Daily Tech & GNSS News Digest - July 1, 2026

Wednesday’s tech news circled the same pressure point from three directions: everyone wants more compute and more chips than the supply chain can currently deliver. Apple is quietly asking Washington to look the other way on blacklisted Chinese memory suppliers, Meta is turning its own capacity crunch into a business by reselling what it doesn’t use, and an AI startup just committed $6.3 billion to rent Nvidia chips inside a SpaceX data center. On the positioning side, Congress spent a hearing cataloguing how far GPS modernization has slipped, while a new transatlantic partnership previewed one path past it—pairing a low-orbit navigation signal with receivers built to shrug off jamming.

Tech News

Apple Lobbies Washington to Buy Memory Chips From Blacklisted Chinese Suppliers

Apple is in negotiations to buy memory chips from two Chinese semiconductor makers on the Pentagon’s blacklist, hoping to blunt a global memory shortage that has already forced it to raise prices across its lineup, per Bloomberg. The iPhone maker is seeking DRAM from ChangXin Memory Technologies (CXMT) and Yangtze Memory Technologies, both flagged on the Pentagon’s 1260H list over alleged ties to China’s military, for use in devices sold in China.

The shortage traces back to the AI boom itself: dominant suppliers Micron, Samsung, and SK Hynix have redirected production from standard consumer memory toward the high-bandwidth memory that AI data centers need, leaving phone and laptop makers short. Apple isn’t currently barred from buying from CXMT, but CEO Tim Cook has personally lobbied Trump administration officials, including Treasury Secretary Scott Bessent, for assurances that CXMT won’t be added to the Commerce Department’s Entity List—a sign of how far the memory crunch is bending Apple’s usual supply-chain caution.

Meta Plans “Meta Compute,” a Cloud Business to Resell Excess AI Capacity

Meta is building a cloud infrastructure business, internally called Meta Compute, to sell both raw AI computing capacity and hosted access to its own models, setting up direct competition with AWS, Azure, and Google Cloud, per Bloomberg. The unit would compete with neocloud players like CoreWeave on raw capacity while also renting out access to Meta’s Muse Spark models, and is being led by infrastructure chief Santosh Janardhan, Meta Superintelligence Labs’ Daniel Gross, and Meta President Dina Powell McCormick.

CEO Mark Zuckerberg told investors in May that reselling excess infrastructure was “definitely on the table” after Meta raised its 2026 capex guidance to $125–145 billion. The market reaction was immediate: Meta shares jumped more than 7% on the news, while CoreWeave and Nebius shares fell 10–12%—an early signal that even a hint of Meta entering the cloud market rattles the pure-play neoclouds.

Reflection AI Signs $6.3 Billion Compute Deal for SpaceX’s Colossus 2

Reflection AI, an open-source AI startup backed by Nvidia, signed a multi-year compute agreement with SpaceX worth up to $6.3 billion, paying $150 million per month starting July 1 for access to Nvidia GB300 chips at SpaceX’s Colossus 2 data center near Memphis, Tennessee, per CNBC and TechCrunch. The deal runs through 2029 if fully executed.

Reflection joins Anthropic, Google, and Cursor as outside tenants of Colossus, which SpaceX has built into one of the largest third-party compute platforms anywhere—committed revenue from outside clients now tops $80 billion through 2029. It’s another data point on how much of the AI buildout is being financed and hosted by companies whose core business isn’t AI at all.

Additional Headlines:

Anthropic’s Claude Sonnet 5 becomes the default free and Pro model: Released June 30, Claude Sonnet 5 is now the default model for Anthropic’s Free and Pro tiers, priced at $2/$10 per million input/output tokens through August 31—well below flagship Opus 4.8—as Anthropic pushes cheaper agentic AI ahead of an expected IPO, per VentureBeat and TechCrunch.
California signs discounted Claude deal for every state agency: Governor Gavin Newsom announced a partnership giving all California state agencies, cities, and counties access to Claude at a 50% discount plus free workforce training—one of the largest state-government AI deployments to date, per CBS Sacramento.
Tech and finance payrolls shrink as AI adoption accelerates: US financial-activities and information-sector jobs have declined by an average of 28,000 per month in 2026, the fastest AI-adoption sectors also showing the fastest payroll contraction, per Bloomberg.
Nasdaq chipmakers slide as investors take profits: Micron fell more than 10% Wednesday even after gaining over 260% year-to-date, as semiconductor stocks that surged more than 80% in the first half of 2026 saw a broad pullback, per CNBC.

GNSS News

Congress Puts the State of US PNT on the Record—and the Picture Isn’t Pretty

The House Energy and Commerce Subcommittee on Communications and Technology held a hearing, “Where Are We? Examining Positioning, Navigation, and Timing Capabilities in the United States,” chaired by Rep. Richard Hudson, that laid out three PNT tracks moving simultaneously: GPS modernization, interference enforcement, and the search for a complementary architecture, per Inside GNSS. Witnesses included Lisa Dyer of the GPS Innovation Alliance, Mariam Sorond of NextNav, and Sam Matheny of newly launched Merkhet Solutions.

The numbers are sobering: eight of the 32 GPS satellites are running on a single string, one subsystem failure from going non-operational, and on April 17 the Space Force terminated the long-delayed GPS Next Generation Operational Control System, which had run more than a decade behind schedule and triggered a Nunn-McCurdy cost breach. Meanwhile, the FCC and Department of Transportation told Congress they lack the budget and staff to enforce existing anti-jamming laws or coordinate a government-wide response to the growing volume of interference incidents—a gap between the threat and the enforcement capacity that the hearing did little to resolve.

US startup TrustPoint and NovAtel, part of Hexagon AB, are collaborating on a hybrid GNSS/LEO PNT effort that pairs TrustPoint’s low-Earth-orbit C-band navigation service with NovAtel’s anti-jamming receiver technology, per Inside GNSS. TrustPoint’s C-band constellation—most recently expanded with its third demonstrator satellite, Time Flies—is being tested against a $1.2 million SBIR Phase II contract awarded under the Navy’s NAVAIR program to build the first receivers capable of processing TrustPoint’s signal.

NovAtel brings adaptive-antenna and signal-processing technology that suppresses jamming while preserving GNSS accuracy, casting the project as a transatlantic effort that pairs American new-space capacity with European receiver engineering. Further demonstrations and interoperability testing are planned into 2026, part of a broader private-sector push toward hybrid LEO-GNSS architectures that complement—rather than replace—legacy GPS and Galileo signals.

Key Takeaways

The AI boom is straining the physical supply chain it depends on: Apple is lobbying Washington to buy blacklisted Chinese DRAM as memory makers redirect capacity to AI, while Reflection AI committed $6.3 billion to rent Nvidia chips inside SpaceX’s data center—two sides of the same compute-and-component crunch.
The cloud market’s biggest players keep blurring their own categories: Meta’s plan to resell excess AI capacity as “Meta Compute” sent its stock up 7% and rival neoclouds down double digits, while Anthropic’s cheaper Claude Sonnet 5 and a discounted California state-government deal show AI vendors racing to lock in volume.
US GPS modernization is falling behind the threat it’s meant to counter: A House hearing detailed satellites running on a single string and a canceled ground-control program, even as TrustPoint and Hexagon’s NovAtel push a hybrid LEO-GNSS signal-and-receiver pairing as one way to build resilience around the aging system.