AI Industry Daily Radar · Monday, July 20, 2026
Executive Summary
- Alibaba previewed Qwen3.8 Max, a 2.4-trillion-parameter model it ranks second only to Anthropic's Claude Fable 5 — the latest salvo in China's accelerating push to close the frontier-model gap.
- Moonshot AI told investors it is preparing a Hong Kong IPO within six months, targeting a $30 billion valuation on the strength of its Kimi K3 open-weight model.
- The European Commission issued binding Digital Markets Act orders forcing Google to open Android to rival AI assistants and share anonymized search data with competitors.
- Databricks signed a $3 billion funding term sheet at a $188 billion valuation, a 40% jump in six months driven by enterprise AI demand.
- SAP closed its acquisition of Prior Labs for over €1 billion, betting that tabular foundation models become as important as chatbots for enterprise data work.
- Google's Gemini 3.5 Pro missed its launch target for the third consecutive time, with reports of a stopgap Gemini 3.6 Flash under consideration.
Top Stories
1. Alibaba Previews Qwen3.8 Max, Claims Second-Only-to-Claude Performance
Alibaba Group Holding previewed Qwen3.8 Max, its largest model to date at 2.4 trillion parameters, during a weekend launch that pushed the company's shares up as much as 5.4% on the Hong Kong Stock Exchange. Alibaba positioned the preview as second only to Anthropic's Claude Fable 5 across overall capability benchmarks, a claim that places it ahead of offerings from OpenAI, Google, and Meta in the company's own evaluation.
The preview version is available through Alibaba's Token Plan subscription. Full weight release details and a complete model card have not yet been published. Both of the week's highest-profile model announcements — this one and Moonshot's Kimi K3 — came from Chinese labs.
Source
2. Moonshot AI Targets $30 Billion Hong Kong IPO Within Six Months
Moonshot AI has told investors it is preparing to list in Hong Kong as early as six months, seeking a valuation around $30 billion, according to Bloomberg and The Information. The company formally initiated the investor-approval process for the IPO this past weekend, shortly after its Kimi K3 model topped a major coding leaderboard and disrupted the U.S. frontier-model market.
Moonshot's last reported private valuation was approximately $5 billion. A successful listing at $30 billion would represent a six-fold increase and rank the company among China's most valuable AI firms. Kimi K3's open-weight release on July 27 — with weights available for free — is expected to serve as a credibility marker for institutional investors evaluating the listing.
Source
3. EU Orders Google to Open Android to Rival AI Assistants and Share Search Data
The European Commission issued two sets of legally binding orders under the Digital Markets Act requiring Google to give third-party AI assistants the same access to Android features that Gemini currently enjoys. The orders cover 11 Android feature groups, including voice activation, cross-app actions, and suggested replies. Rival assistants must be able to activate via voice commands and operate within apps, subject to certification and user consent.
A second order requires Google to share anonymized search ranking, query, click, and view data with competitors on fair, reasonable, and non-discriminatory terms. Search data sharing is scheduled to begin in January 2027; Android interoperability changes are due by July 2027. Google policy chief Kent Walker argued the decisions risk weakening privacy and security protections for European users.
Source
4. Kimi K3: Moonshot's 2.8-Trillion-Parameter Open Model Tops Coding Leaderboard
Moonshot AI released Kimi K3, a 2.8-trillion-parameter model that ranks as the largest open-weight model ever announced. K3 took the top spot on the Frontend Code Arena leaderboard, placing ahead of both Claude Fable 5 and GPT-5.6 Sol Max in human-preference coding rankings. On BrowseComp, a long-horizon information-seeking benchmark, K3 scored 91.2 out of 100 — a state-of-the-art result achieved in a single-agent setup using its 1-million-token context window.
K3 is available now through kimi.com and an OpenAI-compatible API priced at $3 per million input tokens and $15 per million output tokens, with automatic context caching at $0.30 per million tokens. Full open weights are scheduled for release on July 27. K3 ranks approximately ninth on general conversation benchmarks — a coding and agent specialist, not an all-rounder.
Source
5. Databricks Raises $3 Billion at $188 Billion Valuation
Databricks signed a term sheet for a $3 billion strategic funding round led by Coatue Management, pushing its valuation to $188 billion — a 40% increase from its $134 billion Series J just six months ago. The company said the new capital will drive AI innovation across its Unity AI Gateway, Genie, and Lakebase product lines, and will be used to purchase additional GPU capacity.
The round is expected to close later this summer. Databricks offers both proprietary and open-source AI models on its platform and has positioned itself as the data infrastructure layer for enterprise AI deployments. At $188 billion, it ranks among the most richly valued private software companies in the world.
Source
6. SAP Closes Prior Labs Acquisition in €1 Billion Tabular AI Bet
SAP completed its acquisition of Prior Labs, a Freiburg-based startup founded roughly 18 months ago that pioneered tabular foundation models. SAP committed to invest more than €1 billion over four years to scale Prior Labs into a globally leading European AI lab. According to Dealroom, SAP paid approximately €700 million upfront with a €300 million earnout.
Prior Labs will operate as an independent entity within SAP. Its flagship product, TabPFN — a foundation model designed for structured data in spreadsheets and databases — was published in Nature and set state-of-the-art results on tabular benchmarks. SAP is betting that structured-data AI becomes a distinct category alongside text and image models.
Source
https://tech.eu/2026/07/17/sap-acquires-prior-labs-just-18-months-after-launch-in-eur1b-deal/
7. Gemini 3.5 Pro Misses Third Launch Target as Google Weighs Stopgap Release
Google DeepMind's Gemini 3.5 Pro missed its July 17 launch target, the third consecutive delay for the model first previewed at I/O 2026 in May. The original June release window passed, then a revised July target slipped, and the model remains unavailable with no official model card, pricing, or benchmark results published. Alphabet shares fell approximately 4% on the news.
Reports indicate Google is now exploring a stopgap Gemini 3.6 Flash release to field something new while the Pro model's issues are resolved. Three consecutive delays point to a deeper training or quality-bar problem rather than standard engineering caution. Every week of absence pushes enterprise customers toward GPT-5.6, Claude, Grok, and the newly released Kimi K3.
Source
https://9to5google.com/2026/07/16/gemini-3-5-pro-delays/
Industry Trends
Trend 1: Chinese Labs Are Closing the Frontier Model Gap in Real Time
Two of the week's biggest model stories — Alibaba's Qwen3.8 Max and Moonshot's Kimi K3 — both came from Chinese labs. Kimi K3 topped a major coding leaderboard ahead of Claude Fable 5, and Alibaba claims Qwen3.8 Max trails only that same Anthropic model. These are vendor self-assessments that warrant independent verification, but Chinese models are no longer six months behind Western frontier systems. Moonshot's planned $30 billion IPO valuation, up from $5 billion, shows that capital markets are pricing in this shift.
Trend 2: Regulators Are Targeting Distribution, Not Just Models
The EU's DMA orders against Google attack two moats that no competitor could replicate: pre-installed Gemini on billions of Android devices and two decades of search behavior data. This is a different regulatory approach from model-level safety rules or compute caps. By forcing interoperability and data sharing, Brussels is structurally altering the distribution layer that determines which AI assistants reach consumers. Other jurisdictions will be watching whether the January 2027 and July 2027 deadlines hold.
Trend 3: Capital Is Concentrating Around AI Infrastructure and Vertical Foundation Models
Databricks' $188 billion valuation and SAP's €1 billion Prior Labs acquisition share a thesis: the biggest enterprise AI value may sit in specialized infrastructure and vertical-specific foundation models rather than general-purpose chatbots. Databricks is buying GPU capacity and building the data layer. SAP is betting on tabular models for structured business data — a category with no direct chatbot equivalent. The capital flows suggest investors are diversifying away from pure frontier-model bets toward the picks-and-shovels layer.
Featured AI Products
Microsoft Project Perception
Project Perception is an AI cybersecurity platform under development at Microsoft that uses multi-model routing to find and fix software vulnerabilities across code, cloud infrastructure, and endpoints. It sends simple tasks like log parsing and basic triage to cheaper models, reserving frontier models from OpenAI and Anthropic for complex exploit reasoning and live remediation. This routing approach makes always-on vulnerability scanning affordable enough to run continuously, a direct challenge to Anthropic's expanding security business.
Official URL: https://www.microsoft.com/security
TabPFN (Prior Labs / SAP)
TabPFN is a tabular foundation model built for structured data — the spreadsheets, databases, and tables that underpin most enterprise operations. Unlike text-focused LLMs, TabPFN is pre-trained to make predictions on tabular datasets without task-specific fine-tuning. Published in Nature, it set state-of-the-art results on standard tabular benchmarks. Now owned by SAP and backed by a €1 billion four-year investment, it targets a category that has no established incumbent.
Official URL: https://www.prior-labs.com
Kimi K3 (Moonshot AI)
Kimi K3 is a 2.8-trillion-parameter model with a 1-million-token context window, natively multimodal, and built for long-horizon coding and agentic tasks. Its Kimi Delta Attention and Attention Residuals architecture delivers up to 6.3× faster decoding in long contexts. The API is OpenAI-compatible at $3 per million input tokens, and full open weights release on July 27 will make it free to download and self-host — a first for a model at this scale.
Official URL: https://kimi.com
Key Takeaways
- Chinese AI labs released two of the week's most significant models. Alibaba's Qwen3.8 Max and Moonshot's Kimi K3 both claim frontier-level coding performance, and Moonshot is converting that momentum into a $30 billion IPO target.
- The EU's binding DMA orders against Google structurally alter the AI distribution layer in Europe, opening Android to rival assistants and mandating search data sharing with competitors by 2027.
- Enterprise AI capital is concentrating fast: Databricks at $188 billion and SAP's €1 billion Prior Labs deal show investors backing infrastructure and vertical foundation models over general-purpose chatbot plays.
- Google's triple Gemini 3.5 Pro delay is creating a competitive opening that rivals are actively filling — Kimi K3's free July 27 weight release is timed directly into that gap.
- The highest-valued private AI companies are now the ones providing compute, data infrastructure, and specialized models — not consumer-facing chat.