DeepSeek Pauses Funding Round as Chip Deals Stack Up
DeepSeek pauses its round after Liang Wenfeng's remarks spread, as Samsung and Nvidia-SK Group sign chip deals worth hundreds of billions.
July 26, 2026 | Reading time: 8 minutes | Issue #222
Lead
DeepSeek has told investors it is suspending its second funding round after a translated transcript of founder Liang Wenfeng discussing U.S.-China AI competition went viral, according to Bloomberg. The remarks, from a May investor meeting, argued that the main gap between Chinese and American labs is compute power rather than talent, that Nvidia's CUDA moat is disintegrating, and that China's domestic AI-chip ecosystem will prove viable within a year. The leak turned a routine capital-raising exercise into a geopolitical event and sent investors scrambling to interpret whether the pause reflects regulatory caution, strategic recalculation, or both.
The incident lands at a moment when the physical layer of AI is being redrawn at unprecedented scale. Samsung announced a contract worth more than $200 billion to manufacture chips for Broadcom through 2030, focusing on 2nm and below process technologies. Nvidia and SK Group unveiled a $500 billion-plus initiative that includes SK Hynix supplying next-generation memory for Nvidia and joint development of HBM. Verizon disclosed a $1 billion-plus deal to provide dark-fiber connectivity for Google's data centers. Etched, the startup building inference-specific chips, raised a $300 million Series C at a $10.3 billion valuation. Fluidstack, the AI data-center builder, closed an $830 million Series A at a $7.5 billion valuation. The money flowing into compute is no longer about marginal capacity; it is about locking in the supply chain for the next five years.
The contrast between DeepSeek's sudden capital pause and the avalanche of infrastructure deals elsewhere is the central tension of the week. Chinese labs are racing to build capabilities under export controls and capital-market suspicion, while American and allied companies are signing decade-long contracts to secure silicon, memory, and fiber. Liang's bet is that open-source models and domestic chips can neutralize the compute gap. The deals around him suggest his competitors are not counting on it.
Briefs
Samsung Signs $200B+ Broadcom Chip Supply Contract
Samsung will manufacture chips for Broadcom under a contract worth more than $200 billion through 2030, with production focused on Samsung's 2nm and more advanced process nodes. The deal, reported by Bloomberg and surfaced on TechMeme, is one of the largest foundry commitments in recent memory and gives Samsung a marquee customer for its most advanced manufacturing. Broadcom designs the chips; Samsung fabricates them. For Samsung, the contract is a strategic win at a moment when TSMC dominates leading-edge production and Intel's foundry ambitions are faltering.
Nvidia and SK Group Commit $500B+ to AI Data Centers
Nvidia and SK Group unveiled a more than $500 billion AI initiative that includes SK Hynix supplying next-generation memory to Nvidia and joint development of high-bandwidth memory. Separately, Nvidia plans to invest $1 billion in Naver to help finance an AI data center in South Korea, according to Bloomberg, and to partner with SK Group on more than 2 gigawatts of AI data-center capacity. The announcements tie Nvidia more tightly to Korean memory and cloud infrastructure at a time when HBM supply is a critical constraint on GPU deployments.
Etched Raises $300M Series C at $10.3B Valuation
Etched, the AI chip startup designing inference-specific silicon, raised $300 million in a Series C led by Sequoia at a $10.3 billion valuation, up from $5 billion in December. Andreessen Horowitz, SK Hynix, and others also invested. The round is a bet that the inference phase of AI workloads will eventually demand chips optimized for serving models rather than training them. Etched's valuation has more than doubled in seven months, reflecting investor confidence that the market for specialized inference silicon is separate from Nvidia's training stronghold.
Sources: TechCrunch
Travis Kalanick's Robotics Company Atoms Raises $1.7B
Travis Kalanick's robotics holding company, Atoms, raised $1.7 billion led by Andreessen Horowitz, with Uber, Bain Capital, Fifth Wall, and others participating. Ben Horowitz will join the board. Atoms sits atop CloudKitchens and Pronto, the heavy-industry automation company Kalanick acquired earlier this year, and has signaled interest in mining and humanoid robotics. The round reconnects Kalanick with Uber, the company that pushed him out in 2017, and makes Atoms one of the best-funded physical-AI bets of the year.
Sources: TechCrunch
Mobileye Founder Amnon Shashua to Step Down After 30 Years
Mobileye founder and CEO Amnon Shashua plans to step down after nearly three decades at the company, according to a regulatory filing. Mobileye, the Israeli automotive-computer-vision company acquired by Intel for $15.3 billion in 2017 and later spun out, is pushing into robotaxis and humanoid robotics through its acquisition of Mentee Robotics for $900 million. Shashua will remain CEO until a replacement is hired. The transition marks the end of an era for one of the most influential figures in autonomous-driving silicon.
Sources: TechCrunch
Microsoft Launches In-House AI Models, Claims 89% Cost Cut
Microsoft has launched its own in-house AI models and says they cut costs by up to 89% compared with OpenAI, according to VentureBeat. The move is the most concrete sign yet that Microsoft is diversifying its AI supply chain beyond its close partnership with OpenAI, which has powered Copilot and Azure OpenAI Service. In-house models give Microsoft negotiating leverage and a fallback if OpenAI pricing or capacity becomes a constraint. The reported cost savings are large enough that, if verified at scale, they could pressure OpenAI's enterprise pricing strategy.
Sources: VentureBeat
The Map
The week's geography is a map of capital concentration rather than scientific discovery. In East Asia, South Korea captured the largest commitments: Samsung's Broadcom deal and Nvidia's SK Group alliance together exceed $700 billion in announced value, while Naver landed a $1 billion Nvidia investment for a domestic AI data center. China responded with friction rather than fundraising: DeepSeek's capital round stalled after a founder transcript circulated, and Unitree's humanoid-robot shipments and planned Shanghai IPO showed that hardware startups can still advance even as capital markets grow cautious. Hong Kong continued its role as a listing venue for mainland chip firms, with MetaX filing confidentially.
India appeared through enterprise software rather than frontier models. ServiceNow invested $40 million in BusinessNext, a Noida-based banking-software company, at a $700 million valuation. The deal points to India's AI-services sector moving into vertical-specific, on-premise enterprise workflows rather than generic cloud outsourcing. Europe's signal was smaller in dollar terms but strategically specific: Arrakis, a London- and Paris-based startup, raised $38 million to build an AI operating system for industrial sectors including aerospace, energy, and logistics. The geography of the week is clear: frontier model headlines are still American, but the capital and specialization are dispersing across Korea, China, India, and Europe.
The View
The week exposes a split in how the AI industry is pricing risk. American and Korean incumbents are signing long-term, capital-intensive contracts on the assumption that demand for compute will remain robust for years. DeepSeek's funding pause suggests that Chinese startups face a different calculus, where a founder's remarks can freeze a capital round and where access to advanced chips is uncertain enough to shape strategy in public. Both sides are making big bets. Samsung, Broadcom, Nvidia, and SK Group are betting that the hardware supply chain can be locked in before capacity becomes the binding constraint. Liang Wenfeng is betting that open-source models and a domestic chip ecosystem can close the gap without matching that scale of capital. The immediate evidence favors the incumbents: multi-hundred-billion-dollar contracts are hard to dismiss. But DeepSeek's earlier models showed that efficient training can stretch limited hardware further than expected. The question is whether capital, talent, or ingenuity is the binding constraint. This week, capital made the loudest argument.
The Miss
The APEC Chengdu AI statement received modest coverage after it endorsed open-source development while demanding "strong security assurance." Twenty-one member economies, including the U.S. and China, signed the first minister-level APEC AI statement to explicitly include open-source cooperation. It creates no binding rules, but it signals that the Asia-Pacific region is trying to define a middle path between unrestricted open-weight release and closed-model concentration. For model builders, the practical implication is that export markets may increasingly expect transparency and security testing as the price of open-source distribution.
Pull Quotes
"The main U.S.-China AI gap is compute, not talent, and China's domestic AI-chip ecosystem will prove viable within a year." — Liang Wenfeng, DeepSeek, translated by Fred Gao
"Changemakers, the builders of tomorrow's progress machines, will inevitably go up against the final boss, Nature and its fierce resistance to change." — Travis Kalanick, Atoms, on X
"Most AI investment to date has targeted the 30% of workers behind a desk. The real ROI lies in the 70% running industrial operations." — Rafael Quintanilla, CEO, Arrakis, Fortune
Reads & Links
- Bloomberg on DeepSeek pausing its funding round: Bloomberg
- Translated transcript of Liang Wenfeng's investor meeting: Fred Gao
- TIME profile of Unitree and China's humanoid-robot push: TIME
- Bloomberg on Samsung's $200B+ Broadcom contract: Bloomberg
- Reuters on Nvidia and SK Group's $500B+ AI initiative: Reuters
- TechCrunch on Etched's $300M Series C at $10.3B: TechCrunch
- TechCrunch on Travis Kalanick's Atoms raising $1.7B: TechCrunch
- Fortune on Arrakis emerging from stealth with $38M: Fortune
- TechCrunch on ServiceNow's $40M BusinessNext investment: TechCrunch
- APEC Chengdu AI ministerial statement: APEC
The same week showed capital flooding into the machines and a Chinese founder finding that words can move markets as fast as models can.