Bloomberg Tech

2026-07-21 · Hosted by Caroline Hyde, Ed Ludlow · Bloomberg / iHeartMedia

Executive Summary

Chinese AI startup Moonshot rattled markets with the surprise release of Kimi K3, a 2.8 trillion parameter open-weight model that Bloomberg's Peter Elstrom said rivals frontier systems from OpenAI and Anthropic. Chip stocks entered a bear market Friday on fears that more efficient AI needs less compute, but names like Nvidia rebounded Monday as the Nasdaq 100 added almost 1% and gained roughly 2%. Moonshot is now valued above $30 billion and is reportedly preparing a Hong Kong IPO within six months, while rival Alibaba unveiled an updated Qwen model (2.4 trillion parameters) that it says is second only to Anthropic's latest system.

Key Stories & Changes

1. Moonshot's Kimi K3 Shakes Up the AI Race

  • Kimi K3, a 2.8 trillion parameter model from Chinese startup Moonshot, drew praise for rivaling top OpenAI and Anthropic models

  • Demand was strong enough that Moonshot is temporarily limiting new subscriptions while it adds capacity

  • Chip stocks fell into a bear market Friday partly on questions about whether more efficient AI reduces compute/hardware needs

  • Monday: Nasdaq 100 up almost 1%, Nasdaq up 2%, with compute names (Nvidia) and memory names rebounding

  • Bloomberg's Peter Elstrom: China's AI capability gap versus the US, previously estimated at 6-12 months, now looks like "only a couple of months"

  • Alibaba released an updated Qwen model (2.4 trillion parameters) the same weekend, which it says is second only to Anthropic's latest system; Hong Kong-listed Alibaba shares rose on the news

  • Moonshot founder Yang Ji Lin, 33, is a Carnegie Mellon graduate who returned to China; he named the company after Pink Floyd's "Dark Side of the Moon"

  • Yang chose to build a large, frontier-scale model rather than a lean/efficient one

2. Moonshot Preparing Hong Kong IPO

  • Moonshot is said to be preparing a Hong Kong IPO that could price within six months

  • Current valuation is above $30 billion, per Bloomberg's Bailey Lipschultz

  • Rival DeepSeek is reportedly eyeing a mainland China listing sometime in 2027

  • A Hong Kong listing opens a different investor base (Hong Kong, Australia, Singapore) versus a mainland listing, which draws a more retail-driven investing community

  • Comparisons drawn to Alibaba and JD.com, which are dual-listed in Hong Kong and the US

3. SpaceX Targets Thursday for Starship Flight 13

  • SpaceX set Thursday as the new launch date after last week's attempt was scrubbed

  • SpaceX shares fell about 1.5% on the delay; the stock dropped 4% in after-hours trading during last week's aborted attempt

  • The abort was triggered when 4 of 33 Raptor engines on the booster failed to ignite properly; onboard systems stood the vehicle down before liftoff

  • This is Version 3 of Starship, a bigger and more powerful vehicle debuted about two months ago

  • Two engines are being swapped out; Morgan Stanley's Brett Lindsay called the turnaround (about a week) "relatively quick" for a program at this scale

  • Lindsay's thesis: SpaceX's AI, Neo Cloud, Starlink, and orbital data center business lines all depend on Starship being not just functional but rapidly reusable

  • SpaceX currently accounts for about 85% of all mass to orbit over the last three years via Falcon 9

  • Lindsay's $200 price target anchors mainly on Starlink and near-term government/enterprise contracts (including airlines), with orbital data center upside treated as longer-dated optionality (2030+)

  • On key-man risk with Elon Musk, Lindsay said it's "certainly a consideration" but views the company as having real moats and defensibility beyond any one individual

4. Andreessen Horowitz Adds Defense Investor Connor Love

  • Connor Love, a former Army captain and defense/space investor previously at Lightspeed, joined a16z's American Dynamism team

  • Love said American Dynamism represents a "once-in-a-generation opportunity," citing defense technology, energy, and manufacturing alongside software

  • He cited Anduril and Saronic as prior case studies, and pointed to Ukraine's battlefield and the Iran conflict as evidence for a "low-cost autonomy" thesis — cheaper, mass-producible systems reshaping deterrence

  • He referenced Archer's newly shown autonomous VTOL platform as an example of continued product expansion in the space

5. TSMC's $265 Billion US Expansion and Bond Market Plans

  • TSMC is committing a record $265 billion to its US expansion

  • CFO Wendell Huang told Bloomberg's Stephen Engle that TSMC aims to self-sustain CapEx via operating cash flow while maintaining sustainable, steadily increasing dividends

  • TSMC has previously raised $30 billion in the bond market, split roughly half in US dollars and half in NT dollars

  • Huang noted NT dollar funding is cheaper, while US dollar funding is becoming "more and more expensive"; a further bond raise is possible "if the timing is right"

  • NT dollar bond market is "much shallower" than the US dollar market, limiting how much can be raised in NT

6. Interloon's Helium-3 Breakthrough for Quantum Computing

  • Interloon, led by CEO/co-founder Rob Myerson, says it has devised "cold capture," a cryogenic distillation technique to separate helium-3 from commercial-grade helium on Earth

  • Helium-3 is used to chill superconducting quantum computing chips to millikelvin temperatures; historic supply came from decaying tritium, which is no longer produced in meaningful quantities (tritium has a 12.3-year half-life)

  • The Moon's lunar regolith holds an estimated over 1 million metric tons of helium-3; Interloon's Earth-based technique is a "stopgap" derived from its longer-term lunar extraction plans

  • Interloon targets producing 2.5 kilograms of helium-3 annually on Earth in the early 2030s, versus a longer-term goal of 10 kilograms via lunar extraction

  • The company has just under $500 million in purchase orders/contracts extending from 2028 out 7-10 years

  • Interloon plans to leverage NASA's Artemis program and launch/lander partners including SpaceX, Blue Origin, Firefly, and Intuitive Machines

7. CrossLink Capital's Take on Kimi K3 and AI Valuations

  • Mary D'Onofrio, partner at CrossLink Capital, called Kimi K3 "phenomenal" and compared it to last year's DeepSeek moment

  • She said falling inference costs (down 95%, per the discussion) likely won't reduce total dollar spend, because workloads will simply migrate from the most expensive frontier tiers to cheaper models rather than disappearing

  • Kimi K3 pricing ($3 per million input tokens, $15 per million output tokens) is roughly in line with leading models

  • On whether cheap Chinese models undermine trillion-dollar valuations for OpenAI and Anthropic, D'Onofrio said it's too early to know, noting both companies already generate "billions and billions" in revenue across first- and third-party businesses

  • She noted recent IPO activity (SpaceX, SK Hynix ADRs) signals investor appetite for AI infrastructure, but says the reopening IPO window favors long-standing, revenue-generating businesses over speculative ones

8. Cusp AI Raises $450 Million; DeepSeek Founder's Fund Drops; Boeing Delays New Jet

  • Cusp AI: Bezos-backed British AI startup raised $450 million to accelerate chip-materials R&D — N/A (private raise) — Launching an "AI materials foundry" with 48+ organizations including Nvidia, Meta, Hyundai

  • High-Flyer (Liang Wenfeng's fund): DeepSeek founder's investment firm hit after China's quant firms were hit with an AI-related route — -16% — Highlights volatility and bubble concerns in Chinese AI-linked stocks

  • Boeing: CEO Kelly Ortberg told Bloomberg's Guy Johnson at the Farnborough Air Show the market isn't ready for a new jet — N/A — Next-gen jet not expected until "end of the next decade"; focus remains on backlog delivery of current models

9. Week Ahead: Alphabet, Tesla, Intel Earnings

  • Tech earnings kick off this week: Alphabet and Tesla report Wednesday, Intel reports Thursday (IBM already pre-released)

  • Bloomberg Intelligence's Mandi Singh said cloud revenue remains the key metric for hyperscaler AI monetization

  • Debate centers on whether 2027 CapEx growth will be 30-40% or another 50%+ increase year-over-year, with aggregate hyperscaler CapEx approaching the trillion-dollar range

  • For Alphabet specifically: overall revenue growth expected around 20%, but Google Cloud growth expected near 65% year-over-year

  • Singh said Google Cloud is growing roughly 15 percentage points faster than Azure and AWS; OpenAI's deployment on AWS (in addition to Azure) adds a wrinkle for Microsoft

  • Singh cautioned it's still "too early" to see token-level monetization clearly reflected in hyperscaler earnings

1. China's AI Gap Is Closing Faster Than Expected

Kimi K3's reception, alongside Alibaba's updated Qwen model, led Bloomberg's own reporting team to conclude China's AI capability gap versus the US has narrowed from an estimated 6-12 months to just a couple of months. This narrative was reinforced by both the Bloomberg Asia tech desk and CrossLink Capital's Mary D'Onofrio, who separately compared the moment to last year's DeepSeek shock — suggesting this is now a pattern rather than a one-off surprise.

2. Open-Weight Model Efficiency Debate Meets Market Volatility

The core market question all episode-long was whether cheaper, more efficient open-weight models like Kimi K3 reduce demand for compute and chip hardware. Chip stocks briefly fell into a bear market on that fear, then partially rebounded as investors concluded — per D'Onofrio — that falling inference costs shift workloads to cheaper tiers rather than reducing total spend.

3. Public Market Appetite for AI/Space Infrastructure Names

Recent IPO activity, including SpaceX-related coverage and SK Hynix's ADR listing, was cited as evidence that investors are favoring established, revenue-generating businesses tied to the AI buildout rather than purely speculative names. D'Onofrio framed this as a reopening IPO window that still demands financial substance over hype.

4. Reusability as the Bottleneck for the Next Space Economy

Starship's repeated test delays underscore that SpaceX's next wave of business lines — Starlink densification, Neo Cloud, orbital data centers — all hinge on proving rapid reusability at scale, not just successful individual launches. Morgan Stanley's Brett Lindsay framed near-term valuation conservatively around Starlink and government contracts, treating orbital data centers as longer-dated optionality.

5. American Dynamism Broadens Beyond Pure Defense

Connor Love's arrival at a16z reinforces a shift already underway: American Dynamism investing is expanding from core defense primes (Anduril, Saronic) into energy, manufacturing, and space infrastructure, framed as rebuilding America's industrial base rather than a narrow defense-tech bet.

6. Capital-Intensive Infrastructure Financing Strategies Diverge

TSMC's approach — self-sustaining CapEx via operating cash flow, supplemented opportunistically by bond issuance — illustrates how the largest AI-supply-chain players are managing enormous capital needs ($265 billion in TSMC's case) while balancing currency-driven funding costs between cheaper NT dollar debt and costlier US dollar debt. ---

Sentiment Analysis

Overall Market Sentiment: Cautiously Rebounding

Technology stocks recovered part of Friday's bear-market losses in chip names as investors moved from initial fear about AI efficiency undercutting compute demand toward a more constructive read on the same news.

Risk Factors Highlighted

AI compute demand uncertainty: If more efficient open-weight models like Kimi K3 reduce the need for compute, it could pressure chip and hardware names that sold off into a bear market Friday.

China's narrowing AI gap: Bloomberg's own reporting suggests China's capability gap versus the US has shrunk from 6-12 months to just a couple of months, raising competitive and possibly geopolitical concerns.

Chinese AI stock/fund volatility: DeepSeek founder Liang Wenfeng's investment firm fell 16% amid a broader quant-fund route, signaling possible bubble dynamics in AI-linked Chinese equities.

Starship technical risk: A repeat scrub (four Raptor engines failing to ignite) shows the vehicle is still working through technical issues just before a planned relaunch.

SpaceX business-line dependency on Starship reusability: Neo Cloud, orbital data centers, and expanded Starlink capacity are all predicated on Starship not just flying, but being rapidly reusable — a still-unproven cadence.

Key-man risk at SpaceX: Analyst Brett Lindsay acknowledged Elon Musk's outsized role across every division as a consideration in valuing the company.

Uncertain path to token-level AI monetization: Bloomberg Intelligence's Mandi Singh said it remains too early to see clear evidence of token monetization in hyperscaler earnings, despite massive CapEx commitments.

Valuation risk for frontier AI labs: Questions persist about whether trillion-dollar valuations for OpenAI and Anthropic are justified given the emergence of much cheaper competitive models; CrossLink's D'Onofrio said "nobody knows yet."

Currency-driven financing costs for TSMC: Rising US dollar funding costs versus a "much shallower" NT dollar bond market constrain TSMC's flexibility in financing its $265 billion US expansion.

Helium-3 supply bottleneck for quantum computing: Legacy sources of helium-3 (tritium decay) have dried up, threatening the ability to scale superconducting quantum computers unless new extraction methods like Inte

This episode was covered in today's [The Market Signal — 2026-07-21](https://marketsignal.beehiiv.com/p/the-market-signal-2026-07-21), a cross-source synthesis of multiple podcast reports.

Keep Reading