Daily Tech & GNSS News Digest - July 4, 2026

The holiday weekend’s biggest story wasn’t really new—it just got a lot more specific. Newly unsealed court emails show exactly how close the Pentagon and Anthropic came to a deal before the relationship blew up over autonomous weapons, giving the clearest look yet at where the US government wants to draw AI’s lines and where Anthropic won’t move. Meanwhile, Bloomberg’s read on what AI actually costs to run took a leg down, and Micron kept building anyway. On the positioning side, the news skewed toward resilience infrastructure: an industrial LEO-satellite partnership and a new family of timing modules built to keep working when GPS doesn’t.

Tech News

Unsealed Emails Show the Pentagon Called a Deal With Anthropic “Very Close”—the Day Before Blacklisting It

Newly unsealed court filings reveal that a Pentagon official told Anthropic CEO Dario Amodei the two sides were “very close” to a contract agreement just one day before the Department of Defense designated Anthropic a supply-chain risk to national security, according to Gizmodo, TechBrew, and other outlets reporting on documents unsealed July 2. DoD Under Secretary Emil Michael wrote to Amodei that he’d reviewed Anthropic’s latest contract draft with lawyers and thought “we are very close here”—before the department, under Defense Secretary Pete Hegseth, barred defense contractors and suppliers from working with the company and ordered a transition away from its products.

The fight centers on two limits Anthropic maintains on Claude no matter the customer: no fully autonomous weapons systems that can engage targets without a human in the loop, and no domestic mass surveillance. When Amodei objected that the Pentagon’s proposed language would “completely remove our redlines,” Michael reportedly didn’t dispute it, calling the guardrails “just not workable” and pushing for coverage of “all lawful uses.” Federal Judge Rita Lin, who granted Anthropic a preliminary injunction in March, called the contradiction between the friendly email and the hostile blacklist designation “exceedingly difficult to square”—a rare public paper trail on how AI safety commitments actually hold up against the Pentagon’s biggest checkbook.

AI Token Prices Are Falling, and That’s Not Entirely Good News for the Trade

The Silicon Data LLM Token Expenditure Index, which tracks what buyers actually pay for AI model usage, has fallen almost 20% from its May peak after nearly doubling since it launched in December, according to Bloomberg. The index is, in Bloomberg’s framing, the cleanest available read on the more than $700 billion in AI capital spending that hyperscalers have committed to this cycle—and a falling price per token can mean either healthy competition or a sign that pricing power is eroding faster than the capex can be justified.

Silicon Data’s own analysts caution the drop could reflect users shifting toward cheaper, more efficient models rather than a straightforward price war, and note the index has recently stagnated rather than continuing its slide. Still, the timing is pointed: it lands in the same week Micron is spending billions more on memory capacity and Wall Street keeps parsing whether AI infrastructure spending and AI revenue are converging or drifting apart.

Micron Breaks Ground on $9.3 Billion Japan Plant Expansion to Feed the AI Memory Boom

Micron Technology broke ground July 4 on a ¥1.5 trillion (~$9.3 billion) expansion of its Hiroshima, Japan factory, adding capacity to produce high-bandwidth memory used in AI accelerators including Nvidia’s chips, according to Bloomberg and the Japan Times. Japan’s Ministry of Economy, Trade and Industry is covering as much as ¥500 billion of the cost, and the Boise-based company expects the expanded site to begin shipments around summer 2028.

The investment lands squarely inside the memory shortage that’s been rippling through the industry all year—the same crunch that’s pushed Micron shares up more than 260% year-to-date even as they’ve whipsawed on profit-taking, and that has phone and laptop makers scrambling for DRAM supply that memory giants have redirected toward AI data centers. Building new HBM capacity takes years, which is exactly why Micron, Samsung, and SK Hynix keep announcing fabs faster than the current shortage can be resolved.

Additional Headlines:

OpenAI delays wider GPT-5.6 rollout at the government’s request: OpenAI is limiting GPT-5.6 to a vetted “limited preview” after the Office of the National Cyber Director and Office of Science and Technology Policy asked for early access under a new voluntary frontier-model review framework; CEO Sam Altman said he doesn’t object to safety testing but doesn’t want “the government picking the customers,” per TechCrunch.
Trump says he wants AI guardrails, “but as little as possible”: In a CNBC interview, President Trump said some AI standards are needed but that he wants to avoid rules that could slow US firms down against China—remarks that landed the same week the administration reversed export controls on Anthropic and considered OpenAI’s proposed government equity stake, per Bloomberg.
Claude Fable 5 returns globally after 18 days offline: Anthropic restored global access to Claude Fable 5 on July 1 after the US Commerce Department lifted export controls triggered by a jailbreak an Amazon researcher found weeks earlier; the company agreed to self-report future vulnerabilities as part of the deal, per CNBC and VentureBeat.
Z.ai’s GLM-5.2 beats GPT-5.5 on coding benchmarks at a sixth of the cost: The open-weights Chinese model scores 62.1 on SWE-bench Pro versus GPT-5.5’s 58.6 and lands within striking distance of Claude Opus 4.8 on tool-use benchmarks, per VentureBeat—another sign that open-weight Chinese models are closing the gap on frontier coding performance.

GNSS News

Japanese electronics giant Murata Manufacturing and Xona Space Systems signed a memorandum of understanding to develop positioning and timing products that combine Murata’s sensor and module expertise with Xona’s low-Earth-orbit Pulsar PNT service, according to GPS World. Because Xona’s satellites orbit much closer to Earth than traditional GNSS, they deliver stronger signals that improve reception in cities and indoor environments, along with faster convergence and greater resistance to jamming and spoofing.

The companies are targeting applications where GNSS alone struggles: data centers and financial institutions that need precise timing synchronization for 5G/6G networks, and off-road equipment in construction and agriculture where satellite visibility is often poor. Murata previously invested in Xona through its venture arm, so the MOU extends an existing relationship rather than starting cold—part of a broader pattern of established electronics manufacturers betting on LEO constellations as a complement to, not a replacement for, legacy GPS and Galileo.

Furuno Unveils Anti-Jam Timing Modules Built to Keep Critical Infrastructure Synced

Furuno announced its GF-100 Series of dual-band GNSS-disciplined oscillators—four models built for telecom base stations, broadcast systems, satellite terminals, and power-grid monitoring equipment that need precise timing even when GNSS signals are degraded, per GPS World. The modules pair a dual-frequency L1/L5 receiver with a high-stability oscillator that automatically switches to holdover mode when it detects jamming or spoofing, and support Galileo’s and Japan’s signal-authentication services to help verify that GNSS signals are genuine before trusting them.

The compact GF-105S model matches the holdover performance of Furuno’s larger existing module while cutting the mounting footprint by roughly 80%, and the whole line stays hardware-compatible with Furuno’s current products so equipment makers can upgrade without a redesign. Furuno says insights from the annual Jammertest resilience trials in Andøya, Norway—which it’s attended since 2024—shaped the new design, and plans to put the GF-100 Series through this year’s event ahead of a planned November 2026 launch.

Key Takeaways

The Pentagon-Anthropic fight has a paper trail now: Unsealed emails show a DoD official calling a contract “very close” the day before the department blacklisted Anthropic over its refusal to drop limits on autonomous weapons and domestic surveillance—the clearest evidence yet of how AI safety commitments collide with defense procurement.
The AI trade’s pricing signal just turned negative: Bloomberg’s Silicon Data index shows AI token prices down almost 20% from their May peak, complicating the picture even as Micron commits another $9.3 billion to AI memory capacity in Japan.
GNSS resilience is being built into both new and old infrastructure: Murata and Xona’s LEO satnav partnership targets data centers, 5G networks, and off-road machinery, while Furuno’s new anti-jam timing modules aim to keep telecom, broadcast, and power-grid equipment synchronized when GPS is degraded.