLite-On's disclosure that AI power systems will hit 40–50% of total revenue by 2027 with order visibility through 2028 is a concrete datapoint on hyperscaler power-infrastructure buildout, and the 800VDC spec revisions pushing volume production to late 2027 signal a real timing risk for next-gen rack power architectures.

22Wall Street JournalMarkets

The 31-Year-Old Trader Who Called Out Jensen Huang’s ‘China Problem’—and Shorted Nvidia

Culper Research's Christian Lamarco published a May report shorting Nvidia on the thesis that the company has a 'significant China problem,' flagging an 'uncomfortably close' relationship between Jensen Huang and Alice Huang, an executive at a Southeast Asian business partner. The report was initially ignored — Nvidia hit a record high the next day — but six months later, as questions about Alice Huang and suspected Chinese intelligence ties became public, the analysis went viral and Nvidia lost more than $1 trillion in market value, with the stock down 20% since publication. Lamarco, 31, built the case using paid expert interviews with former Nvidia employees, scrubbed code, Singaporean and Malaysian corporate filings, and open-source footage.

The takeA short seller's 40-page China-exposure thesis on Nvidia preceded revelations that CEO Jensen Huang was allegedly involved with a woman suspected of Chinese intelligence ties, and Nvidia has since shed over $1 trillion in market value with shares down 20% — a material repricing of the world's most valuable AI chipmaker driven by export-control and national-security risk.

23ServeTheHomeSemiconductors

AMD Launches EPYC Embedded 9006: Venice Gets an Early Start in Embedded

ServeTheHome details AMD's EPYC Embedded 9006 family, which adopts the SP7 platform wholesale and ships alongside the commercial Venice launch — a reversal from prior generations where embedded trailed. The lineup spans 64 to 256 cores (Embedded 9976 at 256C/512T, 1024MB L3, 600W), moves to 2nm with 32-core CCDs and 128MB per complex, and adds PCIe Gen6 (though lane count drops from 128 to 96), with 7+ year availability and SP7 reuse planned for the next-gen 'Florence' chips. AMD is targeting ATE, high-end firewalls and enterprise storage controllers, framing demand as AI-adjacent, with the 128-core 9756 shipping this month and five more SKUs in late December.

The takeAMD's EPYC Embedded 9006 (Venice, Zen 6c, 2nm) launches simultaneously with commercial EPYC for the first time, scaling to 256 cores/512 threads, 16-channel DDR5-8000/MRDIMM-12800, 96 lanes PCIe Gen6/CXL 3.1 and 8TB memory at up to 600W — a signal that embedded/long-lifecycle infrastructure silicon is now a lead product for AMD, not an afterthought.

24Yahoo Tech & Markets WireMarkets

Anthropic IPO to test Wall Street’s faith in AI as skepticism mounts

Anthropic is marketing a potentially record-breaking IPO at a $1.3–1.4 trillion valuation, below the $1.5–2 trillion investors had floated weeks earlier, with trading expected to begin Nov. 19. The offering lands in a turbulent market — surging Treasury yields, a Microsoft Outlook outage, and the sudden ousting of Nvidia CEO Jensen Huang — testing Wall Street's faith in the AI trade. CEO Dario Amodei struck a measured 'underpromise and overdeliver' tone on timing, while Goldman Sachs and Morgan Stanley lead a substantial institutional marketing push.

The takeAnthropic's IPO, targeting a $1.3–1.4T valuation (down from $1.5–2T contemplated weeks earlier) and expected to trade Nov. 19, is a genuine market event — the clearest test yet of AI appetite amid rising Treasury yields and the shock ouster of Nvidia CEO Jensen Huang, Anthropic's key investor and chip supplier.

25BloombergInfrastructure

Tech Giants Face More Grid Cost Pressure on Democratic Sweep

Anger over rising consumer electricity bills swept Democrats into office from Ohio to Texas on pledges to shield households from datacenter power demand, intensifying pressure on tech giants to guarantee grid infrastructure costs. Datacenter operator Applied Digital is extracting billions in parent guarantees from tenants like Oracle and Meta and even chip suppliers Nvidia and AMD, while capping junk-rated neocloud leasing; NiSource's 'GenCo' structure for Amazon and Google isolates tech loads from consumer rates. Meanwhile Wisconsin regulators' A-/A3 credit threshold forced Oracle to post a letter of credit exceeding $7B (over $100M annually) before it dropped its lawsuit, against a backdrop of $300-320B in expected hyperscaler debt issuance this year.

The takeThe political economy of AI datacenter power is shifting: post-midterm Democrats are forcing hyperscalers to backstop grid costs, with Applied Digital demanding parent guarantees from Oracle, Meta, Nvidia and AMD, and Oracle facing a >$7B letter-of-credit demand in Wisconsin. This reprices who can build AI capacity and how it gets financed.

26Hacker NewsInfrastructure

Will We Put Data Centers in Space?

The report models 1 GW of compute across terrestrial and orbital scenarios, concluding space solar in dawn-dusk sun-synchronous orbit yields 3-5x terrestrial panel output and that ODCs become cost-competitive within 3-5 years if Starship stays on track, though unlikely to hold meaningful compute share before 2030. It flags that inter-satellite bandwidth limits ODCs to inference workloads early on, that chip production may become the binding constraint rather than launch, and that SpaceX's dual role as sole viable launch provider and xAI owner raises concentration-of-power and AI-governance concerns around jurisdictional ambiguity for orbital compute.

The takeA rigorous technical-economic analysis finding orbital data centers reach cost parity with off-grid terrestrial power at ~$250/kg launch cost and beat all terrestrial energy at ~$50/kg (vs ~$1,500/kg today), contingent almost entirely on Starship achieving full reusability — with radiative cooling a manageable 2-5% of cost and chip-failure overbuy modeled at ~38-40%.

27DigiTimes AsiaInfrastructure

Why a North Dakota town backed AI data center development

In the November 3, 2026 midterms, voters weighed 47 local AI data center measures; of 45 counted, 44 backed restrictions or opposition, including five full bans (CA, FL, OH), twelve partial bans (KS, ME, OH), and three six-month moratoriums in Maine. Harwood, North Dakota was the sole rejection of tighter review, where 76% of 430 voters opposed new scrutiny for large developments — preserving a 320-acre Applied Digital hyperscale site rezoned in September 2025 amid local concerns over noise, water, air quality, power bills and traffic.

The takeThe 44-of-45 rejection rate for data center restrictions across 10 states is a hard signal that local permitting risk is becoming a systemic constraint on US AI compute buildout — five outright bans plus twelve partial bans and three moratoriums directly threaten siting pipelines, making Harwood's 76% outlier vote (325-105) the exception that proves the rule.

28DigiTimes AsiaSemiconductors

Hoya reportedly unable to meet Samsung's full 2027 EUV mask blank request

Samsung asked Hoya in September for additional 2027 EUV mask blanks after internal simulations showed a sharp consumption increase, but Hoya said it cannot fully accommodate the request due to surging demand from other chipmakers. Hoya holds 70-75% of the global EUV mask blank market (AGC 20-25%), and quality/spec issues have pushed Hoya above 90% of Samsung's procurement; Hoya's new Singapore plant (~JPY42bn) won't produce until fiscal 2028. Samsung is qualifying South Korea's S&S Tech as a second source, with particle contamination still to be resolved before first domestic use.

The takeHoya's 70-75% grip on EUV mask blanks is becoming a hard constraint on Samsung's roadmap: Samsung will burn ~5,500 blanks this year growing ~25% annually, yet mask counts per node are exploding (1b DRAM 4-5 → 1c 6-7 → 1d ~10; foundry 4nm 15-16 → 2nm 25-27) and PSM blanks take 1.5-2x longer to make, while Hoya's Singapore fab only ramps in FY2028. This is a genuine upstream bottleneck on EUV lithography economics, not just a procurement hiccup.

29Hacker NewsCybersecurity

Lovable apps expose passport details and more than 200K private messages

Security researchers, working with Handelsblatt, analyzed 159,041 Lovable projects from Common Crawl (Dec 2025–Sep 2026); 34,545 used Supabase, and 24,664 served database records on page load without login. Additional read-only requests surfaced email addresses in 10,560 tables across 7,403 sites (30% of those examined), plus exposed passport numbers, ticket and payment data, contracts, invoices, and 280,000 customer emails. The authors stress the 30% figure is not a confirmed leak rate and their sample is not representative, but argue Lovable's recurring integration patterns let attackers reuse one misconfiguration approach across many apps built by non-technical users.

The takeA systematic scan of 159,041 Lovable-built projects found 24,664 loading Supabase records without authentication and 7,403 sites exposing email-bearing tables, including a case leaking 1,000+ passport-number records and a forum exposing 200,000+ chat messages — evidence that AI app generators reproduce identical access-control misconfigurations at scale, making one exploit template reusable across thousands of apps.

30Wall Street JournalSemiconductors

How Donald Trump and Jensen Huang Forged Their Powerful Alliance

WSJ details how Jensen Huang privately urged Trump to act against a patchwork of state AI regulations, leading to a Truth Social post and executive order, and describes Huang's outsized role in shaping the president's AI strategy, global chip deals and state visits. The piece reports fresh scrutiny after Trump stood by Huang despite affairs, with lawmakers probing what the administration knew about Huang's business partner Alice Huang—suspected of working for Chinese intelligence—and the failure to stop AI chips going to China.

The takeThe WSJ reports that Nvidia CEO Jensen Huang directly pushed Trump in the Oval Office to preempt state AI rules, prompting an executive order—evidence that a single chip CEO shaped federal AI policy and chip-export decisions, now under congressional scrutiny over suspected Chinese intelligence ties and AI chips reaching China.

31Tom's HardwareGeopolitics & Defense

U.S. Treasury Secretary questions whether Nvidia knowingly sold AI GPUs to shell companies, says company itself is responsible — ‘I think it is important that anyone at the company there be held accountable for that’

In a CNBC interview, Treasury Secretary Scott Bessent said U.S. export controls suffer 'a lot of leakage' and reframed the Nvidia story around whether the company knew its clients and sold to 'obviously pass-through shell companies,' calling for accountability for anyone involved while saying it is unclear whether CEO Jensen Huang knew. He criticized Nvidia general counsel Timothy Teter for contacting Commerce about Megaspeed's Alice Wong detention, and disclosed that Treasury, War, State and USTR backed adding Megaspeed to the Entity List for months, but BIS reportedly stalled the blacklisting as Commerce Secretary Howard Lutnick feared worsening trade tensions with China.

The takeTreasury Secretary Bessent publicly questions whether Nvidia knowingly sold AI GPUs to pass-through shell companies and calls for individual accountability, while revealing interagency support for blacklisting Megaspeed that BIS allegedly stalled over China trade-tension fears. This is a material escalation of export-control enforcement pressure on the AI chip supply chain, with direct implications for Nvidia's compliance posture and GPU diversion to China.

32DigiTimes AsiaSemiconductors

Etron plans strategic stake in Powerchip Innovation Investment

Memory IC designer Etron Technology posted Q3 2026 revenue of NT$6.792 billion (US$206.6 million), up 38.6% sequentially and 535% year-over-year, on strong memory demand and rising product prices. Net income attributable to the parent reached NT$2.493 billion, or EPS of NT$7.59, reversing a NT$1.74 per-share loss in Q3 2025. The company is also planning a strategic stake in Powerchip Innovation Investment.

The takeEtron's Q3 2026 revenue of NT$6.792 billion (US$206.6M) jumped 38.6% QoQ and 535% YoY with EPS of NT$7.59 versus a NT$1.74 loss a year prior, a dramatic swing that quantifies the memory upcycle now lifting even niche specialty DRAM designers — and the strategic stake in Powerchip Innovation Investment signals Etron converting windfall into supply-chain positioning.

33Financial TimesGeopolitics & Defense

US floats Nato membership for Donbas withdrawal deal, Zelenskyy says

Zelenskyy said Kushner and Witkoff asked whether Kyiv would cede the ~20% of Donbas it still holds in exchange for NATO admission, a first, with European capitals backing a 'maximalist' posture ahead of renewed Abu Dhabi trilateral talks. He also accused SpaceX's Pentagon-funded Starshield of imposing 161kph speed limits that hamper strikes and drone interception, allegedly only over occupied territory, which Ukrainian officials read as pressure tied to the negotiations; SpaceX called the claim 'false'. Denmark pledged a €645mn package including F-16 ammunition and €200mn for drone production.

The takeThe US floating a Donbas-withdrawal-for-NATO-membership trade marks a genuine shift in the war's diplomatic architecture, and the Starshield speed-cap dispute (161kph limit, Pentagon-funded Starlink, SpaceX denying any change) shows commercial space infrastructure is now a live lever in peace negotiations — a concrete case of tech statecraft.

34Tom's HardwareSemiconductors

Malaysia plans new law using data center power consumption to catch AI GPU smugglers

Malaysia's trade minister acknowledged companies rented space in Malaysian data centers to legally import GPUs that were then re-exported in violation of U.S. export bans, and the Digital Ministry is reviewing whether authorities can act against such re-exporters. Proposed legislation would require operators to disclose GPU import counts, power consumption, facility locations, and operator/hyperscale customer identities, with lifetime tracking rather than approval-stage oversight, plus mandatory payment for all reserved electricity capacity regardless of use. Malaysia has approved 117 data center projects, 55 operational, and will share facility information with U.S. counterparts on request, though it is unclear whether GPU models, serial numbers, or on-site inspections will be required.

The takeMalaysia's plan to police GPU diversion by cross-referencing disclosed accelerator imports against data center power draw is a novel enforcement mechanism, but the article's own analysis shows it is technically leaky: H100/H200 SXM at 700W and B200 SXM at 1,000W TDP vary widely under real workloads, and without server configs, PUE, or serial-number tracking, power-based inference of GPU counts is imprecise. The accompanying rule forcing operators to pay for reserved-but-unused capacity (e.g., paying for all 100 MW when only 50 MW is drawn) is the sharper lever, raising the cost of phantom capacity used to launder diverted hardware.

35Yahoo Tech & Markets WireMarkets

The bond market just turned on the AI trade — and chip stocks lost more than $650 billion in two days: Chart of the Day

A Yahoo Finance basket of 62 chip stocks lost more than $650 billion in market value from Wednesday's close through Friday, with only three rising, even as the PHLX Semiconductor Index still finished the week up ~8%. Nvidia fell ~13% in the week Jensen Huang resigned as CEO — its worst week since April 2025 and sixth consecutive weekly loss — wiping out ~$570 billion as investors weighed its $530 billion web of AI deals; partners Nebius and CoreWeave also slid. Critically, the bond market stopped cushioning chip selloffs: the 10-year yield rose on all three down days for semis, jumping ~30bp to ~5.15% (biggest weekly rise since April 2025), as hot CPI and PPI data pushed Polymarket's odds of a December rate hike to ~98%.

The takeThe AI trade's shock absorber broke: chip stocks shed over $650B in two days as the 10-year yield jumped ~30bp to 5.15% and rate-hike odds hit 98%, ending the regime where Treasury rallies cushioned semiconductor selloffs. Nvidia's ~13% weekly drop (worst since April 2025, sixth straight loss) erased ~$570B against its $530B AI deal web, showing the AI capex story is now hostage to Fed policy.

36Tom's HardwareAI & Models

Researchers allege DeepSeek, Kimi, and other Chinese AI models were trained using banned Nvidia Blackwell GPUs

Independent researchers at UnemploymentLabs claim that open-weight Chinese models including DeepSeek V4 and Kimi K3.1 were trained on Nvidia Blackwell GPUs, which are barred from export to China. The method replicates weight initialization with matching SM and thread counts and finds the resulting statistical distribution of weights precisely matches the Chinese models, breaking down when SM/thread counts are altered; AMD cards were ruled out for DeepSeek V4. The authors hedge that the work proves correlation only — not where the chips were or that controls were broken, noting renting Blackwell hardware abroad may be legal.

The takeA GPU 'fingerprinting' technique exploiting cuRAND seed determinism and SM/thread-count signatures (148/152 SMs, 2,048 threads per SM) statistically matches the weight-initialization distributions of DeepSeek V4, Kimi K3.1, MiniMax-M3 and Ling-3.1-Flash to banned Blackwell silicon, and even estimates card counts — a novel forensic method for enforcing export controls.

37Tom's HardwareGeopolitics & Defense

Pentagon ‘struggles to account for’ $31 billion in missing F-35 aircraft parts, weeks after shipment was swiped by China

A new GAO report says the Pentagon's F-35 inventory system is 'outdated, inaccurate, and incomplete,' leaving roughly 26 million spare parts worth an estimated $31 billion spread across 700 government and contractor facilities poorly tracked. The watchdog notes 'minimal progress' on recommendations dating to 2019, including unimplemented physical inventory checks and no reliable way to verify contractor data, while over 1 million parts worth $85 million are lost or missing and ~17,000 repair parts worth up to $42.5 billion go untracked. The report lands weeks after a classified F-35 parts shipment routed through Hong Kong was allegedly swiped by China, and as the JPO claims it is building real-time interfaces with prime contractors.

The takeThe GAO's finding that the Pentagon cannot account for ~26 million F-35 spare parts worth $31 billion — with 17,000 repair parts worth up to $42.5 billion untracked and 'minimal progress' since 2019 — exposes a systemic supply-chain accountability failure in the U.S.'s most expensive weapons program, made more acute by the alleged Chinese interception of a diverted parts shipment through Hong Kong.

38DigiTimes AsiaMarkets

Pegatron sees networking recovery in 2027 as consumer electronics growth eases after peak

Pegatron's November 12 earnings call showed Q3 2026 consolidated revenue of NT$389.073B (+41.8% QoQ, +50.9% YoY), gross margin 4.5% and operating margin 2.1%, though net profit fell 12.5% YoY to NT$3.977B on weaker non-operating income. Communications revenue (~NT$225B, 58% of total) rose ~50% YoY on smartphone launches and networking recovery, while information products (~NT$80B) more than doubled on server ramp; servers are guided to ~10% of 2026 revenue and several-fold growth in 2027, with networking recovery results expected in 2027. Heavy capex for North America/Taiwan capacity and a liquid-cooling lab, plus strategic inventory buildup, pushed debt ratio to 70% and operating cash flow to a NT$128.4B outflow, with ECB financing approved as contingency.

The takePegatron's Q3 2026 revenue hit a record NT$389.07B (+50.9% YoY) with operating profit up 149.7% YoY as its first-year server business reaches ~10% of revenue, but the real signal is the balance-sheet strain: inventories up NT$159B and short-term borrowings up NT$151.7B since year-end, driving a NT$128.4B operating cash outflow and debt ratio to 70%. Management guides server revenue to grow several-fold in 2027 across VR/GB/HGX/compute trays/storage and 4-5 self-developed ASIC customers ramping with visibility into 2028, while flagging that liquid-cooling facility costs and new North America/Taiwan capacity are still offsetting gross margin.

39WiredCybersecurity

Beware the New AI-Fueled Ransomware Boom

Wired's Kernel Panic newsletter reports that claimed ransomware incidents rose sharply over the summer, with some estimates putting leak-site postings near 3,000—a 20-40% increase since spring—and Halcyon counting 111 groups making claims between July and September. Researchers from Halcyon, Recorded Future and ReliaQuest say criminals are using LLMs for phishing lures, negotiation and code checking, fueling fragmentation into smaller and lone-wolf operations (Global Secret Group claimed 48 victims, Storm 71 in 20 days), though established groups like Qilin, Akira and the Gentlemen remain largely human-run.

The takeResearchers attribute a 20-40% summer spike in claimed ransomware attacks—nearly 3,000 leak-site postings and 111 active groups in Q3—partly to LLM use in phishing and negotiation, with lone-wolf operators like a teenager netting $800K in under two weeks, even as full AI-driven ransomware has yet to appear.

40Wall Street JournalGeopolitics & Defense

The Massive Chinese Spy Operation Targeting Western Tech and AI

WSJ reports that U.S. officials believe Beijing has mobilized government agencies, companies and citizens in a vast information-gathering operation to obtain Western technology know-how, and that the campaign has allegedly entangled former Nvidia CEO Jensen Huang. Syndicated BBC coverage details the arrest of two Chinese nationals in south-western France accused of operating a parabolic antenna to capture Starlink and military satellite data near defense and aerospace installations, with French authorities increasingly publicizing such cases.

The takeThe WSJ piece alleges China has ensnared now-former Nvidia CEO Jensen Huang in a state-orchestrated technology-theft campaign of 'unparalleled scale' aimed at upgrading its military and economy — a claim with enormous implications for AI-chip export controls and corporate security. The bundled BBC report adds concrete tradecraft: two Chinese nationals charged in France for allegedly intercepting Starlink and military satellite data near Airbus, Thales and Dassault sites, showing espionage targeting Western space and defense technology.

41QbitAI (量子位)AI & Models

Breaking: DeepSeek Opens Full-Suite API, Including Voice and Agentic Web Search

DeepSeek released a full-suite application API giving developers standalone, separately billed access to web search, deep-thinking search, file upload, link reading, OCR, and — for the first time — official ASR and TTS endpoints, running on the V4.1-Flash model across Messages, Responses and Chat Completions interfaces. Unlike the October Search/Fetch release, the model now autonomously decides when to search, what to fetch and how to synthesize results, removing the orchestration code developers previously had to write. DeepSeek also open-sourced a lightweight chat frontend demo supporting import of conversation history, explicitly stating developers can replicate the full DeepSeek App, and framed its own chat product as merely 'a relatively basic form' of the technology.

The takeDeepSeek is unbundling its entire consumer app stack — agentic web search, Fetch, OCR, ASR and TTS — into independently billable APIs, with ASR at ¥0.39/hour (roughly 1/4.6 of OpenAI's GPT-Transcribe and 1/16.5 of Google Cloud STT) and TTS at ¥1.5 per 10k characters. Open-sourcing a chat frontend that can 1:1 replicate the DeepSeek App signals a deliberate strategy shift: cede the application layer to developers and compete purely on model plus infrastructure economics.

42QbitAI (量子位)AI & Models

「Strictest Ever」Wet-Lab Prediction Benchmark Released! Maxing Out Reasoning and 5x Compute Is Still Useless

UniPat AI, with advisors from Stanford, Cornell, Harvard, Peking University and HKUST, released SPU, a benchmark of 100 public tasks from 85 recent papers plus 16 private unpublished wet-lab tasks (VHH nanobodies, electrolytes, Magrolimab variants) where models get 8 hours on an H20 GPU to autonomously predict experimental result rankings, scored by Spearman correlation against held-out measurements. Across 12 frontier models and 13 configurations, no model approaches reliable prediction, and the study documents that extra reasoning effort, more tool calls and more GPU use do not improve accuracy because models fail at selecting the right physical quantity to rank on. The dataset and code are open-sourced.

The takeSPU's 116 tasks and 4,207 wet-lab experiments show frontier models barely clear random chance — best score 0.486 (Claude Opus 5.5), only 34-52% of tasks above the random threshold — and GPT-6 Astra at max reasoning burned 4.8x runtime and 5.3x API cost for essentially zero score gain, a concrete indictment of brute-force test-time compute scaling.

43WiredAI & Models

No Pilots. Few Air Traffic Controllers. What Could Go Wrong?

The article traces how AI is entering aviation faster than the sector's traditionally conservative safety culture can absorb: autonomous drone delivery scaling to hundreds of Amazon and Walmart sites, eVTOL air taxis moving through the FAA's eIPP demonstration phases, and the FAA deploying Air Space Intelligence's SMART predictive tool to manage airspace. It balances industry optimism against warnings from a retired FAA employee, pilot unions, and researchers about neural-network failure modes, hallucinated safety-critical data, and the risk that AI-driven efficiency justifies shrinking the controller workforce.

The takeWired's deep dive documents the concrete milestones of the AI aviation build-out—Joby's 3,200-mile pilotless Cessna flight, the FAA's SMART predictive airspace tool, and the eIPP program—while surfacing the real safety and labor stakes, including the FAA cutting its controller staffing target from 14,633 to 12,563 and lawmakers demanding a GAO review over hallucination fears.

44Wall Street JournalAI & Models

Big Tech Wants Everyone to Use an AI Agent. But Who Shoulders the Cost?

WSJ reports that Meta, OpenAI, SpaceX, Anthropic and Instinct are gating agent access behind paid tiers or limited launches because background agents are far costlier to run than chatbots — Google built a new platform for Gemini agents and delayed Gemini Spark over engineering issues. Concrete figures: Deutsche Bank estimates $25–50 in capital cost per Muse user (each runs its own VM), Vals AI estimates heavy free-tier Muse use could exceed $50/month, and Zuckerberg projects >10x efficiency gains within a year to reach profitability. The piece ties this to last week's votes for data-center moratoriums, after which Meta and OpenAI executives said building datacenters will get harder, making inference efficiency the decisive variable for value capture.

The takeThe unit economics of always-on agents are emerging as the binding constraint on the AI buildout: Zuckerberg says Muse is 'very compute-intensive' and targets >10x cost-efficiency gains, Deutsche Bank pegs per-user agent capex at $25–50, and Musk admits Grok Bot serving costs are 'unsustainably high relative to revenue' — all as post-election data-center moratoriums harden the compute supply ceiling.

45Tech in AsiaCybersecurity

A Mythos quest? SEA fighting AI threats with obsolete security

Anthropic's Project Glasswing gave roughly 40 organizations early access to Claude Mythos Preview, a frontier model that can identify and exploit zero-day vulnerabilities in decades-old code, triggering alarm from regulators across Asia including MAS, Hong Kong and South Korean central banks and Singapore's CSA. Security vendors including Palo Alto Networks, Akamai, Lumen and Rubrik say the model collapses vulnerability discovery into exploit weaponization in minutes, with AI agents and autonomous attacks expected to follow, leaving banks and super apps most exposed. Regional enterprises remain vendor-loyal and reliant on legacy systems, with PwC projecting a 60% rise in cybersecurity risk investment in 2026 as firms weigh proactive spend against breach costs.

The takeAnthropic's Claude Mythos Preview compresses the attacker workflow from weeks of manual infrastructure mapping to minutes of discovery-to-weaponization at machine speed, prompting MAS, Hong Kong and South Korean central banks to force bank reviews while Palo Alto Networks warns such capabilities 'will not stay contained' and expects them commonplace within six months. The piece quantifies the exposure: Philippine banks lost US$95.7 million to cyberattacks in 2024 and Singapore absorbed over 1,800 attacks weekly, against enterprise security spend stuck at 5-10% of opex on legacy systems.

46BloombergAI & Models

White House Urges Easing Liability to Encourage AI Disclosure

At a Thursday White House meeting, Trump's AI task force (the Super Intelligence Force) recommended agencies limit regulatory liability for AI companies for at least 12 months to encourage reporting of critical cyber incidents, reversing last month's warnings that developers must report and resolve incidents or face consequences. Chaired by DNI Clayton with Pentagon vice chair and attended by Commerce Secretary Lutnick, FTC Chair Ferguson and OSTP's Kratsios, the group also discussed expanding the voluntary industry accord, incident-reporting protocols, and voluntary pre-release model testing run by the NSA and Commerce's Center for Advancing Innovation and Standards for Super Intelligence. The meeting occurred amid scrutiny following the ouster of a CEO over a relationship with a woman suspected of spying for China.

The takeThe White House's Super Intelligence Force is reversing course from threatening enforcement to offering a 12-month liability shield so AI labs will self-report critical incidents — a significant shift in US AI safety governance that trades accountability for visibility into rogue-model breaches. The move, paired with voluntary pre-release testing by the NSA and Commerce's new superintelligence standards center, signals the administration is building a disclosure regime on incentives rather than law.

47Tech in AsiaAI & Models

DeepSeek rolls out V4.1-Flash update with faster outputs

Hangzhou-based DeepSeek released V4.1-Flash-1017 across web, app and its developer platform, describing it as a minor update that nonetheless delivers substantial improvements in reasoning, coding, agentic and visual tasks with faster, shorter outputs, per a Hugging Face technical report. The release follows a roughly five-hour outage this week and mounting slowdowns, as Chinese AI providers compete aggressively on affordability.

The takeDeepSeek's V4.1-Flash-1017 iteration claims gains in reasoning, coding, agentic and visual tasks with shorter, faster outputs — a competitive move in the Chinese price war where Alibaba Cloud alone pushed 100+ models in September 2024. The five-hour outage and rising service disruptions signal capacity strain at a lab whose releases now set regional pricing and capability benchmarks.

48Hacker NewsGeopolitics & Defense

Ukraine knocks out 36% of Russia's oil refining capacity in under two weeks

Between Oct. 30 and Nov. 11, 2026, Ukrainian drone strikes hit at least nine major Russian refineries, including the Saratov plant, the Kirishi refinery (Russia's second-largest) and the Ust-Luga Baltic export terminal, per defense manufacturer Fire Point. The targeted plants' combined capacity exceeds 120 million metric tons of crude annually, or 36% of Russia's total refining capacity; Reuters reports roughly 30% of operational capacity knocked out and acute fuel shortages in at least six oblasts. The campaign was framed by Zelenskyy as retaliation for Russian strikes on Ukraine's energy grid.

The takeUkrainian long-range drone strikes destroyed refineries and export terminals representing over 120 million metric tons of annual crude processing—36% of Russia's refining capacity—in 13 days, with Reuters corroborating a ~30% operational hit and fuel shortages across six oblasts. This is a material demonstration of drone deep-strike capability against strategic energy infrastructure, a high-tech warfare and critical-infrastructure event with direct geopolitical and oil-market consequences.

49Institute for the Study of WarGeopolitics & Defense

Russian Offensive Campaign Assessment, November 13, 2026

ISW reports that Ukrainian officials say Starshield connectivity disruptions, tied to a 161 km/h in-motion speed cap, have been degrading Ukraine's ability to intercept Russian drones and conduct intermediate-range strikes since November 10, with all missions reportedly failing by November 13 — while SpaceX denies any change. Separately, Bloomberg-reported data shows at least six Western firms under Kremlin 'temporary management' have donated roughly 7.1 billion rubles ($86.86M) to Russian military charities funding drones and equipment. The report also details Kremlin maneuvering to limit European participation in peace talks to a NATO-dismantling security-architecture agenda, a false-parallelism push for a Kostyantynivka 'humanitarian' ceasefire, and Ukrainian Neptune strikes on Russian drone production in Rostov-on-Don.

The takeThe ISW assessment carries hard, verifiable detail on two thesis-relevant fronts: SpaceX's Starshield 161 km/h in-motion speed restrictions are reportedly degrading Ukraine's drone-interception and deep-strike targeting along the entire frontline, a concrete satellite-comms capability constraint in active warfighting, and Russia is funneling ~7.1 billion rubles ($86.86M) from seized Western assets into military charities that buy drones, ATVs and comms gear.

50Tech in AsiaMarkets

ByteDance plans up to $6b share buyback at $400b valuation

ByteDance has proposed buying back up to $6b of shares at $241.35 apiece, implying a ~$400b valuation, roughly 20% above last year's program, with some secondary trades near $500b. It also wants to extend the deadline for an IPO or full shareholder exit by four years to December 11, 2030, and will grant employees in its AI division 300 million tracking shares tied to that business's value.

The takeByteDance is repricing itself at ~$400b (up ~20% YoY, with secondary trades near $500b) while pushing its IPO/exit deadline to 2030, and the 300m tracking shares for its AI division are a concrete signal that Doubao/Seed is being carved out as a separately valued asset — a marker of how Chinese frontier AI is being capitalized outside public markets.

51Tech in AsiaSemiconductors

Samsung to produce DeepX’s next-generation AI chip

DeepX has completed design of the DX-M2 and handed final plans to Samsung's foundry for 2nm production, budgeting over US$100 million with first silicon expected in March 2027 and mass production targeted for 2028. The chip targets robots, industrial equipment and AI PCs, running generative AI models at roughly 5 watts to keep inference on-device rather than in remote datacenters. It follows the DX-M1 AI-of-things chip shown at CES 2024.

The takeA Korean fabless startup taping out a 2nm edge-AI chip at Samsung Foundry with >$100M committed and first silicon targeted for March 2027 is a concrete data point on Samsung's advanced-node customer traction and on the shift of generative AI inference to ~5W on-device silicon.

52Tech in AsiaMarkets

CapitalG leads $300m round for AI agent startup LangChain

LangChain raised $300 million in a Series C led by CapitalG, valuing the US startup at $3 billion, with Singapore's GIC, DFJ Growth, IVP and Sapphire Ventures participating after a $125 million Series B. Founded in 2022 as an open-source project for LLM task handling, the company now sells agent-management software, and its LangSmith monitoring and testing product counts 7,000 customers including Nvidia, Toyota, Salesforce, Harvey, Rippling and Clay.

The takeA $300M Series C at a $3B valuation, more than double the prior $125M Series B, with Singapore sovereign fund GIC participating, marks LangChain's pivot from open-source LLM tooling to enterprise agent-management infrastructure with 7,000 customers including Nvidia and Salesforce — a material signal of where agent-orchestration capital is concentrating.

53The Next PlatformSemiconductors

Cerebras Readies CS-5 Systems For 2027 Launch As It Bags Another Hyperscaler

Cerebras disclosed on its Q3 2026 earnings call that the WSE-5 waferscale processor is back from TSMC's 5nm node ahead of schedule and already running models in system test, targeting CS-5 production in early summer 2027 with more than 2x CS-4 performance. Q3 hardware sales fell 28.8% to $68.9M with gross margins of just 11% as the company ramps manufacturing and absorbs legacy G42 pricing, but Cerebras ended the quarter with $7.36B in cash and a $26.1B backlog, largely from its OpenAI inference deal, with AWS, Oracle and Microsoft Azure contracts not yet counted. Management expects to roughly triple sales in 2027 from an estimated ~$885M GAAP revenue in 2026.

The takeCerebras' WSE-5 taped out June 15 and returned from TSMC's 5nm fab ahead of schedule, running models in-chassis less than five months later, with CS-5 systems entering production in early summer 2027 at more than double CS-4 speed. Backed by $7.36B cash and a $26.1B backlog (mostly OpenAI's 750MW-to-2GW inference deal), plus AWS, Oracle and Azure wins, Cerebras is positioning waferscale as a credible decode-inference alternative to Nvidia GPUs and Grok LPUs.