Inkling Open Source Launch | Rewire Morning News Briefing

Murati built a model with 975 billion parameters from zero in 9 months; SpaceX fell below its IPO offering price in less than a month. AI’s speed of “creating gods” and the retreat of capital are running in the same summer.

1｜Thinking Machines releases Inkling, Murati’s first move

Eight months after leaving OpenAI, Murati unveiled his first play. Thinking Machines released Inkling, an open-source mixed-expert model with 975 billion parameters. It was trained on 45 trillion tokens of text, images, audio, and video, with native support for four modalities of reasoning. Each call activates about 41 billion parameters only, directly matching closed-source giants on speed and cost.

It took only 9 months to go from zero to release. Compared with OpenAI and Anthropic’s development cycles of several years, this pace itself is a declaration. Thinking Machines is currently valued at $12 billion, with Nvidia as an investor. Murati’s logic isn’t to build the smartest general-purpose model, but the easiest model to customize. Full weights are now live on Hugging Face, and enterprises can directly fine-tune on the Tinker platform. When model capabilities start to converge, the next round of competition may not be about parameter size, but about who can become an enterprise’s private AI faster.

（Source: Fortune / Axios / TechCrunch / FT / WSJ / SiliconANGLE）

2｜The Iran–U.S. war enters a new phase, and the Pentagon’s bills have exceeded $100 billion

The United States has resumed a maritime blockade against Iran, while also launching a new round of airstrikes to hit Iranian military facilities used to threaten navigation through the Strait of Hormuz. Overnight airstrikes killed at least 7 Iranian servicemembers and injured more than 260. Iran announced the peace agreement is void, calling it a “existential war,” and claimed it had launched strikes against 85 U.S. military targets in Bahrain and Kuwait.

On the same day, Iran released an American female citizen who had been detained since December 2024. In escalations, taking a de-escalatory posture is Tehran’s classic move. But the intensity of clashes is far beyond the ceasefire framework from June. Intelligence officials estimate that the Pentagon’s war costs have already exceeded $100 billion, and the Trump administration has yet to publish an official estimate. Roughly one-fifth of the world’s oil and natural gas shipments pass through the Strait of Hormuz. Clashes along this waterway are adding a war premium to global energy pricing. （Continued from yesterday’s report）

（Source: Fortune / Al Jazeera / CNN / Wired / Axios）

3｜SpaceX drops below its IPO offering price; the honeymoon ends in weeks

SpaceX’s stock price fell below its $135 IPO offering price and briefly touched $132.15. After surging nearly 50% in the first three days following the IPO last month, it then fell by nearly a quarter over the next three days. Market cap dropped from a peak of about $2.6 trillion to about $1.75 trillion, vaporizing more than $850 billion from the high. Investor fervor lasted less than a week.

A $86 billion IPO is one of the largest tech listings in history. But SpaceX’s problem isn’t fundamentals—it’s expectations. Even a highly profitable rocket monopoly can’t sustain the valuation bubble after an IPO. For AI unicorns waiting in line, it’s a mirror. The capital markets’ patience for the “next trillion-dollar story” is being shortened on a day-by-day basis.

（Source: Bloomberg / CNBC / TechCrunch / FT）

4｜Apple Intelligence approved to enter China; the AI backbone switches to Alibaba

China’s Cyberspace Administration has approved Apple Intelligence to launch in China, with the underlying model using Alibaba’s Tongyi Qianwen. iOS, iPadOS, macOS, and visionOS will be fully covered, and Baidu AI will also be integrated. The approval was granted on July 15, but no specific launch date was specified. Alibaba’s U.S.-listed shares rose 4% on the day.

In other global markets, Apple Intelligence uses Apple’s in-house models and a collaboration model with OpenAI, but in China it must switch to a local AI provider. The phone can be Apple’s; the AI must be a company in China. The essence of this swap is data sovereignty. When Apple hands over control of the AI backbone to Alibaba, its AI capability in the China market no longer gets defined by itself. For Apple, becoming the default AI engine for iPhones is a ticket to an entry point at the billion-device scale.

（Source: TechCrunch / The Information / WSJ / CNBC）

5｜DTCC tokenized securities move into live trading; Wall Street rewrites its own pipeline

The Depository Trust & Clearing Corporation (DTCC) handled the first batch of live trades for tokenized securities, covering Russell 1000 stocks, major ETFs, and U.S. Treasuries. More than 50 institutions participated, including JPMorgan, BlackRock, Goldman Sachs, and Vanguard. In December 2025, the SEC cleared this path with a no-objection letter. The pilot will continue through the end of summer, with full commercial rollout in October.

DTCC clears and settles nearly all U.S. securities transactions. When the clearing-and-settlement infrastructure itself begins to tokenize, this isn’t a “crypto enters Wall Street” story—it’s Wall Street using blockchain technology to rewrite its own settlement pipeline. Tokenization doesn’t change the assets themselves; it changes how assets are delivered and settled. If the full rollout in October proceeds smoothly, the traditional time framework for clearing and settlement could be changed permanently.

（Source: CoinDesk / WSJ / Crypto Briefing / Genfinity）

Also worth knowing ↓

Anthropic and Blackstone officially launched Ode. Ode with Anthropic is a $150 million AI implementation company. Investors include Goldman Sachs, Apollo, and Sequoia. The core team comes from Fractional AI, which was acquired in May. Anthropic provides the models; Ode provides on-site engineers for deployment. Their bet is that the next trillion-dollar market for AI lies in implementation, not in the models themselves. （Source: TechCrunch / BusinessWire）

OpenAI created an AI hacker to attack its own models. GPT-Red evolved its attack capabilities through self-play, finding attack paths for 84% of scenarios in benchmark tests; human safety researchers found only 13%. It also discovered a previously unseen prompt-injection attack. OpenAI calls it a “fake chain of thought.” （Source: MIT Technology Review）

ChangXin Memory Technologies launched on the Shanghai STAR Market with an $8.6 billion IPO, priced at 8.66 yuan per share. This is the largest semiconductor IPO in A-shares history, surpassing the fundraising scale of SMIC in 2020. ChangXin is the world’s fourth-largest DRAM manufacturer, with a market share of about 7.7%. It will begin trading on July 27. On the same day that SpaceX fell below its offering price in the U.S., China’s chip localization mobilization accelerated as well. （Source: Reuters / Bloomberg）

June PPI unexpectedly fell by 0.3%, showing signs that wholesale inflation is easing. Commodity prices plunged 1.4%, energy prices fell 6.4%, and gasoline fell 12%. Year over year, it dropped from 6.5% in May to 5.5%, well below economists’ expected 6.2%. Energy costs boosted by war got a temporary breather, but analysts warned that the impact of Iran’s war after re-escalation has not yet been factored in. （Source: BLS / CNN / Advisor Perspectives）

Intel became the first company worldwide to mass-produce High NA EUV logic chips. Some Panther Lake layers have completed double verification of ASML 0.55 NA scanners on the 18A process. On the same day, Google’s reported 9th-generation TPU selected Intel’s EMIB packaging solution instead of TSMC’s CoWoS. Intel’s manufacturing business has started getting real leverage in foundry orders. （Source: Tom’s Hardware）

IBM CEO Krishnan issued a public letter admitting, “We haven’t adapted and acted fast enough.” The letter did nothing to prevent a 25% plunge, setting the company’s largest single-day drop in 115 years. Multiple analysts interviewed by Fortune believe Krishnan’s own words were more deadly than any outside judgment. （Continued from yesterday’s report） （Source: Fortune）

The next day, Federal Reserve Chair Waller testified before Congress, saying he “often” meets with the Trump administration. He said AI spending may push up prices but will not cause persistent inflation, while continuing to refuse to disclose the direction of interest rates. On the same day, Buffett said Trump’s choice of Waller was a “good choice.” （Continued from yesterday’s report） （Source: CNBC / Fortune）

Click to learn about the jobs at Lydong BlockBeats

Welcome to join Lydong BlockBeats’ official community:

Telegram subscription group: https://t.me/theblockbeats

Telegram group chat: https://t.me/BlockBeats_App

Twitter official account: https://twitter.com/BlockBeatsAsia