Daily Tech & GNSS News Digest - July 17, 2026

Friday belonged to Beijing’s one-two punch: Moonshot AI dropped the largest open-weight model in history overnight, and Xi Jinping personally opened the World AI Conference hours later with a 29-nation governance body headquartered in Shanghai—models and rules, launched within the same 24 hours. The chip selloff the model triggered pushed the semiconductor index into bear-market territory, while OpenAI quietly killed its browser to bet everything on ChatGPT as a single super app. On the positioning side, two pieces of next-generation infrastructure moved forward: Xona’s LEO constellation got a certification program for the devices that will use it, and Europe’s upgraded satellite-augmentation system passed its first real-time test.

Tech News

Moonshot’s Kimi K3 Is the Largest Open Model Ever—and It Deepened the Chip Rout

Moonshot AI launched Kimi K3 late Thursday, a sparse mixture-of-experts model with roughly 2.8 trillion total parameters and a 1-million-token context window—the largest open-source-track model ever released, per VentureBeat and MarkTechPost. The model ships in two variants at $3 per million input tokens and $15 output, with open weights promised by July 27; early independent estimates place it in the same tier as top closed US models. The architecture is new, not just bigger—built on Kimi Delta Attention with native vision and a tunable reasoning-effort control.

Markets read a Chinese frontier-class model as a threat to the AI trade’s margins: Bloomberg reports the chip rout deepened after the launch, with the Nasdaq 100 closing down 1.5% for its worst week in almost a month and the Philadelphia Semiconductor Index falling 1.6%—now 20% off its June record, meeting the definition of a bear market. When the K3 weights go free in ten days, every closed lab faces the question at 2.8 trillion parameters: what exactly justifies the premium?

Xi Jinping Opens the World AI Conference—With a 29-Nation Governance Body

Chinese President Xi Jinping delivered the opening keynote of the 2026 World AI Conference in Shanghai on Friday—his first appearance since the event began in 2018—a day after 29 countries signed the agreement establishing the World Artificial Intelligence Cooperation Organisation (WAICO), a new intergovernmental body headquartered in Shanghai, per Bloomberg, CNBC, and Caixin. Founding members skew Global South—Indonesia, Brazil, Malaysia, South Africa, Senegal, Russia, and Pakistan among them—and UN Secretary-General António Guterres attended the launch.

Xi pitched AI development as “a symphony of international cooperation,” framed China’s open-source models as a public good, and pledged 5,000 AI training opportunities for developing countries over five years, per CNBC and Al Jazeera. Across town, Huawei showcased its Atlas 950 SuperPoD domestic AI computing cluster on the WAIC floor. The package—chips, models, and now rules, all made in China—lands the same week Washington has no equivalent governance proposal on the table, and the Kimi K3 launch the night before the speech was not a coincidence of calendars.

OpenAI Kills Its Atlas Browser to Bet Everything on the ChatGPT Super App

OpenAI discontinued Atlas, its AI web browser, consolidating browsing, work tools, voice, and agents into ChatGPT as a single super app, per Build Fast with AI’s roundup. Atlas launched as OpenAI’s bid to own web navigation; its shutdown says the company now believes the chat app itself replaces the browser as the front door to the internet—a conclusion Microsoft reportedly reached the same week, opting for deeper ChatGPT integration over building a rival browser.

The discipline angle matters as much as the strategy: OpenAI is weeks from an expected IPO filing, and shutting down a product that wasn’t winning—rather than letting it limp along for optics—is what companies do when they start answering to public markets. A year ago every lab wanted its own browser; now the fight is over which single app users open first each morning.

Additional Headlines:

Update: Gemini 3.5 Pro is expected to launch today—after a full rebuild: Reporting this week explains the delay covered here Thursday: Google scrapped the original base model entirely and restarted pretraining after finding structural failures in recursive tool-calling, per TechTimes. Google has not officially confirmed the date, specs, or pricing.
Mira Murati’s Thinking Machines released Inkling: The former OpenAI CTO’s first major public release is an open-weight model anyone can download and fine-tune—an American marquee founder joining the open-model surge, per Build Fast with AI.
NotebookLM is now Gemini Notebook—and it got a cloud computer: Google’s rename adds a secure cloud environment that runs code inside notebooks for the tool’s 30 million-plus users, with Google Search integration coming, per TechCrunch and Google’s announcement.
Netflix tumbled on slowing growth: Shares fell after the streamer warned of slowing sales growth for a second straight quarter, per Bloomberg.
Microsoft’s record Patch Tuesday fixed 570 flaws—found partly by AI: The largest single patch release in company history, with internal AI systems credited for identifying and prioritizing a significant share, per Build Fast with AI’s roundup.

GNSS News

Xona Launches Pulsar Verified, With Safran’s Skydel Simulators as First Members

Xona Space Systems launched Pulsar Verified, a certification program for device manufacturers, chipset makers, and OEMs building products for its planned LEO navigation constellation—and Safran Electronics & Defense’s Skydel simulation engine, along with all Skydel-based simulators, is among the first to complete the technical verification, per GPS World and Inside GNSS. The program confirms that hardware can receive and operate with the Pulsar signal, which Xona is building to deliver centimeter-level precision and authenticated protection against spoofing, with signals up to 100 times stronger than GPS from low Earth orbit.

The simulator-first sequencing is the practical part: Skydel can now generate high-fidelity simulations of Xona’s full LEO constellation alongside legacy GPS and Galileo in one integrated tool, letting receiver makers design and test Pulsar-compatible products before the constellation itself is fully deployed. After a string of Xona partnerships—Trimble on corrections, Furuno on timing, Murata on industrial modules—a formal certification ecosystem is the sign a constellation operator gives when it’s preparing for real customers rather than demos.

Airbus Shows Europe’s Next-Gen EGNOS Working in Real Time

Airbus demonstrated real-time dual-frequency, multi-constellation (DFMC) corrections for EGNOS V3, the next generation of Europe’s satellite-based augmentation system that will augment both GPS and Galileo signals, per Inside GNSS. With V3’s dedicated ground stations not yet available, the Airbus team processed live data from public IGS and EUREF reference stations, generated SBAS corrections in real time, and achieved average vertical positioning error of around 0.3 meters—then went further, uplinking the corrections to a geostationary satellite and broadcasting them on the GPS L5 frequency to user antennas across Europe.

The current EGNOS safety-of-life service has supported LPV-200 precision approaches since 2011; V3 keeps that and layers the DFMC service on top, aimed squarely at aviation’s most demanding phases—approach and landing in low visibility. In a month dominated by jamming bulletins and GPS-backup announcements, this is the quieter half of the resilience story: making the primary civil navigation layer itself more accurate and harder to degrade, using two frequencies and two constellations instead of one of each.

Key Takeaways

China launched models and rules within the same 24 hours: Moonshot’s Kimi K3—2.8 trillion parameters, open weights by July 27—dropped the night before Xi Jinping’s first WAIC keynote and the debut of a 29-nation, Shanghai-headquartered AI governance body, a coordinated demonstration that Beijing competes at every layer of the stack.
The AI trade keeps repricing around open models: K3’s launch deepened a chip rout that put the semiconductor index 20% below its June record—bear-market territory—while OpenAI’s shutdown of its Atlas browser shows even the leading lab consolidating around what wins ahead of public-market scrutiny.
Next-generation navigation infrastructure is moving from plans to proof: Xona’s Pulsar Verified program—with Safran’s Skydel simulators first through—builds the device ecosystem before the LEO constellation flies, while Airbus’s real-time EGNOS V3 demonstration hit 0.3-meter vertical accuracy augmenting GPS and Galileo together.