Sign in Subscribe
AI Intelligence

ASML Raises Guidance Again on AI Chip Demand

Vijay Bhagwati

• 7 min read
Share

DeepSeek targets a $71 billion IPO, Apple tests on-device model compression, and Washington launches an AI cyber clearinghouse.

July 15, 2026 | Reading time: 9 minutes | Issue #211

Lead

ASML hiked its full-year forecast for the second time this year after second-quarter revenue and profit beat estimates, the Dutch lithography giant said Wednesday. The company now expects 2026 sales between €43 billion and €45 billion, up from a prior range of €36 billion to €40 billion, and gross margin of 54% to 56%, up from 51% to 53%. Net profit came in at €2.9 billion against €2.6 billion expected, and net sales hit €9.3 billion versus €8.8 billion expected. The stock rose more than 7% at the open and has doubled in 2026.

ASML is the only producer of the extreme-ultraviolet machines that etch the most advanced processors, so its order book is a direct read on how aggressively TSMC, Samsung, and Intel are expanding leading-edge capacity. Chief executive Christophe Fouquet said customers are "accelerating their capacity expansion plans" and that the company is adding 30% to its 2026 low-NA EUV and DUV immersion capacity to keep up. The figures land days after TSMC reported a 68% jump in June sales and said it would add two advanced packaging plants in Taiwan.

The upgrade also highlights a tension that runs through the rest of today's stories. Demand for AI hardware is clearly accelerating, but the economics of that buildout are not evenly distributed. ASML's China business still accounts for roughly 20% of sales despite tightening export controls, and a U.S. bill that would further restrict DUV shipments to Chinese chipmakers is working through Congress. Fouquet noted that past restrictions have sometimes triggered a rush of pre-emptive buying. The bet embedded in ASML's raised guidance is that the current capex cycle has enough momentum to absorb whatever geopolitical friction comes next.

OpenAI Researcher Miles Wang Eyes $2B Drug Discovery Startup

OpenAI researcher Miles Wang is leaving the lab to start an AI drug-discovery company, TechCrunch reported. Wang is in talks to raise about $200 million at a $2 billion valuation, with Lightspeed in discussions to lead. Several other OpenAI researchers are expected to join. Wang disputed the funding figures and company description but did not provide corrected numbers.

The startup is reportedly focused on finding new uses for existing drugs, including compounds that previously failed in trials. That path can reach revenue faster than building new molecules from scratch because safety profiles are already established. Wang joined OpenAI in 2024 after leaving Harvard and co-authored work on using AI to accelerate biological research. The deal follows Chai Discovery's $400 million raise at a $3.8 billion valuation and Isomorphic Labs' $2.1 billion Series B, confirming that AI life-science startups can still command premium prices.

Apple Evaluates PrismML's On-Device Model Compression

Apple is in early talks with PrismML, a Caltech spinout backed by Khosla Ventures, about technology that can shrink large AI models enough to run directly on an iPhone, PrismML chief executive Babak Hassibi told CNBC. The company on Tuesday released a compressed version of Alibaba's open-source Qwen model that it said fits all 27 billion parameters into under 4 GB, down from roughly 54 GB.

PrismML reduces each stored value from 16 bits to one or three possible values. It claims 10 to 15 times less memory use, six to eight times faster response generation, and three to six times lower energy consumption than conventional versions. The trade-off is a few percentage points of overall performance, with factual recall weakening before reasoning, math, and coding. Apple did not comment. The release coincides with the public beta of iOS 27, which includes Apple's long-delayed Siri overhaul and a push to keep more AI processing on the device.

White House Stands Up "Gold Eagle" AI Cyber Clearinghouse

The Trump administration launched a federal clearinghouse for sharing AI-discovered cyber-threat information between government and the private sector, CyberScoop reported. "Gold Eagle," created by executive order last month, is managed by the Treasury Department with input from CISA, DHS, and the Pentagon, plus open-source software providers and critical-infrastructure operators.

A senior White House official said the project is already receiving vulnerability intelligence and prioritizing patches. It will use models including Anthropic's Mythos to find flaws in victim systems and software, and has built a Vulnerability Information and Coordination Environment platform with Carnegie Mellon's Software Engineering Institute. The administration framed the effort as a way to stay ahead of adversaries as AI tools make vulnerability discovery faster. Former White House cyber coordinator Michael Daniel told CyberScoop the work is still experimental and that policymakers need to learn whether AI-generated threats are structurally different from conventional phishing and exploitation.

EU Eases Battery Rules for Smart Glasses Under US Pressure

The European Commission on Tuesday proposed exempting smart glasses, smartwatches, fitness trackers, and electric toys from rules requiring removable batteries, Politico Europe reported. The delegated act, which needs two months of review by Parliament and member states, removes a regulatory barrier that had delayed Meta's newest smart glasses in Europe.

US Ambassador to the EU Andrew Puzder had publicly criticized the battery rules in March, calling them so restrictive that they blocked a "wonderful, jointly developed, US-European product." The Commission denied giving in to pressure, saying the exemption was based on safety and technical limits. More than seven million pairs of Meta smart glasses sold worldwide in 2025, but distribution in Europe, the Middle East, and Africa remained slow. Consumer groups warned that the change weakens protections and cited privacy concerns about always-on cameras.

Eastern Front

DeepSeek is in talks for a new funding round at a $71 billion pre-money valuation, the Financial Times reported via Business Times. That is a sharp step up from the roughly $50 billion valuation the Chinese AI lab secured in early June, when it closed a $7 billion round backed by Tencent and Contemporary Amperex Technology. Founder Liang Wenfeng has told investors that DeepSeek will prioritize research over short-term commercialization and keep developing open-source models.

The new valuation would make Liang the world's richest AI model creator, with a net worth estimated at $36 billion by the Bloomberg Billionaires Index. It also underscores how Chinese labs are capitalizing on Western constraints. DeepSeek's rise has been fueled partly by US export controls that limited access to advanced Nvidia chips, forcing efficiency-focused training approaches that later impressed Western researchers. The company is also reportedly developing its own AI chip. A $71 billion price tag would place DeepSeek in the same valuation neighborhood as some top US frontier labs, even though its revenue model remains thin. The open-source pledge matters strategically: it gives Beijing a path to global distribution without relying on app stores or cloud partnerships that Washington can disrupt.

Capital Flows

Sovereign AI infrastructure is becoming a distinct venture category. London-based Valarian raised a $50 million Series A led by NEA, bringing total funding to $70 million, Fortune reported. The company's ACRA software wraps AI workloads and sensitive applications in a controlled layer that sits on top of Amazon, Microsoft, or Google cloud infrastructure and governs what data leaves, who touches it, and who can shut the system off. NEA called it the firm's first defense and dual-use investment in Europe.

Valarian's pitch is that sovereignty cannot be a settings toggle inside someone else's data center. The argument gained urgency for European buyers after the Trump administration cut off Anthropic's access abroad earlier this year. Meanwhile, India's Tata Consultancy Services said it plans to deploy up to 8,900 AI deployment engineers and is scouting acquisitions, according to Channel NewsAsia. The two moves show how AI capital is flowing into control layers, not just model training. One region wants to escape American cloud jurisdiction; the other is scaling the workforce that will install AI inside global enterprises.

From the Lab

A team at the Technical University of Denmark has combined a generative AI model with a printer-sized quantum computer from British startup Orca Computing to generate novel peptides that bind to specific proteins, Wired reported. The hybrid system produced more successful peptides than a purely classical version, with the largest gains where training data was scarce. The researchers tested the generated peptides in a wet lab.

The work is still small-scale. Quantum hardware is not powerful enough to run full-sized antibody models, and finding a binding peptide is only one step in drug development. But the study is a rare example of a near-term, experimentally validated use for quantum computing in generative biology. The team is now exploring synthetic antidotes for snakebite venom and personalized immunotherapies. Orca is also working with BP on chemistry and Toyota on design processes.

The View

The day's stories trace the same fault line: AI is spreading faster than the institutions built to govern it. ASML's raised guidance shows the hardware buildout is real and accelerating. DeepSeek's $71 billion target shows that capital will finance alternative supply chains regardless of export controls. Washington's Gold Eagle clearinghouse is an admission that automated vulnerability discovery is now too fast for conventional coordination. And the EU's smart-glasses exemption shows regulation bending under commercial and diplomatic pressure before it has settled basic privacy questions.

The underlying pattern is a shift from model competition to infrastructure competition. Model capabilities are compressing into smaller weights, as PrismML's Qwen demo suggests, while the systems around models, sovereignty layers, cyber clearinghouses, and national chip subsidies, are becoming the main arena. The labs that win may not be the ones with the biggest training clusters, but the ones that can secure supply, distribution, and regulatory accommodation at the same time.

The Miss

Anthropic released a new advertisement titled "There's hope in hard questions" that has drawn criticism for doomer imagery including burning houses, surveillance scenes, and Arlington National Cemetery, TechCrunch reported. Chief executive Sam Altman called it satire on X. The episode is a side note, but it captures the difficulty of selling safety as a product when competitors can frame the same message as fear-mongering.

Pull Quotes

  • "Customers are accelerating their capacity expansion plans. This is translating into customer commitments across our product portfolio, providing ASML with increased visibility into longer-term demand." — Christophe Fouquet, ASML CEO, via CNBC
  • "No one could use their model anymore, because the president of another country had shut it off." — Max Buchan, Valarian co-founder, on the Anthropic export disruption, via Fortune
  • "I was a huge quantum skeptic. I believed any application to my work would be decades away." — Timothy Patrick Jenkins, DTU professor, via WIRED

The clearest signal today is not a new model release, but ASML's order book: the physical buildout of AI infrastructure is still accelerating faster than almost anyone expected.