By Dongyan Zheng in AI — 23 Jun 2026

ChinaBiz Briefing | Alibaba vs ByteDance AI Video Battle, Compute Crunch, and CATL's UK Debut

China's tech and industrial sectors delivered a dense set of signals on June 23: Alibaba and ByteDance traded blows in AI video generation, the structural cost crisis inside China's AI model industry came into sharper focus, CATL extended its battery infrastructure ambitions into the UK, XPeng revealed a supply chain overhaul embedded in its next SUV, and Morgan Stanley placed a high-conviction bet on China's domestic GPU ecosystem. Taken together, the day's news maps the fault lines of China's technology economy in 2026 — where ambition is outrunning infrastructure, and where global expansion is accelerating even as domestic constraints tighten.

Alibaba Fires First in AI Video, Hours Before ByteDance Could Define the Narrative

Alibaba released HappyHorse 1.1 on the evening of June 22 — deliberately timed to preempt ByteDance's Seedance 2.1 showcase at the Volcano Engine Force Conference the following day. The upgrade delivers five targeted refinements: improved motion consistency in high-kinetic sequences, multi-character subject stability supporting up to nine reference images, stronger prompt fidelity, granular skin-texture realism, and dynamic audio prosody. Alibaba simultaneously launched a creator competition with a prize pool exceeding RMB 300,000 in cash and 3.1 million platform credits, with top awards including brand deals worth up to RMB 1 million.

Why it matters: The two companies are pursuing divergent technical philosophies — HappyHorse optimizes for commercial workflow utility in short-drama and e-commerce contexts; Seedance targets physical realism for higher-budget cinematic applications. The competition's 50% HappyHorse content threshold is a deliberate ecosystem lock-in mechanism. But creator loyalty in the AIGC space has proven transactional, and the window between "technically impressive" and "production-stable" remains wide across all models.

China's AI Labs Are Rationing Tokens — and That's a Structural Warning Signal

Zhipu AI, Moonshot AI's Kimi, and MiniMax all launched their most capable coding models in June 2026, then almost immediately imposed purchasing restrictions on those same products. Kimi and MiniMax APIs are running at persistent overload; Zhipu's subscription plans require daily queuing and have been repriced three times in a year. MiniMax abandoned flat monthly subscriptions entirely for per-token billing after its M3 coding agent — which ran autonomously for nearly 12 hours on a single task — exposed the unsustainable economics of software-style pricing for agentic workloads.

Why it matters: When a software company limits how much of its product customers can buy, it has reclassified its output from infinitely replicable code to an industrially constrained commodity. China's AI companies face this problem with a harder constraint than their U.S. counterparts: a supply chain that is simultaneously sanctioned, capacity-constrained, and architecturally immature. Zhipu AI has accumulated losses of approximately RMB 6.2 billion over three and a half years; MiniMax carries roughly RMB 7 billion in losses over the same period. DeepSeek is responding by moving to build gigawatt-scale proprietary compute, with founder Liang Wenfeng personally committing approximately RMB 20 billion in the company's first funding round. The token rationing visible today is not a temporary inconvenience — it is the first audible signal of a structural cost ceiling.

Morgan Stanley Raises China AI Chip TAM to $91B, Bets on Domestic GPU Champions

Morgan Stanley's Greater China semiconductor team raised its China AI GPU total addressable market forecast by 36% to US$91 billion by 2030, implying a 23% CAGR. The revision is driven by US$80–100 billion in projected capex during 2026–2027 and a revised sovereign/SOE TAM of US$9 billion, underpinned by China's reported plan to deploy RMB 2 trillion in national data center infrastructure over five years. The bank projects China's AI chip self-sufficiency ratio to rise from 42% in 2025 to 70% by 2030.

Why it matters: The note's most consequential assumption is that Chinese AI chip vendors will not merely substitute domestically but will follow Chinese cloud capital overseas — modeling Chinese GPU penetration of overseas CSP data centers rising from zero in 2027 to 20% by 2030. That is a structural bet, not a consensus view. It implies U.S. export controls are functioning as a forcing function for Chinese semiconductor self-reliance rather than a ceiling on it.

CATL Takes Battery-Swap to the UK, Using Freight Corridors as a Proving Ground

CATL and Octopus Energy announced a joint venture on June 22 to build a commercial heavy-truck battery-swapping network across the United Kingdom — the first overseas deployment of CATL's Qiji platform. The first demonstration stations are targeted for 2027, with expansion to more than 30 stations across England, Scotland, and Wales by 2035, covering major freight corridors, logistics ports, and distribution centers. The Qiji system completes a full battery exchange in approximately five minutes using standardized packs, and is designed to absorb surplus renewable electricity, reducing grid stress.

Why it matters: CATL is positioning the UK project as a replicable template for broader European expansion. The battery-swap model is a direct commercial challenge to the megawatt-class fast-charging approach favored by most European operators, competing on operating economics, battery cycle life, and grid compatibility. For CATL, establishing infrastructure ownership in Western markets is as strategically important as selling cells — it deepens dependency and creates recurring service revenue.

XPeng's MONA L03 Reveals a Supply Chain Overhaul, Not Just a New SUV

XPeng confirmed the naming of the MONA L03 compact SUV on June 22, with a commercial launch expected before year-end. The vehicle replaces BYD's Fudi blade cells with CALB batteries — severing cost dependency on a direct competitor — and sources its 183 kW electric motor from Luxshare Precision, the Apple Tier-1 supplier making its first major EV powertrain entry. The L03 carries XPeng's proprietary Turing AI chip across all trims, running the VLA 2.0 vision-only autonomous driving stack. The MONA M03 sedan currently accounts for approximately 46% of XPeng's total group volume — a concentration risk the L03 is explicitly designed to address.

Why it matters: The L03 is XPeng's test of whether its supply chain leverage can travel upmarket. The M03 proved XPeng could optimize existing domestic suppliers for cost at the entry tier; the L03 is testing a different thesis — that XPeng can function as a supply chain architect, selecting partners for strategic fit and co-investment potential rather than unit economics alone. Luxshare's entry into EV motors, backed by XPeng's volume mandate, is one of the more significant cross-sector pivots in China's industrial supply chain this year.

Nine Chinese brands combined for approximately 1,800 registrations in Norway in May 2026, capturing close to 12% market share in a month when BEVs accounted for 97.8% of all new vehicle sales. BYD ranked seventh with 567 units; XPeng posted 551 units, up 74.4% year-on-year; MG registered 368 units. Tesla reclaimed the top position with 3,345 units and 21.5% share. Norway's total May registrations rose 9.1% year-on-year to 15,560 units, the strongest May in four decades.

Why it matters: Norway's near-universal EV adoption rate strips away the combustion-era brand advantages that protect incumbent automakers in less electrified markets. A 12% combined share indicates Chinese brands have cleared the initial threshold of market entry in Europe's most demanding EV environment. Sustaining that position will depend on after-sales network depth and consumer trust — assets that take years to build and cannot be accelerated by product launches alone.

What to Watch Next

ByteDance's Seedance 2.1 showcase response to Alibaba's preemptive move will set the tone for China's AI video competitive cycle through Q3. DeepSeek's infrastructure build timeline — and whether its self-sufficiency model can be validated before the capital cycle tightens — is the most consequential unresolved variable in China's AI industry. On the hardware side, Cambricon's MLU690 launch in Q4 2026 will be the first real-world test of Morgan Stanley's revised TAM thesis. And for XPeng, the MONA L03's sales trajectory will determine whether its supply chain architecture ambitions are commercially validated or remain a strategic hypothesis.

Related Coverage:

Alibaba Launches HappyHorse 1.1 Ahead of ByteDance’s Seedance 2.1 in AI Video Race Morgan Stanley Raises China AI Chip TAM to $91B by 2030, Bets Big on Domestic GPU Champions CATL Partners With Octopus Energy to Build UK Heavy-Truck Battery-Swap Network XPENG's MONA L03 SUV Signals a Strategic Pivot: From Cost Cutter to Supply Chain Architect China's AI Firms Are Rationing Tokens. That's the Bubble Warning Nobody Wants to Hear.Chinese EV Brands Capture 12% Share in Norway's Near-Total Electric Market