Moonshot AI sparks $1T capex race as token costs plunge globally

Executive Summary

• AI Model Disruption: Chinese startups like Moonshot are rapidly closing the technological gap with US tech giants at lower token costs.

• Hardware Rebound: Semiconductor and server stocks rose as hyperscalers ramp up infrastructure investments toward $1T by 2027.

• Aerospace Advances: Commercial space and defense firms are advancing satellite infrastructure, autonomous defense platforms, and lunar supply chains.

Corporate Dynamics

Moonshot: Launches 2.8 Trillion Parameter AI Model and Prepares Hong Kong IPO

• News: Chinese AI startup Moonshot unveiled Kimi K3—a massive 2.8 trillion parameter AI model—at China's World AI Conference in Shanghai, pricing tokens at ~$3 per input and $15 per output million. Due to an overwhelming surge in subscription demand, Moonshot temporarily limited new subscriptions while expanding capacity. Founded by 33-year-old Carnegie Mellon graduate Yang Zhilin (who named the firm in Chinese after Pink Floyd's Dark Side of the Moon), Moonshot is preparing a Hong Kong initial public offering (IPO) within the next 6 months at a valuation exceeding $30 billion, targeting Asian and international investors from Singapore, Australia, and Hong Kong.

• Outlook: Kimi K3 narrows China's technological capability gap with top US frontier models from OpenAI and Anthropic to a couple of months (down from 6 to 12 months prior), driving enterprise workload migration toward lower-cost open-weight models while shifting proprietary frontier compute toward reasoning-heavy tasks.

BABA (Alibaba): Unveils Flagship 2.4 Trillion Parameter AI Model

• News: Alibaba released its latest flagship Qwen model featuring 2.4 trillion parameters, claiming it ranks second globally only to Anthropic's latest system. The launch drove a surge of over 4% in Alibaba's Hong Kong-listed shares.

• Outlook: Strengthens Alibaba's competitive positioning in China's advancing AI ecosystem, bolstering investor sentiment and driving domestic technology capital flows.

DeepSeek: Navigates Equity Volatility While Targeting 2027 Onshore IPO

• News: Amid an AI equity rout among Chinese quant firms, DeepSeek founder Liang Wenfeng’s investment firm dropped 16%. Unlike peers seeking international listings, DeepSeek expects an IPO around 2027 targeted on Mainland China's A-shares exchange to tap local onshore and retail capital.

• Outlook: Highlights near-term equity volatility across domestic Chinese AI firms while establishing a distinct, retail-focused onshore funding runway separate from cross-border capital markets.

SpaceX (SPCX): Targets 13th Starship Flight Following Ignition Abort and Secures Outperform Rating

• News: SpaceX targeted Thursday for its 13th Starship test flight after an automated system aborted a launch attempt last week right at ignition when 4 of 33 Raptor engines failed to ignite simultaneously. SpaceX identified the root cause within hours and swapped out 2 engines. SpaceX shares fell 1.5% (and dropped up to 4% in immediate after-hours trading), trading below its IPO price to under $121. Mizuho initiated coverage with an Outperform rating and a $200 price target based on 2029 valuation metrics, noting SpaceX enabled ~85% of all global orbital mass delivered over the past three years.

• Outlook: Leveraging a 90% launch cost advantage over competitors, SpaceX's Starship Version 3 launch cadence remains essential forStarlink constellation densification, aviation connectivity, and planned 2030 orbital data centers.

Interlune: Unveils Breakthrough Cryogenic Extraction for Quantum Cooling

• News: Space infrastructure startup Interlune introduced "Cold Capture," a cryogenic distillation technique extracting Helium-3 isotope from Earth-based Grade-A commercial helium to yield ~2.5 kg annually. The breakthrough addresses severe global shortages caused by dwindling tritium decay (12.3-year half-life) and has secured over $500 million in binding purchase orders spanning 7 to 10 years starting in 2028. Long-term plans aim to harvest 10 kg from lunar regolith in the early 2030s out of >1 million metric tons deposited by solar winds.

• Outlook: Unlocks commercial revenues in quantum computing supply chains, creating an immediate terrestrial bridge to long-term lunar extraction operations.

TSM (TSMC): Funds Record $265 Billion US Expansion via Cash Flow

• News: Taiwan Semiconductor Manufacturing Co. (TSMC) CFO Wendell Huang confirmed the company intends to fund its record $265 billion US expansion primarily through operating cash flow to protect shareholder dividends, though New Taiwan Dollar or US dollar bonds remain an option if market conditions dictate.

• Outlook: Preserves cash dividend returns for investors while expanding North American fabrication capacity, balancing internal capital discipline with global supply chain commitments.

CUSP AI: Raises $450 Million and Launches AI Materials Foundry

• News: Bezos-backed British startup CUSP AI raised $450 million to develop next-generation chip materials via AI, launching the AI Materials Foundry alongside partners Nvidia, Meta, and Hyundai.

• Outlook: Accelerates material science breakthroughs for semiconductor hardware through cross-industry strategic partnerships.

BA (Boeing): Postpones Brand-New Commercial Jet into Late 2030s

• News: Speaking at the Farnborough Airshow, Boeing CEO Kelly Ortberg stated the company will need until the end of the 2030s to introduce a brand-new commercial aircraft.

• Outlook: Directs capital allocation toward short-term financial stabilization and clearing existing customer production backlogs over costly new airframe R&D.

GOOGL (Alphabet): Advances Custom Chips, Proposes Regulations, and Targets 65% Cloud Growth

• News: Alphabet shares rose 1.5% (peaking up 3% intraday) following reports of developing a new server chip to boost AI model efficiency. Google Cloud revenue is expected to grow 65% year-over-year (outpacing AWS and Azure growth) with overall Alphabet revenue forecasted to grow 20% ahead of Wednesday earnings. Meanwhile, DeepMind CEO Demis Hassabis advocated before US congressional leaders for a federal pre-release AI testing regime for models like Gemini 3.5 Pro, following prior Trump administration directives forcing Anthropic to pull Fable 5 models and voluntary testing by OpenAI.

• Outlook: Rapid cloud revenue expansion and vertical chip integration help absorb heavy infrastructure capex while proactive regulatory engagement reduces federal compliance hurdles.

AMD: Unveils "Helios" AI Server Platform Co-Developed with Microsoft

• News: AMD introduced "Helios," its first rack-scale AI server system co-developed under a partnership with Microsoft. Supporting 72 AMD Instinct MI455X GPUs powered by Epyc CPUs and Pensando Data Processing Units (DPUs), AMD also secured supply agreements with Meta and Oracle.

• Outlook: Gains ground in high-density data center hardware, challenging rival rack systems by supplying integrated compute-and-networking architectures to major cloud providers.

NVDA (Nvidia): Develops Next-Gen "Kyber" Platform to Counter Competitor Hardware

• News: Nvidia shares advanced over 2% early before trimming gains, as the broader Sox index rose 2%. To protect market share against rival rack systems, Nvidia is developing its next-generation "Kyber" platform designed to double GPU capacity beyond current NVL72 systems (which support 72 GPUs across Blackwell, Grace Blackwell, and Vera Rubin architectures).

• Outlook: Neutralizes competitive rack-scale threats by doubling compute density per unit, defending its premium market share across enterprise data centers.

AVGO (Broadcom): Leads Semiconductor Rebound Following Oversold Call

• News: Broadcom shares gained over 2%, leading semiconductor equities higher following a bullish note from J.P. Morgan indicating the chip sector had become oversold.

• Outlook: Rebounds alongside broader sector sentiment as institutional buyers re-enter chip manufacturing equities ahead of tech earnings reports.

INTC (Intel): Surges Over 3% Amid Chip Sector Rebound

• News: Intel shares advanced more than 3% during today's trading session recovery across semiconductor equities.

• Outlook: Captures broad sector tailwinds as investors return to semiconductor hardware enablers ahead of quarterly reporting.

MRVL (Marvell): Re-establishes Buy-Side Interest Following Valuation Cooling

• News: Marvell recorded gains as Tematica Research CIO Chris Versace confirmed rebuilding positions in the stock following profit-taking in early July due to stretched valuations.

• Outlook: Reattracts growth-focused capital as pullback price levels offer renewed margin of safety for institutional investors.

ANET (Arista Networks) & ETN (Eaton): Highlighted as Key AI Infrastructure Enablers

• News: Wall Street analysts and fund managers highlighted Arista Networks (networking) and Eaton (power infrastructure) as vital AI-adjacent hardware plays.

• Outlook: Benefit from indirect capital flows into physical data center buildouts requiring expanded electrical capacity and high-speed network bandwidth.

ACHR (Archer Aviation) & Anduril Industries: Unveil Co-Developed Autonomous Aircraft

• News: Archer Aviation and Anduril Industries revealed a co-developed autonomous aircraft platform engineered for military and commercial applications under a 2024 partnership.

• Outlook: Broadens commercial and defense market opportunities by deploying mass-producible autonomous flight hardware.

Castelion & Saronic: Target Defense Hardware Growth via Venture Capital Expansion

• News: Venture firm Andreessen Horowitz (a16z) appointed Connor Love (former Army captain and Lightspeed investor) to its American Dynamism practice to back core US national infrastructure, including defense tech startups Castelion and Saronic.

• Outlook: Positioned to secure expanding defense hardware budgets for low-cost autonomous platforms proven in recent operational theaters like Ukraine and the Red Sea.

AXON (Axon Enterprise): Gains Tailwinds from ICE Adoption and Police Shortages

• News: Axon Enterprise continues to experience tailwinds driven by law enforcement labor shortages and planned adoption of body cameras announced by the director of ICE.

• Outlook: Expands high-margin recurring government revenue streams through regulatory mandates and agency technological adoption.

LMT (Lockheed Martin): Debuts Low-Cost Missiles and Invests Up to $9 Billion in Production Ramp

• News: At the Farnborough International Air Show, Lockheed Martin debuted the PAC-3 ACE missile (a PAC-3 MSE derivative targeting early 2028 flight prototyping at 1/3 to 1/4 the cost) and the MORFIUS X-Rotor (autonomous counter-drone system neutralizing up to 50 drones per pass set for release in months). Lockheed is investing $8 billion to $9 billion to triple or quadruple production for THAAD, PAC-3 MSE, and PrSM systems via US, European, and Middle Eastern partnerships.

• Outlook: Meets surging global air defense demand by drastically lowering per-unit interceptor costs while rapidly scaling manufacturing throughput.

AAPL (Apple): Pulls Back Slightly Following Recent Record Highs

• News: Apple shares retreated slightly during today's session, giving back a portion of gains achieved after touching record highs last week.

• Outlook: Undergoes standard equity consolidation following historic high valuations without altering broader consumer hardware expectations.

TSLA (Tesla): Slides 3% Pre-Earnings Amid Auto Margin Scrutiny

• News: Tesla shares declined 3% ahead of its upcoming quarterly earnings report.

• Outlook: Reflects investor caution regarding automotive profit margins and execution timelines for autonomous software.

MU (Micron) & SanDisk: Face Downside Risk Pending Guidance Updates

• News: Market analysts warned that a failure to raise guidance during quarterly calls could trigger 15% to 20% stock corrections in Micron and SanDisk near key technical support levels.

• Outlook: Exposes memory providers to valuation drawdowns if operational execution fails to match high market expectations.

ASML: Anchors European Tech Allocation in Semiconductor Equipment

• News: ASML was highlighted as a primary international technology holding for semiconductor equipment exposure alongside selective Japanese AI companies.

• Outlook: Capitalizes on global fab construction cycles as an indispensable equipment supplier across international supply chains.

Industry Trends

Open-Weight Model Disruption and Token Deflation

• Analysis: Open-weight Chinese models (Moonshot's 2.8T Kimi K3, Alibaba's 2.4T Qwen) are rapidly narrowing the technology gap with US frontier leaders (OpenAI, Anthropic) to just a couple of months. Chinese models currently account for 45% of token usage among US companies due to significant price efficiency (~$3 input / $15 output per million tokens). Despite benchmark evaluations across 900 math problems showing reasoning capability ~10 months behind prior Anthropic models, the lower token cost is driving mass commercial adoption.

• Outlook: Lower token costs accelerate the migration of enterprise workloads to cheaper open models while shifting expensive frontier compute toward complex reasoning tasks. This structural shift pressures valuation multiples for proprietary model builders while expanding total API demand.

Hyperscaler Capex Cycles and Rack-Scale Hardware Convergence

• Analysis: Tech industry AI capital spending is projected to surpass $1 trillion by 2027 as tech enterprises deploy infrastructure to monetize a $2 trillion backlog in cloud service agreements. Major tech companies report top-line growth is currently bottlenecked by physical capacity constraints rather than demand. Hardware design is rapidly shifting from individual GPUs to integrated rack-scale systems—such as AMD's Helios (72 MI455X GPUs) and Nvidia's upcoming Kyber platform—while adjacent providers like Arista Networks (networking) and Eaton (power infrastructure) capture spillover demand.

• Outlook: Compute and power constraints will sustain high capital expenditure across the entire semiconductor supply chain, benefiting component vendors, power grid enablers, and equipment makers like ASML across US, European, and Japanese markets.

Commercialization of Autonomous Defense and Deep Tech Infrastructure

• Analysis: Defense and space technology are pivoting toward mass-producible, lower-cost autonomous platforms (Lockheed's PAC-3 ACE missile, Archer/Anduril's joint aircraft, and a16z's American Dynamism practice backing Anduril, Castelion, and Saronic). Concurrently, commercial space relies on SpaceX's 90% launch cost advantage (~85% of global mass to orbit over the past 3 years), while startups like Interlune ($500M+ binding orders) commercialize critical cooling isotopes like Helium-3 for quantum computing.

• Outlook: Private and public capital flows are prioritizing space supply chain providers—where profit margins are roughly double those of end-system integrators—and scalable autonomous hardware proven in operational theaters like Ukraine and the Red Sea.

Market Sentiment

Hyperscaler Capex Confidence vs. Earnings Scrutiny

• Analysis: Equity recoveries in chipmakers (Broadcom +2%, Intel +3%, Alphabet +1.5%) demonstrate buy-side appetite anchored in massive cloud service backlogs ($2 trillion) and high cloud revenue expectations (Google Cloud expected up 65% YoY). However, market strategist Matt Orton (RJF (Raymond James)) notes that investor confidence requires clear proof of effective monetization and managed capex growth. Analysts caution that inadequate guidance updates could trigger 15% to 20% drawdowns in memory stocks sitting near key technical levels, such as Micron and SanDisk.

• Outlook: While the market broadly agrees on long-term $1 trillion capex trajectories by 2027, near-term equity performance will diverge based on company-specific guidance and margin preservation during quarterly earnings calls.

Token Pricing Deflation vs. Closed Model Monetization

• Analysis: Market concerns following ultra-low-cost Chinese model releases (Kimi K3) initially triggered a sell-off in compute equities over spending concerns. Sentiment stabilized as market participants realized that lower token costs accelerate overall enterprise adoption rather than reducing overall infrastructure spend. Proprietary model developers (OpenAI, Anthropic) maintain market support via direct consumer and enterprise third-party API revenue streams.

• Outlook: Institutional capital continues to favor hardware enablers (Nvidia, AMD, TSMC) and vertically integrated cloud platforms over standalone software wrappers, prioritizing concrete revenue backlogs over pure parameter scale.

  
  

Weekly In-Depth Analysis：

→ Read Weekly TechTrends：Google’s cash drain makes tech’s huge AI spend hard to justify