ChaiA.sonic💥 retweeted
A few weeks ago, I was asked if I was planning to add more $RKLB shares since the price had dropped so much from a high of $150 not long ago. This is what I had to say at the time when the price was in the mid-to-high $60 range.
Now that the Neutron delay is official and the price has tanked to the low $50 range, I'm starting to re-add to my position. Today, I added ~9% more. If it goes further down, I might do more depending on how the rest of my portfolio is doing.
I continue to believe that Rocket Lab is in an incredibly good position to capitalize on the growing need for access to space in both satellite technologies and launch technologies. The Neutron delay doesn't change the core thesis long-term.
Imagine seeing this entire stack and deciding $RKLB lives or dies because Neutron slipped into 2027.
CEREBRAS $CBRS: AI’S SPEED KING AND THE $1,000 BULL CASE FOR 2029–2030
My bull case for 2029–2030 come to roughly $800–1,000 per share. The financial model uses 2029, with growing customer evidence supporting the opportunity.
Start with capacity.
Cerebras’s SEC filing shows 750MW committed by OpenAI, with deployment expected during 2026–2028, plus an option for another 1.25GW by the end of 2030. That provides a contractual path to a potential 2GW. My 2029 scenario assumes an accelerated buildout and additional cloud customers bring average productive cloud capacity to 2GW during that year.
The infrastructure expansion is tangible. Cerebras announced a Finland facility scaling in phases to 165MW of contracted IT capacity, backed by seven-year service orders, with construction underway on the initial 50MW phase. This supports the expansion thesis. Achieving 2GW of productive average capacity in 2029 requires substantial execution.
On the August 12 Q2 earnings call, management said more than 600MW was live or contracted for delivery by the end of 2027, with an expansion pipeline measured in gigawatts. Manufacturing capacity was expected to increase more than 10× in 2026. Management also expects core revenue to more than triple in 2027 and continue growing at multiples in subsequent years. Those statements support the ramp thesis; $30–40B in 2029 is the bull-case projection.
Pricing?
I assume $15–20M of annual core revenue per MW, equivalent to $15–20B per GW.
There are real industry benchmarks behind that assumption. Nebius $NBIS disclosed four landmark deals at $20–25M of annual contract value per MW. CoreWeave $CRWV subsequently announced short contracts at approximately $40M/MW in annualized revenue. Those CoreWeave contracts run three to six months and reflect premium pricing.
These industry benchmarks involve different hardware and contract structures. My $15–20M of annual core revenue per MW remains an assumption Cerebras must achieve across its fleet, but it is realistic. The $25.4B backlog spans several years and includes certain pass-through amounts. Reaching $30–40B of annual core revenue in 2029 requires substantial additional bookings, timely capacity delivery and achievement of the modeled revenue yield. That being said, we are very early in the freshman year of the IPO and early in the roll out and scale up.
Market size gives this model context.
Cerebras’s IPO filings, citing Bloomberg Intelligence, estimate its addressable inference opportunity growing from approximately $66B in 2025 to $292B in 2029. My $30–40B revenue scenario equals approximately 10.3–13.7% of that 2029 forecast. At the $35B midpoint, the comparison is almost exactly 12%.
The specialized opportunity is substantial too. Public reporting of Citi $C analyst Atif Malik’s research puts fast inference at $130B in 2030. My revenue range equals approximately 23–31% of that later-year estimate, with the $35B midpoint at roughly 27%. These forecasts use different market definitions and include hardware or infrastructure alongside services. The 2030 date also matters when assessing the pace required to reach my 2029 target.
The physical footprint offers another perspective. McKinsey projects global inference demand of 71.5GW in 2029. My modeled 2GW fleet equals approximately 2.8% of that forecast. This is a nominal capacity comparison across all providers. Architecture, utilization and pricing determine the economic value of each MW and the resulting revenue share.
My thesis centers on Cerebras building a substantial, profitable position in fast inference. That specialty spans coding, agents, finance, security and scientific workloads, giving a focused architecture broad potential demand.
The customer evidence is growing too.
OpenAI identifies Jane Street, Podium, Basis and Rogo as early users of Cerebras-powered Ultrafast. Faster inference is already enabling more responsive development, voice and financial research workflows. These are customers of OpenAI’s service, powered by Cerebras capacity.
At Supernova on August 18, OpenAI’s Thibault Sottiaux said “we would love for Ultrafast to just be the default.” He acknowledged that this remained an ambition. That is a significant product aspiration from the customer already anchoring Cerebras’s capacity expansion.
SemiAnalysis (in their paywalled article we learned about during the “UltrafastGate” scandal that hit this weekend) reported on October 1 that Ultrafast service to Jane Street and a few other customers was being delivered at $200M/MW. That remains a reported outlier: I have not found independent confirmation, and the claim leaves the time period and Cerebras’s revenue share unspecified.
In his August 25, 2026 Dwarkesh Podcast interview, Dylan Patel separately discussed Jane Street capturing $300–500M/MW in economic value from AI, without documenting the calculation. My model uses the $15–20M annual core revenue per MW assumption above.
There are more disclosed commercial relationships worth including.
Cerebras signed cloud capacity agreements with Cognition and Lovable. It names Block $XYZ, Figma $FIG, AlphaSense and GSK $GSK as customers using fast inference for agentic workflows. Lovable’s agreement specifically includes dedicated Cerebras capacity for selected latency-sensitive workloads. On the Q2 call, Feldman disclosed six deals above $30M signed during the quarter.
The customer testimony helps explain why speed commands value. At Supernova, Figma’s AI research product lead Pavi Bhatter said “speed can become our strategy.” Cognition’s Silas Alberti described up to 4× faster completion of code-search-assisted tasks while preserving answer quality. These are examples of faster inference improving useful work.
General Compute announced a multi-year agreement to deploy Cerebras systems, initially targeting agentic coding. Gimlet Labs announced a Cerebras deployment for Gimlet Cloud. Callosum is integrating Cerebras silicon into its platform for heterogeneous agentic inference. These relationships expand the channels through which developers can access Cerebras performance. Their disclosed announcements provide no contract revenue figures.
Cerebras has also named IBM $IBM, Mistral, Notion and Mayo Clinic among organizations that have chosen its technology. Its SEC filing identifies G42 and Mohamed bin Zayed University of Artificial Intelligence as existing significant customers. These relationships broaden the established customer foundation.
The business is already producing revenue through both channels. Q2 core cloud and other services revenue grew 287% to $127.7M, while core hardware revenue reached $82.1M. CFO Bob Komin also disclosed several late-stage hardware opportunities representing hundreds of millions of dollars in the pipeline. Those prospective deals provide additional hardware upside as orders and revenue develop.
Cerebras earns revenue by selling systems and by selling access to its compute. Dedicated cloud contracts generally carry take-or-pay commitments, while on-demand customers pay for usage. My 2GW model assumes capacity monetized through Cerebras’s cloud services. External system sales and leases have separate economics. Systems deployed into Cerebras’s own cloud provide the infrastructure that earns service revenue, so that hardware investment and the resulting cloud revenue must be accounted for consistently.
AMD $AMD belongs in the thesis.
The announced joint architecture uses AMD Helios for high-throughput prefill and Cerebras for low-latency decode. The companies project up to 5× higher tokens per second per watt relative to a Cerebras-only configuration in their disclosed modeling. Cerebras’s Q2 update targeted production in Q4 2026.
At Supernova, AMD CTO Mark Papermaster said: “The users get better economics and low latency and that quick response.” He also described a partnership built over years, including AMD EPYC processors already used in Cerebras clusters. Arista Networks $ANET supports the networking layer, as Feldman explained at the event.
The commercial direction matters financially. Cerebras plans to buy and deploy AMD Helios systems. On the Q2 call, Feldman affirmed that Cerebras purchases the racks and receives the revenue from customers using the resulting cloud service. Discussing the joint offering, he added: “we have buyers for it already.” Buyer identities and order values were not disclosed. I found no announced AMD purchase commitment for Cerebras systems. The demonstrated opportunity is improved cloud economics and capacity to serve more customers.
Cybersecurity adds another growth opportunity.
CrowdStrike $CRWD and Cerebras have a confirmed strategic partnership. CrowdStrike will use Cerebras inference to help power Falcon AI Detection and Response. Cerebras’s Q2 update describes inline LLM security covering a large portion of enterprise traffic.
At Supernova, CrowdStrike VP of Engineering Keith Culley said fast inference enables “5x to 10x more inspection inside the same acceptable window of time.” Cerebras also highlighted Armis, now part of ServiceNow $NOW, reporting a code-security scan completed in roughly one-third the time, with more vulnerabilities found at lower cost. That comparison was presented by Cerebras; no contract value was disclosed.
My speculation is that broader deployment through CrowdStrike, larger security models and more agent activity could materially expand Cerebras compute demand. Gartner forecasts $244B in worldwide information-security spending for 2026.
CrowdStrike’s 2026 Global Threat Report found AI-enabled adversary activity increased 89% year over year, with average eCrime breakout time falling to 29 minutes and the fastest observed case taking just 27 seconds.
Inline security puts protection directly in the path of live enterprise AI traffic, making response time critical to keeping those workflows usable. Faster inference can support larger, smarter security models running continuously. That is the foundation for a potentially substantial, recurring inference business. Future deployment volumes and revenue remain undisclosed, and additional cybersecurity expansion beyond the modeled 2GW fleet remains unquantified upside.
Demand from national security customers also has a concrete foundation.
Sandia deployed four Cerebras CS-3 systems as the first part of a planned eight-node Kingfisher system, supporting the NNSA’s AI for Nuclear Deterrence strategy. Sandia, Lawrence Livermore and Los Alamos subsequently entered a third round of collaborative research with Cerebras through the Advanced Memory Technology program.
Cerebras also has a DOE memorandum of understanding supporting exploration of further AI and HPC collaboration under the Genesis Mission. Its Carahsoft partnership and inclusion in the Department of Defense’s Tradewinds Solutions Marketplace provide government procurement channels.
On the Q2 call, Feldman also described government technology partnerships involving stacked memory and integrated wafer-scale optical solutions, extending the relationship into future hardware development.
My speculation is that these existing relationships could expand into larger secure AI deployments, scientific agents, intelligence analysis and cyber defense. I leave that future expansion unquantified.
AWS and Meta add further upside.
Amazon’s AWS $AMZN has announced a hybrid inference architecture combining Trainium for prefill with Cerebras for decode, deployed inside AWS and accessed through Bedrock. Cerebras’s Q2 update targets availability in Q1 2027. That creates a distribution channel to enterprises already buying compute through AWS.
The commercial foundation is documented. Cerebras’s June 10-Q describes an initial multi-year AWS lease, options for future procurement and manufacturing capacity commitments. It explicitly says additional AWS purchases could make AWS a material percentage of revenue. My bull case includes Bedrock adoption translating into larger deployments and repeat orders as coding and agent workloads grow.
Feldman made a useful clarification on the Q2 call: “our $25 billion in RPO does not reflect any backlog of business from AWS or any other hyperscaler at this time.” That leaves hyperscaler expansion outside the backlog figure investors already know.
Meta $META also has an established connection. Its April 2025 LlamaCon announcement introduced experimental access to Cerebras-powered models through Llama API. That provides a technical precedent and an existing relationship for future expansion.
Meta’s newer Muse platform emphasizes agents, coding and reducing task latency, all areas where faster inference can matter. Broader Cerebras deployment across those workloads is a plausible upside scenario. Any Cerebras role in Muse remains speculative.
Alphabet’s Google $GOOGL deserves continued research.
Google named Cerebras as a supporter of its MLIR compiler infrastructure, and Cerebras now serves Google DeepMind’s Gemma 4 model. Those are documented technical connections. At Supernova, Cerebras CMO Julie Choi also mentioned demos from DeepMind, adding another ecosystem connection worth tracking.
On the Q2 call, management reported progress in discussions with other hyperscalers; an analyst also asked about Google and Microsoft $MSFT. I found no public Google purchase agreement or quantified Cerebras revenue commitment. I assign no Google revenue to this model.
These relationships give Cerebras additional paths to customer diversification and sustained demand into the 2030s.
The technology roadmap supports the economics too.
At Supernova, CTO Sean Lie described a CS-4 design with 50% fewer components, 60% more manufacturing automation and up to 3× faster system deployment. The roadmap targets twice the inference speed each year and solutions with up to 20× higher throughput by 2027. These are company claims and targets. Their financial significance is the potential to deliver more billable tokens per MW at lower cost as utilization and pricing evolve.
At Hot Chips 2026, Cerebras detailed CS-5, targeted for 2027, aiming for up to 5,000 output tokens per second per user on the largest frontier models. Its subsequent CS-6 roadmap adds 3D-stacked DRAM so more of each model can fit within each system, potentially reducing the infrastructure needed per workload and improving capital efficiency as those products are delivered.
My financial model:
2GW × $15–20B annual core revenue/GW = $30–40B revenue.
At 40% core operating margin = $12–16B core operating profit.
At 20× projected 2029 core operating profit = $240–320B enterprise value.
That multiple is equivalent to 8× projected 2029 core revenue.
Cerebras management has outlined a medium-to-long-term profitability target of approximately 40% core operating margin. Applying it in 2029 is part of my bull-case assumption. Core figures are non-GAAP and exclude stock-based compensation, customer-warrant amortization, data-center pass-through revenue and costs, and certain other items.
The Q2 call helps explain the margin thesis. Feldman said: “More tokens per system at lower cost means more revenue and more gross margin.” Komin reported core operating margin improving from negative 42% to negative 16% year over year, and described a path toward 60%+ core gross margin through higher pricing, improved throughput and a shift from higher-cost rented systems to owned capacity. Profitability remains a future milestone.
Capital efficiency matters just as much. Komin said Cerebras has “much lower net capital expenditure per megawatt” than most AI cloud providers. His explanation: Cerebras deploys its own hardware at bill-of-materials cost, while its largest customer reimburses a meaningful portion of data-center fit-out spending. That is management’s comparison, and it provides a concrete reason the company could retain more cash from its growth.
For valuation context, Nvidia $NVDA and Broadcom $AVGO currently trade around 28.5× and 39.8× trailing operating profit, measured using enterprise value. Their published forward P/E ratios are approximately 19.4× and 20.5×. Those figures provide context across different accounting measures and time periods.
My 20× assumption values Cerebras in 2029 using projected 2029 core operating profit, which includes equipment depreciation. At a 40% operating margin, that equals 8× revenue. The valuation requires sustained growth and cash generation after ongoing equipment and data-center investment. With an illustrative 20% tax rate and zero interest expense, 20× operating profit corresponds to approximately 25× earnings.
The midpoint is straightforward:
$35B revenue × 40% margin × 20 = $280B enterprise value.
Equivalently:
$35B revenue × 8 = $280B enterprise value.
My approximately $903 midpoint assumes 310M diluted shares and roughly zero net debt in 2029, with cash and liquid investments offsetting borrowings. The cloud buildout’s future funding mix remains uncertain. Holding enterprise value and share count constant, each $1B of net debt reduces equity value by approximately $3.23 per share.
A $250–300B equity value translates to $806–968 per share. At 360M diluted shares, it becomes $694–833. The share count and net-debt position are future assumptions that materially affect the result.
Cerebras ended June 2026 with approximately $7B of financial net cash, excluding restricted cash and operating lease liabilities. Management also describes lower net capital requirements per MW because it deploys its own hardware at bill-of-materials cost and receives reimbursement from its largest customer for a meaningful portion of data-center fit-out spending. Those advantages provide a plausible path to financing expansion through existing liquidity, customer funding and cash generated as profitability develops.
Customer usage within the 2GW cloud model gets counted once. AMD-enabled cloud usage follows that same rule. AWS leases and hardware purchases, along with systems sold to other operators, require separate revenue and margin assumptions as their economics become available.
Additional AWS, Meta, cybersecurity and government expansion remains unquantified upside here.
Capacity delivery, realized pricing, margins and dilution determine the outcome. The demand thesis centers on shorter time to decision across coding, agents, finance and security.
ChaiA.sonic💥 retweeted
THE FULL SPACE VALUE CHAIN
Really interesting seeing how differently major space companies are positioned across launch, manufacturing, spectrum & services heading into 2028:
• $RKLB combines rockets, satellite manufacturing & defense with Iridium global spectrum and 2.55M+ existing subs
• $SPCX controls launch, satellite manufacturing & Starlink through its vertically integrated network
• $ASTS connects everyday smartphones directly to satellites using large arrays & carrier spectrum
• $PL operates Earth observation satellites & sells imagery, data & analytics
• $FLY builds launch vehicles, lunar landers & defense systems
• $YORK manufactures satellites for Pentagon constellations & defense programs
• $RDW supplies spacecraft components, sensors & mission hardware
• $BKSY operates imaging satellites & delivers real time geospatial intelligence
You can also see below how the bigger opportunity in space is shifting toward owning scarce spectrum, infrastructure and recurring customer relationships.
ChaiA.sonic💥 retweeted
My interview with Frank Klein and Dr. Brad Clevenger premieres 30 minutes before the market opens! 🚀
youtu.be/CIMQvRFW2y0
I’m excited for you all to see this conversation from my visit to Mynaric in Germany.
I recorded an introduction to give you some context from the visit. It’s a little longer than I originally planned, but the factory tour will be released separately after Rocket Lab has reviewed it, so I felt that context was worth including ahead of the interview.
A huge thank you to Rocket Lab for organizing the visit, and to Frank and Brad for taking the time to speak with me.
Looking forward to hearing what you think!
@RocketLab
#RocketLab #Mynaric #RKLB
ChaiA.sonic💥 retweeted
MY $BABA VALUATION MODEL
PRICE TARGET: $192 --> $203
POTENTIAL UPSIDE: +61% ✅
"The stock remains undervalued modeling $100B cloud revenue in 5 years and a decent operating leverage"
ASSUMPTIONS:
LTM Revenue: $149.282B
(𝘢𝘴𝘴𝘶𝘮𝘪𝘯𝘨 𝘜𝘚𝘋/𝘊𝘕𝘠 = 𝟩.00)
5Y Revenue CAGR: 10%
2031 Profit Margin: 17%
2031 PE Ratio: 18
Shares outstanding: 2.416B
Shares reduction: 1%/year
VALUATION:
Q2 2031 $BABA SHARE PRICE =
149.282 * (1.10)^5 * 0.17 * 18 /
[2.416 * (0.99)^5] = $320
You can now choose the discount rate that you prefer, for Alibaba I want to use 12%
ACTUAL PRICE: $128
FAIR VALUE: $182
PRICE TARGET (1Y): $203
POTENTIAL UPSIDE: 61%
EXPECTED RETURNS: 20.1%/year
DIVIDEND YIELD: 0.8%
- - - - - - - - - - - - - - -
𝘛𝘩𝘪𝘴 𝘱𝘰𝘴𝘵 𝘪𝘴 𝘧𝘰𝘳 𝘪𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘰𝘯𝘢𝘭 𝘢𝘯𝘥 𝘦𝘥𝘶𝘤𝘢𝘵𝘪𝘰𝘯𝘢𝘭 𝘱𝘶𝘳𝘱𝘰𝘴𝘦𝘴 𝘰𝘯𝘭𝘺.
𝘐'𝘮 𝘯𝘰𝘵 𝘢 𝘧𝘪𝘯𝘢𝘯𝘤𝘪𝘢𝘭 𝘢𝘥𝘷𝘪𝘴𝘰𝘳 and all the 𝘪𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘰𝘯 𝘤𝘰𝘯𝘵𝘢𝘪𝘯𝘦𝘥 𝘪𝘯 𝘵𝘩𝘪𝘴 𝘱𝘰𝘴𝘵 𝘪𝘴 𝘯𝘰𝘵 𝘧𝘪𝘯𝘢𝘯𝘤𝘪𝘢𝘭 𝘢𝘥𝘷𝘪𝘤𝘦.
- - - - - - - - - - - - - - -
Do you own $BABA shares?
Follow me: @nataninvesting
ChaiA.sonic💥 retweeted
$RKLB presents one of the more ambitious vertical integration strategies in aerospace because each business has the potential to strengthen the economics of the next.
Owning launch, spacecraft, components, network infrastructure, and data services creates opportunities to compress development timelines and retain more value inside the organization.
The largest financial benefit comes from reducing the cost of building and operating Iridium rather than recording additional revenue between internal business units.
Every internally designed spacecraft, component, and Neutron launch can improve capital efficiency while giving management greater control over the network roadmap.
That creates a feedback loop where recurring service revenue helps finance future infrastructure, which in turn supports the next generation of products and capabilities.
The investment case ultimately depends on whether Rocket Lab can convert organizational scale into lasting economic leverage instead of simply becoming a larger company.
ChaiA.sonic💥 retweeted
A List of Robotics Stocks:
Humanoids
$TSLA Tesla
$XPEV Xpeng
$AGLT Agility
$XIACF Xiaomi
$HYMTF Hyundai
Sensors
$VPG Vishay Precision
$OUST Ouster
$ARBE Arbe Robotics
$MVIS MicroVision
$AUR Aurora
$CGNX Cognex
$INVZ Innoviz
Chips
$NVDA Nvidia
$AMBQ Ambiq
$AMBA Ambarella
$INDI indie
Logistics
$AMZN Amazon
$SYM Symbotic
Deep Sea Robotics
$KRKNF Kraken Robotics
$OII Oceaneering
Healthcare Robotics
$ISRG Intuitive Surgical
$SYK Stryker
$MDT Medtronic
$PRCT PROCEPT
Industrial Robotics
$HON Honeywell
$TER Teradyne
$LECO Lincoln Electric
Robotics Automation
$PEGA Pegasystems
$ROK Rockwell Automation
$ABBN ABB
$ZBRA Zebra Technologies
$CGNX Cognex
$PATH UiPath
Defense Robotics
$AVAV AeroVironment
$KTOS Kratos
$LMT Lockheed Martin
$NOC Northrop Grumman
$BA Boeing
$GD General Dynamics
Consumer & B2B Robotics
$XIACF Xiaomi
$RR RichTech Robotics
$SERV Serve Robotics
Batteries
$EOSE Eos Energy
$QS QuantumScape
$MVST Microvast
$FLNC Fluence
$KULR KULR
$SLDP Solid Power
$AMPX Amprius
ChaiA.sonic💥 retweeted
📣📣 Meet Qwen-AgentWorld — a native language world model that simulates 7 agent environments (MCP, Search, Terminal, SWE, Web, OS, Android) within a single model. Environment modeling is the training objective from day one, not a post-hoc adaptation.
🤔 LLMs are trained to be better agents — better at acting in environments. But nobody has trained them to model the environments themselves.
🗺️ Our roadmap: investigate how language world modeling can push the boundaries of general agent capabilities, along two routes:
1️⃣ Build a foundation model for environment simulation — outperforming Claude Opus 4.8 and GPT-5.4 on AgentWorldBench
2️⃣ Investigate how world modeling enhances agent training:
🔬 Controllable Sim RL (agentic RL with LWM as environments) surpasses training in real environments
🧠 Learning to predict environments (LWM warm-up) makes agents stronger — remarkably, even without any agent-specific training, this predictive knowledge transfers to agentic tasks with zero fine-tuning
📑 Paper: arxiv.org/abs/2606.24597
📖 Blog: qwen.ai/blog?id=qwen-agentwo…
💻 GitHub: github.com/QwenLM/Qwen-Agent…
🤗 HuggingFace: huggingface.co/collections/Q…
🧩 ModelScope: modelscope.cn/collections/Qw…
ChaiA.sonic💥 retweeted
$NOW is a once-in-a-lifetime opportunity.
2030 price targets:
• Bear case: $315
• Base case: $407
• Bull case: $523
The bull case implies ~22% 5-year FCF CAGR - and it’s achievable through:
• Consistent beat-and-raises through 2030
• Operating leverage through AI efficiency unlocks and revenue scaling faster than expenses
• NDRR >115% + accelerated token consumption monetization
Other key factors:
• CEO buying shares while calling for a $1T valuation by 2030
• Strong Rule of 40 performance (~60%),
• Multiple “massive” new logo deals going live on the Now platform this quarter
Do you think $NOW reaches $500?
ChaiA.sonic💥 retweeted
Alibaba’s AI Moment Is Finally Here And The Market Still Doesn’t Fully Get It
I’ve been covering Alibaba $BABA for years, and honestly, for a long time, the story around this company felt stuck in the mud.
Every quarter looked the same.
Investors obsessed over slowing Chinese consumption, regulatory fears, weak sentiment, and whether Alibaba had permanently lost its identity as a growth company.
The market stopped viewing BABA as a technology platform and started valuing it like a mature retailer trapped inside a struggling macro environment.
But after going through Alibaba’s fiscal Q4 2026 results in detail, I think something important just changed.
Not the headline numbers. Those actually looked terrible at first glance.
I’m talking about the underlying narrative.
For the first time in years, Alibaba is showing clear evidence that its AI investments are no longer just experimental projects or flashy demos designed for investor presentations.
The monetization engine is finally starting to appear in a meaningful way.
And once you dig beneath the ugly earnings headlines, you start realizing why the stock reversed from a sharp premarket selloff to finishing the session up more than 8%.
In my opinion, this was one of the most important quarters Alibaba has reported since the regulatory crackdown era began.
The Headline Numbers Looked Ugly
If you only looked at the surface-level metrics, you would think Alibaba had delivered a disaster.
Revenue grew just 3% year-over-year to RMB243.4 billion, while adjusted EBITA collapsed 84% YoY.
Non-GAAP EPS also fell dramatically and free cash flow turned deeply negative as the company accelerated spending across AI infrastructure, cloud expansion, quick commerce, and user acquisition for its Qwen ecosystem.
Initially, algorithms reacted exactly how you would expect.
The stock sold off aggressively in premarket trading because the quarter looked like a classic margin destruction story.
But that interpretation completely missed what was actually happening underneath the hood.
Alibaba is intentionally sacrificing near-term profitability to build what could become China’s dominant AI-cloud ecosystem.
And unlike previous AI narratives that were built mostly around theoretical future potential, this quarter showed actual commercialization momentum.
This Wasn’t A Retail Quarter It Was An AI Quarter
The real story here wasn’t e-commerce.
It was Cloud Intelligence Group.
Cloud revenue surged 38% YoY to RMB41.6 billion, while external customer revenue accelerated to 40% growth.
More importantly, AI-related product revenue continued posting triple-digit growth for the eleventh consecutive quarter.
That matters a lot more than people realize.
For years, the bearish argument around Alibaba’s cloud business was that growth had structurally peaked.
Investors believed Alibaba Cloud had become little more than a slower-growing infrastructure utility tied to China’s weak economy.
This quarter shattered that thesis.
What we’re seeing now is Alibaba evolving beyond traditional cloud infrastructure into something much more valuable.
The company is transitioning from basic IaaS-style compute and storage into full-stack AI monetization.
And that changes the economics entirely.
Management made it very clear during the earnings call that demand is rapidly shifting toward Model-as-a-Service offerings, enterprise AI agents, orchestration software, inference workloads, and production-scale AI deployments.
This is where the real money in AI infrastructure will likely be made over the next decade.
Anybody can rent storage.
Very few companies can provide an integrated ecosystem combining proprietary LLMs, AI chips, cloud infrastructure, orchestration software, enterprise agents, and consumer distribution at scale.
Alibaba is trying to become all of that simultaneously.
Qwen Is Quietly Becoming A Serious Asset
I think the market still underestimates how important Qwen could become.
For a while, Western investors largely ignored Alibaba’s AI models because the assumption was simple: Chinese AI companies would remain permanently behind OpenAI, Anthropic, and Google.
But that narrative has started breaking apart.
Alibaba’s Qwen models are now showing increasingly competitive performance across reasoning, coding, multimodal applications, and enterprise workflows.
The company also launched specialized models focused on video generation and real-time interactive world models.
More importantly, Alibaba is no longer treating Qwen as a standalone chatbot product.
They’re integrating it directly into their commerce ecosystem.
That’s a major strategic advantage.
Taobao and Tmall services are now integrated into the Qwen app itself, while Alibaba also launched AI shopping assistants and enterprise AI agents designed for merchants.
This creates something Western investors may not fully appreciate yet:
Alibaba owns one of the largest real-world commercial datasets in China.
Every search, every transaction, every merchant interaction, every logistics workflow, every product recommendation, and every consumer behavior signal feeds directly into the ecosystem.
That data advantage becomes incredibly powerful in the AI era.
Amazon figured this out years ago.
Alibaba is now pushing aggressively in the same direction.
The Market Is Missing The Bigger Picture
One thing I keep noticing with Alibaba coverage is that investors are still analyzing this business primarily through a retail lens.
But Alibaba is increasingly becoming an AI infrastructure and platform company.
And if that transformation succeeds, the valuation framework changes dramatically.
At roughly 21x forward earnings, Alibaba is no longer dirt cheap like it was during the peak pessimism phase when it traded closer to 10-12x earnings.
But I still think the stock remains materially undervalued relative to what this business could become over the next several years.
Because if Alibaba successfully scales its AI-cloud monetization engine, the market will eventually stop valuing it like a Chinese retailer and start valuing it more like a strategic AI infrastructure provider.
That multiple difference matters enormously.
The market already gives companies like Microsoft $MSFT , Amazon $AMZN, and Nvidia $NVDA premium valuations because investors understand they sit at the center of the AI compute stack.
Alibaba is trying to build the Chinese equivalent of that ecosystem.
And frankly, they may be further along than many people realize.
Their AI Chip Strategy Is Flying Under The Radar
Another thing that caught my attention this quarter was Alibaba’s growing vertical integration strategy.
Most investors know Amazon has been aggressively building its own AI chips through Trainium and Inferentia.
Alibaba is following a similar playbook.
Its proprietary Zhenwu PPUs are already being deployed across Alibaba Cloud infrastructure, and management disclosed that over 100,000 chips have been deployed on its public cloud platform.
This is important for several reasons.
First, China’s access to advanced Western semiconductors remains uncertain due to export restrictions.
Second, proprietary silicon meaningfully improves long-term cloud economics.
Third, owning both the infrastructure layer and the chip layer creates stronger ecosystem lock-in.
And finally, AI demand itself is becoming so large that every hyperscaler increasingly wants control over its own hardware stack.
That’s not unique to Alibaba.
That’s becoming the new industry standard.
The company also announced partnerships tied to intelligent driving and autonomous systems, which shows Alibaba is expanding beyond generic cloud workloads into higher-value AI compute markets.
Again, this is exactly the kind of ecosystem expansion investors reward in U.S. mega-cap AI companies.
The Profitability Collapse Looks Scary But It’s Probably Temporary
Now obviously, there are real risks here.
You can’t ignore an 84% collapse in adjusted EBITA.
Free cash flow also deteriorated sharply because Alibaba is spending aggressively on AI infrastructure, quick commerce subsidies, and ecosystem expansion.
And yes, the quick commerce business is still pressuring margins.
But when I step back and look at this quarter strategically, I don’t see a collapsing business.
I see a company entering investment mode again.
That distinction matters.
There’s a huge difference between margin compression caused by structural weakness versus margin compression caused by deliberate reinvestment into high-growth opportunities.
Amazon went through this exact phase multiple times.
So did Meta.
Wall Street punished both companies during heavy investment cycles before eventually rewarding them once monetization scaled.
Alibaba now appears to be entering a similar transition period.
The Most Important Signal From This Quarter
The biggest takeaway from this quarter wasn’t the revenue growth.
It wasn’t the EPS miss.
It wasn’t even the cloud acceleration itself.
The most important signal was this:
Alibaba finally proved that AI demand is materially driving customer acquisition and external monetization growth.
That’s the inflection point investors were waiting for.
For years, the AI story around Chinese tech felt speculative.
Now we’re seeing actual evidence that enterprise customers are adopting Alibaba’s AI services at scale.
Management even stated that Model Studio customers grew eightfold year-over-year as enterprises accelerated deployment of AI applications and production-scale workloads.
And once commercialization starts showing up consistently, valuation frameworks begin changing fast.
Valuation Still Looks Attractive To Me
Even after the recent rally, I still think Alibaba offers a compelling long-term setup.
Consensus estimates currently project fiscal 2028 EPS around the high single-digit range, and if Alibaba successfully executes on its AI-cloud transition, I think the market could eventually justify a materially higher earnings multiple than where the stock trades today.
Not because Alibaba suddenly becomes a pure AI company overnight.
But because the mix of the business starts improving.
Higher-margin AI services, enterprise software, cloud infrastructure, inference workloads, and AI agents are structurally more valuable than low-margin retail operations.
And once investors believe those segments can drive sustainable growth again, the re-rating potential becomes significant.
I don’t think the market fully appreciates how large China’s enterprise AI opportunity could become over the next decade.
The demand for AI compute, enterprise automation, inference infrastructure, and industry-specific AI deployment is still in the very early innings.
Alibaba is positioning itself right at the center of that trend.
ChaiA.sonic💥 retweeted
🇨🇳A railway maintenance robot in China.
Detects scratches on rail surfaces down to 0.1mm units while automatically repairing them with a grinding tool at a maximum of 10,000 rotations.
It finds internal defects via ultrasonic flaw detection.
ChaiA.sonic💥 retweeted
Tencent 4Q25 Earnings
- Rev ¥194.4b +13% ↗️🟡
- GP ¥108.3b +19% ↗️🟢 margin 55.7% +314 bps ✅
- Adj EBITDA ¥83b +19% ↗️🟢 margin 42.7% +238 bps ✅
- NG EBIT ¥69.5b +17% ↗️🟢 margin 35.8% +128 bps ✅
- EBIT ¥60.3b +17% ↗️🟢 margin 31.0% +119 bps ✅
- NG Net Inc ¥64.7b +17% ↗️🟢 margin 33.3% +121 bps ✅
- Net Inc ¥59.1b +15% ↗️🟡 margin 30.4% +55 bps ✅
4Q25 Revenue by Segment
- VAS ¥89.9b +14% ↗️🟡
- VAS (Social Networks) ¥30.6b +3% ➡️🟠
- VAS (Domestic Games) ¥38.2b +15% ↗️🟢
- VAS (International Games) ¥21.1b +32% ↗️🟢
- Marketing Services ¥41.1b +17% ↗️🟢
- FinTech and Business Services ¥60.8b +8% ↗️🟡
- Others ¥2.5b +10% ↗️🟡
- Fair Value (listed investee) ¥672.7b
- Fair Value (unlisted investee) ¥363.1b
Biz Metrics
- MAU (Weixin and Wechat) 1.42b +2% ↗️🟡
- Mobile MAU QQ 508m -3% ↘️🟠
- Fee-based VAS Subscriptions 267m +2% ↗️🟡
- Time Spent (Mini Programs) +20% ↗️🟢
FY25 Earnings
- Rev ¥751.8b +14% ↗️🟡
- GP ¥422.6b +21% ↗️🟢 margin 56.2% +332 bps ✅
- Adj EBITDA ¥336.4b +21% ↗️🟢 margin 44.8% +280 bps ✅
- NG EBIT ¥280.7b +18% ↗️🟢 margin 37.3% +131 bps ✅
- EBIT ¥241.6b +16% ↗️🟢 margin 32.1% +61 bps ✅
- NG Net Inc ¥259.6b +17% ↗️🟢 margin 34.5% +81 bps ✅
- Net Inc ¥229.8b +17% ↗️🟢 margin 30.6% +81 bps ✅
- OCF ¥303.1b +17% ↗️🟢 margin 40.3% +108 bps ✅
- FCF ¥182.6b +16% ↗️🟢 margin 24.3% +45 bps ✅
FY25 Revenue by Segment
- VAS ¥369.3b +16% ↗️🟢
- VAS (Social Networks) ¥127.7b +5% ↗️🟡
- VAS (Domestic Games) ¥164.2b +18% ↗️🟢
- VAS (International Games) ¥77.4b +33% ↗️🟢
- Marketing Services ¥145b +19% ↗️🟢
- FinTech and Business Services ¥229.4b +8% ↗️🟡
- Others ¥8.1b +4% ↗️🟡
1 | VAS grew +14% dragged by slower growth in social network +3%, long-form video +1%, offset by faster growth in Music Subscription +13%, Domestic Games +15% (Delta Force, Valorant, and Wuthering Waves) and even faster growth in International Games +32% (Supercell, PUBG Mobile, and Wuthering Waves).
Value-added service revenue was CNY 90 billion renminbi, up 14% YoY. Our social network revenue grew 3% YoY to CNY 31 billion renminbi, driven by increased revenue from Video Accounts, live streaming, and from music subscriptions. Music subscription revenue increased 13% YoY on ARPU and subscriber growth. Long-form video subscription revenue increased 1% YoY as video subscribers grew slightly year-on-year, benefiting from the drama series Love's Ambition, the variety show Natural High: Season 3, and the animated series Renegade Immortal. Each of these ranked first by video views in their respective genres across all video platforms in China for the quarter. Domestic games revenue grew 15% YoY, primarily driven by Delta Force, the Valorant franchise, and Wuthering Waves. International games revenue increased 32% YoY, primarily driven by Supercell's titles, PUBG Mobile, and Wuthering Waves.
2 | Continue to strengthen Weixin’s commerce with Mini Shops and Mini Programs, saw user time spent grow +20%
Moving to communications and social networks, we strengthened Weixin's commerce experience by upgrading features for users and tools for merchants in the Mini Shops. The upgraded e-commerce gateway page allows users to check their shopping carts, see what friends are recommending, and receive notifications from their favorite shops, and generated substantial GMV during the quarter. Through the new Likes for Discounts feature, users can discover products liked by friends and receive and share discounts via the e-commerce gateway page, chats, and moments.
For Mini Programs, total user time spent increased over 20% year-on-year, driven by workplace productivity tools, mini-games, and novels. We added Tencent CodeBuddy to our developer toolkit, enabling developers to create mini-programs using natural language input, and we provided developers of AI native mini-programs with free compute resources.
3 | Marketing was strong, saw more time spent with Video Accounts, better conversion rates in Mini Shops, and more engagement in Mini Programs.
Entering 2026, we have deepened collaboration with e-commerce platforms, facilitating their merchants advertising within Tencent, and we've increased the inventory for rewarded video ads and Video Accounts, which have contributed to faster year-on-year marketing services revenue growth in the first quarter to date versus in the fourth quarter of last year. At a product level, Video Accounts total time spent increased due to upgrades to the content recommendation algorithm, enabling faster growth in ad impressions while our ad load remained lower than peers. Better conversion rates contributed to more marketing spending for Mini Shops merchants. For Mini Programs, consumers engaging more with mini-games and mini-dramas attracted more marketing spend from the mini-game and mini-drama studios. Weixin Search overall query volume grew at a rapid rate due to AI enhancements to search results, driving growth in commercial query volume, while search pricing also increased.
4 | Fintech remained robust driven by commercial payment and wealth management.
On fintech and business services, segment revenue was CNY 61 billion, up 8%. We grew fintech services revenue by a single-digit percentage year-over-year, and fintech gross profit at a higher rate driven by wealth management and commercial payment services. Commercial payment volume sustained positive year-on-year growth, supported by a higher number of transactions and a narrowed decline in value per transaction. For wealth management, which is the second biggest contributor to fintech revenue, average assets per user and number of users each increased year-on-year.
5 | Saw strong gross profit margin improvement across all business segments.
Overall gross margin was 56%, up 300bps YoY by segment. VAS gross margin was 60%, up 400bps YoY, primarily driven by greater contributions from internally developed high-margin games. Marketing services gross margin was 60%, up 200bps YoY as AI-powered marketing services drove strong growth in high-margin revenue streams, particularly Video Accounts and Weixin Search. Fintech and business services gross margin was 51%, up 400bps YoY,, benefiting from growing scale of cloud services and improved revenue mix in Fintech services alongside enhanced cost efficiency.
6 | Tencent’s franchises are resilient in the age of AI, have network effects, deep supply chain integration, stringent regulatory and licensing requirements, scarce or unique resources, sticky, and private data that is closed and interactive in nature.
First of all, I would like to talk about our key franchises are very resilient in the age of AI. As we know, AI will affect every part of the technology industry, but some products and services are inherently more resilient than others. We believe that some of the characteristics of resilience would include network effects arising from consumer to consumer to content creator, and consumer to business interactions in descending order of strength. That's number one. Number two, deep supply chain integration linking the worlds of bits with the world of atoms. Number three, stringent regulatory and licensing requirements. Number four, scarce or unique resources, including physical and intellectual properties. Number five, tick rates that are low compared to value provided or cost of switching. And number six, private data that is closed and interactive in nature.
7 | Thus see communication services as continuing to remain extremely sticky against non-AI services and resilient against AI-based services.
We believe this need for human interaction, together with the network effects and closed nature of the data arising from these interactions, have resulted in communication services being extremely sticky in the face of competing non-AI services in the past and will continue to be resilient versus AI-based services in the future.
8 | Network effects also apply to their multiplayer games.
Moving on to our games. They are also very resilient as our multiplayer games, especially PVP games, also enjoy network effects. Similar to sports, they are team-based in nature, and players play with and against other players. Just as people prefer to participate themselves or watch the teams they support compete in sports rather than watching AI sports, game players continue to enjoy the interaction with other humans that our games provide. Our games also cultivate strong IPs.
9 | While AI will enable more games to be made faster, the highest quality and the most innovative will be the best, with more time, more games will be played.
While AI will enable more games to be made faster, the game industry is already in a position of excess supply, with 200,000 new games on mobile and 18,000 new games released on Steam every year. The limiting factor is that new games need to be high quality and more innovative than the best existing games, which in turn requires human creativity on top of cutting-edge technology. Game is a natural beneficiary of AI proliferation, also when people have more time at hand.
10 | Fintech services are also resilient to AI as licenses are limited present barriers to entry, and the payment network with partner banks is hard to replicate, and they have the lowest mobile payment take rates in the world.
Our fintech services are also resilient as they depend on difficult to secure and retain licenses which are limited in nature and also set the boundary on how innovations can be introduced in an industry. We have also invested decades building a payment network of difficult to replicate rails into partner banks, merchants, and connecting them with more than 1 billion consumers, which brings its own network effects. Our mobile payment take rates are already among the lowest in the world, which we believe makes competing with us on price highly uneconomical.
11 | Using AI to accelerate in-game content production, and realistic game play, and with better content creation and ad targeting in marketing.
In games, we are deploying generative AI to accelerate in-game content production, enabling us to produce more content within our big games. We're using generative AI to facilitate new user acquisition and existing user retention through measures such as targeted ads and personalized daily highlight reels. We're enriching the core gameplay experience with AI features such as virtual teammates in PVP games and realistic non-player characters in PVE games
12 | Using AI In marketing to serve more relevant and better ads, create more ads to smaller subset of users more efficiently,, allowing marketing revenues to grow at 19% faster than 14% for the industry.
For marketing services, we scaled up our advertising foundation model to provide more relevant ads to more targeted users, boosting ad conversions for advertisers and providing better user experiences at the same time. We provide generative AI-powered ad creative solutions, enabling advertisers to create more ads which are more relevant to smaller set of users and more efficiently. We introduced our automated ad campaign solution, AiM Plus, under which advertisers can automate targeting, bidding, and placement, improving their return on marketing investments and increasing their budget allocation to us. These initiatives contributed substantially to Tencent's marketing services revenue growth of 19% in 2025, outstripping the overall China ad industry growth of 14%.
13 | Video Accounts is now the 2nd largest by DAU in China (after Douyin), and time spent grew by more than 20% in 2025, AI improving content production, distribution, and recommendation.
For Video Accounts, deploying a longer sequence AI model which captures more of a user's signals to enhance content recommendation is boosting user growth, engagement, and content distribution. Total time spent on Video Accounts increased more than 20% in 2025, and Video Accounts is now the second-largest short video service by DAU in China. For digital contents, we utilize AI in content production, improving production workflow efficiency, and providing visually compelling special effects. AI also helps in content distribution through more intelligent content recommendations across music, videos, and literature.
14 | Now using more AI in enterprise software like note-taking, and lightweight AI models to enhance credit-scoring processes and facilitate fraud detection.
We're using AI in enterprise software to provide features such as AI agents that can take notes on and summarize concurrent meetings for users, and AI agents that generate intelligent summaries of customer service history for merchants. Our enterprise software products, WeCom and Tencent Meeting, are leaders in their categories in China in terms of usage and revenue.
For Fintech, we utilize lightweight AI models to enhance credit scoring processes and facilitate fraud detection, contributing to us sustaining better than industry non-performing loan rates. Now that our core businesses are benefiting operationally and financially from integrating AI, we believe we are in a position of strength to add development of new AI products to our priorities.
15 | While Tencent was not a first-mover in LLMs, they are pushing hard with HunYuan 3.0, which is going to significantly better in capabilities and multi-modal.
Although we're not the first mover in large language models, having already revamped our team, improved our data quality, and rebuilt our AI infrastructure for pre-training and reinforcement learning, we're now iterating more intelligent models at a faster pace. HunYuan 3.0 is in internal testing and currently represents a bigger step in capabilities versus HunYuan 2.0 than 2.0 was versus 1.0. For multimodal capabilities, our 3D text-to-image and world models are early category leaders and will increasingly benefit from leveraging our proprietary data and abundant use cases.
16 | Don’t see AI chatbots as the only means to bring AI to users, as it can come through many ways, AI chatbots are largely competing with search rather than every other application, focusing on find PMF and use cases instead.
Some observers in Chinese tech are single-mindedly focused on AI chatbots as the only means for bringing AI to users. We believe this mindset is overly simplistic because AI can help people in a multitude of ways beyond powering an information advice app. We believe that AI chatbot applications are largely competing with search applications rather than with every other application. For Yuanbao, our own AI chatbot app, we're focused on finding product market fit and use cases which belong in chatbot AI app. We're rapidly iterating Yuanbao to enhance its user experience by providing better search integration, improved speech recognition, easier access to multimodal capabilities, and exploration around group chat, which we believe will increase usage and user retention of the app.
17 | OpenClaw allows AI to be decentralized not via a single point, the browser, or the chatbot, but now allows many companies to have their Claws with different models.
I think OpenClaw is actually a very exciting concept, right? You know, it actually sort of presents a decentralized model or a decentralized regime for, you know, how AI works in this world. You know, so there's some parallel to sort of how the internet evolves, right? You know, in the very beginning when internet first appears, right, you know, there seems to be sort of, you know, there is one entry point, which is the browser. You know, there is sort of one distribution point, which is the search engine. Over time, you know, it, you know, there are different services which evolve, right?
The chatbot being sort of, you know, the single entry point. Now with Claw, you can see, you know, it opens up a completely decentralized regime where, you know, many companies can have their own Claw, and the Claw can be using all kinds of different models, right?
18 | Excite with OpenClaw and see opportunities with personal agents, improving HunYuan will allow them to offer new unique agentic capabilities through Weixin and QQ
The excitement around OpenClaw illustrates that people recognize AI can unlock computer use capabilities to improve their daily lives but also illustrate the risks around unleashing unsupervised AI. We want AI agents in Weixin to deliver AI productivity that's beneficial to the general public as well as early adopters, and which will boost ecosystem activity and naturally generate revenue. AI agents are currently powered by a multiplicity of foundation models, and we expect that users at the application level will continue to have access to a range of models. However, improving the performance of HunYuan will enable us to offer new, unique to Weixin agentic capabilities.
Users control this new generation of AI tools through command line interfaces in their existing communication apps, which generally means Weixin and QQ in China, as it's the most efficient for users to interact with digital agents in a place and format where they are already interacting with human contacts. The new AI products that I described above require substantial and increasing investment, which we believe will generate significant return for us over the long run.
19 | Spent CNY 7bn and CNY 18bn on HunYuan and Yuanbao in 4Q25 and FY25, and expect to 2X these investments in 2026, and these exclude external GPUs via Tencent Cloud, and will be funded from earnings.
Our spending on our two biggest new AI products, HunYuan and Yuanbao, was CNY 7 billion in the fourth quarter of 2025 and CNY 18 billion for the full year. These figures are only for HunYuan and Yuanbao and exclude AI initiatives supporting our existing products and services, as well as exclude costs arising from providing GPUs to external customers via Tencent Cloud. We expect to more than double these investments in HunYuan, Yuanbao, and other new AI products in 2026, which we intend to fund from increasing earnings from our core businesses.
20 | Breaking out these investments, as they see them as upfront lumpy rather than ongoing recurring expenses and thus prefer to separate the reporting, confident of monetization of the usage from these new AI products.
In this transformational period, we are breaking out our investment in new AI products because we view these strategic investment conceptually similar to investment in an affiliate or to CapEx. These are upfront investments required to build the necessary foundation to unlock new value as opposed to ongoing operating expenses. As such, we believe the impact of these investments should be viewed separately from the profits generated by our existing businesses. Over time, we're confident that monetization will follow usage for these new AI products.
21 | Expects Tencent Cloud, which is #2 to follow games, payments, and long-form video to be #1, and similarly, its new AI products. Tencent Cloud’s choice to focus on high-quality rather than high-revenue, low-value-add activities has allowed it to achieve EBIT breakeven in 2024 and RMB 5bn EBIT profitable in FY25.
Lastly, I would like to present a case study on Tencent Cloud as the latest example on how we develop our services into market leaders with economic returns over time. That would follow games, payments, and long-form video. We expect it will be the same for our new AI products. Tencent Cloud was a relative late entrant in cloud services. However, we committed to a patient and long-term investment strategy, believing that it had scale from the start due to Tencent itself being the biggest single end user for a range of technology infrastructure in China, and that it could provide differentiated services arising from Tencent's unique insights, ecosystem, and capabilities.
After a period where Tencent Cloud prioritized the revenue growth somewhat misguided by other industry participants, in 2022, we aggressively restructured Tencent Cloud to focus on high-quality services rather than chasing high revenue but low-value-added activities such as reselling and customizing projects. This pivot cost us several quarters of revenue growth, but it enabled Tencent Cloud to achieve operating profit breakeven in 2024, up from significant losses in prior years. During 2025, although Tencent Cloud continued to face revenue headwinds due to limited availability of GPU for external customers as we prioritize our internal needs, it grew revenue and sharply improved earnings, achieving RMB 5 billion adjusted operating profit.
22 | Tencent Cloud is growing faster and higher compute will drive even faster revenue growth, initial losses were incubating the business that will generate strong profitability.
In recent months, we're seeing a better pricing environment, especially for memory and CPU, which, along with robust AI demand and overseas expansion, allowing Tencent Cloud to grow revenue at a faster rate. Moving through the year, we have ordered a substantially higher volume of compute, which should also facilitate revenue growth.
Overall, we think Tencent Cloud is becoming another example of how Tencent competes on our own terms and pace and how our incubation investment cycle works. We view the initial losses in Tencent Cloud as a fixed sum of cash investment necessary to incubate a successful new business, but ultimately generating good economic returns. We view the initial investment in new AI products, code-named MAP, in the same way.
23 | China Cloud Service Providers struggled historically with low margins because they can source directly, but now it is becoming a monopsony, where the suppliers are priortizing the biggest and most regular customers, the hyperscalers, and finally can raise prices and be profitable.
for years the industry has suffered because the cloud services providers in China were operating at very low margins. One of the reasons they operated at very low margins was because, you know, if there was a new entrant or if the customers wanted to source infrastructure directly, they were able to telephone the supplier and, you know, order the infrastructure that they wanted from the supplier of, you know, CPU or GPU or DRAM. You know, that's no longer the case. You know, now, the supply is booked out months, quarters, in some cases, years in advance. You know, the supplier is prioritizing the biggest, most regular customers, which are the hyperscalers such as ourselves.
Therefore, you know, the smaller cloud providers no longer have certainty that they can source supply, and they need to come to the hyperscalers. You know, the hyperscalers have been operating at low margins and so, you know, when the demand picks up, then, you know, we almost sort of as an industry have no choice but to pass through higher prices. You have seen a number of price increases in China cloud in the last 24 hours as a result. In terms of, you know, how we sort of value capture, you know, in this dynamic environment, then, you know, one broad principle is, you know, we seek to deliver, you know, more value through, you know, enrichment.
24 | Tencent chose not to focus on low-value, low-margin bare metal compute but higher-value, higher-margin compute instead.
Enrichment means that, you know, at a minimum, if you have, you know, compute, you can rent it out bare metal and you get a certain low price and low margin. You know, preferably you rent out. You subdivide it and virtualize it into tokens, and then you get a higher price and higher margin per unit of compute. Ideally, you bundle it into a platform as a service or software as a service. Then you can get, you know, the best pricing and the best margins. That's part of the journey that we've been on, and that's part of, you know, how Tencent Cloud has moved from a very substantial losses four years ago to pretty substantial profits last year. We'll continue on that journey of, you know, moving from bare metal to token to platformization and to software.
ChaiA.sonic💥 retweeted
ROIC is one of the biggest drivers of quality factors.
But yet it is not just high ROIC alone that one should seek, but also if it can keep rising and stay high, and if there are sufficient opportunities to reinvest.
visioninvesting.substack.com…
Nice chart shared by @ReneSellmann.
This is my favorite chart from last year ⬇️
In a nutshell, clearly not all "Quality" is created equal.
Since 1985, the "most alpha" has been found in two places:
✅ Capital Efficiency: ROIC, ROE & ROA
✅ Solvency: Interest & Debt Coverage
Buying "High Gross Margin" sounds sexy, but the market rewards those who actually generate returns on their capital.
ChaiA.sonic💥 retweeted
Chinese battery makers had 70% of the global EV battery market in 2025. South Korean battery makers are struggling to remain profitable.
Top 5 battery makers:
1. 🇨🇳 CATL 39.2%
2. 🇨🇳 BYD 16.4%
3. 🇰🇷 LG Energy 9.2%
4. 🇨🇳 CALB 5.3%
5. 🇨🇳 Gotion 4.5%
Chart: asia.nikkei.com/business/aut…
ChaiA.sonic💥 retweeted
Just published my latest deep dive:
Alibaba, Meituan and JD's Quick Commerce War and How Grab and Sea Will React
Co-written with @the_zack_zhu after months of work. China is the clearest preview of what’s coming to Southeast Asia and understanding that dynamic is crucial.
