Murati built a 975 billion-parameter model from scratch in 9 months, while SpaceX fell below its IPO price in less than a month. The rapid rise of AI and the capital downturn are racing in the same summer.
1｜Thinking Machines Launches Inkling, Murati's Opening Move
After leaving OpenAI eight months ago, Murati revealed his opening move. Thinking Machines has launched Inkling, an open-source hybrid expert model with 975 billion parameters, trained on 450 trillion tokens of text, image, audio, and video, natively supporting four-mode reasoning. Each call activates only about 410 billion parameters, directly benchmarking against closed-source giants in terms of speed and cost.
It took only 9 months from inception to launch. Compared to OpenAI and Anthropic with development cycles often spanning years, this speed itself is a declaration. Thinking Machines is currently valued at $12 billion, with participation from NVIDIA. Murati's strategy is not to create the smartest general model but the most easily customizable one. The full weights are already live on Hugging Face, allowing enterprises to fine-tune directly on the Tinker platform. As model capabilities begin to converge, the next round of competition may not be in parameter size but in who can quickly convert into an enterprise's private AI.
(Source: Fortune / Axios / TechCrunch / FT / WSJ / SiliconANGLE)
2｜US-Iran War Enters New Phase, Pentagon Bill Surpasses $100 Billion
The U.S. has reimposed a naval blockade on Iran and launched a new round of airstrikes, targeting Iranian military facilities used to threaten the navigation of the Strait of Hormuz. Overnight airstrikes resulted in at least 7 Iranian military casualties and over 260 injuries. Iran has announced the nullification of the peace agreement, calling it an "existential war," and claimed to have struck 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. Making a de-escalatory gesture amidst escalation is a classic Tehran move. However, the intensity of the exchange has far exceeded the ceasefire framework of June. Intelligence officials estimate the Pentagon's war cost has surpassed $100 billion, with the Trump administration yet to disclose an official estimate. Approximately one-fifth of the world's oil and natural gas transportation passes through the Strait of Hormuz, and the ongoing clashes in 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 Falls Below IPO Price, Honeymoon Period Ends in Weeks
SpaceX's stock price fell below the $135 IPO price, briefly touching $132.15. After initially skyrocketing nearly 50% in the first three days post-IPO, it then dropped almost a quarter in the following three days. The market cap plummeted from a peak of around $2.6 trillion to about $1.75 trillion, evaporating over $850 billion from its high. Investor frenzy lasted less than a week.
The $860 billion IPO was one of the largest tech debuts in history. However, SpaceX's issue was not fundamentals but expectations. Even a profitable rocket monopoly couldn't sustain its post-IPO valuation bubble, serving as a mirror to the upcoming AI unicorns in line. The capital markets' patience for the "next trillion-dollar story" is shrinking by the day.
Source: Bloomberg / CNBC / TechCrunch / FT
4 | Apple Intelligence Approved to Enter China, AI Backbone Switches to Alibaba
The Cyberspace Administration of China approved Apple Intelligence to launch in China, using Alibaba's Tongyiyiqianwen underlying model. It covers iOS, iPadOS, macOS, and visionOS entirely, with Baidu AI also being integrated. The approval was granted on July 15, but no specific launch date was specified. Alibaba's US shares rose by 4% on the same day.
While Apple Intelligence uses Apple's in-house models and OpenAI's collaborative models in other global markets, it must switch to local AI vendors in China. The phones can be Apple's, but the AI must be from a Chinese company. The essence of this exchange is data sovereignty. When Apple relinquishes control of the AI backbone to Alibaba, its AI capabilities in the Chinese market are no longer self-defined. For Alibaba, becoming the default AI engine for iPhones is a ticket to access billions of end-users.
Source: TechCrunch / The Information / WSJ / CNBC
5｜DTCC Tokenized Securities Go Live in Production Trading, Wall Street Rewiring Its Pipeline
The Depository Trust & Clearing Corporation (DTCC) has facilitated the first production trades of tokenized securities, covering the Russell 1000 stocks, major ETFs, and U.S. treasuries. Over 50 institutions participated, including JPMorgan Chase, BlackRock, Goldman Sachs, and Vanguard. The SEC greenlit this path with a no-action letter in December 2025, with the pilot running through the end of summer and a full commercial launch in October.
DTCC settles virtually all U.S. securities trades. When the clearing infrastructure itself begins to tokenize, this is not a story of "crypto entering Wall Street" but of Wall Street rewriting its own settlement pipeline using blockchain technology. Tokenization does not change the assets themselves; it changes how assets are delivered and settled. If the full-scale launch in October progresses smoothly, the traditional settlement timelines could be permanently altered.
(Source: CoinDesk / WSJ / Crypto Briefing / Genfinity)
Also Worth Knowing ↓
Anthropic and Blackstone Officially Launch Ode. Ode, with Anthropic, is a $1.5 billion AI implementation company, with investors including Goldman Sachs, Apollo, and Sequoia, and a core team from Fractional AI, acquired in May. Anthropic provides the models, while Ode offers on-site engineering deployment. Their bet is that the next trillion-dollar market for AI lies in on-the-ground implementation, not just in the models themselves.
(Source: TechCrunch / BusinessWire)
OpenAI Has Built an AI Hacker to Attack Its Own Models. GPT-Red continuously evolves its attacking capabilities through self-play and has found attack paths in 84% of benchmark scenarios, compared to human security researchers finding only 13%. It has also discovered a previously unseen form of hint injection attack, termed by OpenAI as "fake chain of thought."
(Source: MIT Technology Review)
ChangXin Memory Debuts on Shanghai's STAR Market with an $8.6 Billion IPO, priced at $8.66 per share. This is the largest semiconductor IPO in A-share history, surpassing SMIC's fundraising in 2020. ChangXin is the world's fourth-largest DRAM manufacturer, with a market share of around 7.7%. Trading commenced on July 27. While SpaceX struggled in the U.S. on the same day, the capital mobilization for China's semiconductor self-sufficiency continues to accelerate.
(Source: Reuters / Bloomberg)
In June, the PPI unexpectedly fell by 0.3%, signaling a retreat in wholesale inflation. Commodity prices plummeted by 1.4%, energy prices dropped by 6.4%, and gasoline fell by 12%. The year-over-year rate decreased from 6.5% in May to 5.5%, well below economists' expectations of 6.2%. The temporary relief in energy costs driven by the war has eased, but analysts warn that the impact of a re-escalation in the Iran conflict has not been factored in. (Source: BLS / CNN / Advisor Perspectives)
Intel has become the world's first company to mass-produce High NA EUV logic chips. Portions of the Panther Lake layer have undergone dual validation on the 18A process with ASML's 0.55 NA scanner. On the same day, Google's ninth-generation TPU reportedly chose Intel's EMIB packaging solution over TSMC's CoWoS. Intel's manufacturing business is now starting to gain a real edge in contract orders. (Source: Tom's Hardware)
IBM CEO Krishna issued a public letter acknowledging that "we did not move quickly enough to adapt and take action." This letter failed to prevent a 25% plunge, marking the company's largest single-day drop in its 115-year history. Several analysts interviewed by Fortune believe that Krishna's own words were more damning than any external assessment. (Continued from yesterday's report) (Source: Fortune)
On the second day of his congressional testimony, Fed Chair Powell stated that he "frequently" meets with the Trump administration. He mentioned that AI spending could drive up prices but would not lead to persistent inflation, while continuing to refuse to disclose the direction of interest rates. On the same day, Buffett remarked that choosing Powell was a "good choice" for Trump. (Continued from yesterday's report) (Source: CNBC / Fortune)
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
