TECH BRIEF

Daily Tech Brief

Daily Tech Brief from Front Research

April 16, 2026

Good morning from Front Research, here's what's moving in tech today.

• ASML raises 2026 revenue forecast to €36 to €40B on AI demand, but a light Q2 outlook and fresh China restrictions send the stock lower

• Meta commits to a 1GW custom AI chip buildout with Broadcom through 2029, running on the industry's first 2nm accelerator, and Hock Tan exits Meta's board

• Jane Street commits $7B to CoreWeave with a $1B equity stake and a $6B cloud deal, becoming the neocloud's first major financial services anchor

• TSMC reports Q1 at the top of its January guide with record $35.7B sales, with today's earnings call expected to confirm a step-up in 2026 capex and the 2nm ramp

Into the brief.

ASML Raises Full-Year Forecast on AI Orders, but Q2 Guide and China Rules Hit the Stock

• ASML reported Q1 net sales of €8.8 billion and net income of €2.8 billion, beating consensus, and lifted its 2026 revenue forecast to €36 to €40 billion from €34 to €39 billion

• Q2 revenue guidance of €8.4 to €9.0 billion came in below the €9.07 billion consensus, softening the headline beat

• Memory applications rose to 51% of system sales from 30% in the prior quarter, while South Korea surpassed China as the largest market at 45% of net system sales

• ASML expects to ship 60 low-NA EUV tools in 2026, up roughly 25% from 2025, with capacity for 80 in 2027

• Why it matters: The raised full-year guide confirms the AI capex cycle is not peaking, but the light Q2 guide and shrinking China mix are a reminder that the pull-through is uneven and geopolitical restrictions are reshaping the customer base

ASML reported first-quarter net sales of €8.8 billion and net income of €2.8 billion, beating consensus of roughly €8.5 billion in sales and €2.5 billion in profit, and lifted its 2026 revenue forecast to a range of €36 to €40 billion from €34 to €39 billion. CEO Christophe Fouquet said "the semiconductor industry's growth outlook continues to solidify, driven by ongoing AI-related infrastructure investments" and that demand for chips is outpacing supply, with customers pulling forward capacity expansion plans. (Reuters via Investing.com, ASML press release, Bloomberg)

Under the headline beat there were two complications. First, ASML guided second-quarter sales to a range of €8.4 to €9.0 billion, below the €9.07 billion consensus, a softer print than the raised full-year number implied. Second, CNBC reported the stock sold off on tightening US restrictions on shipments into China, with China sinking further in ASML's customer mix. South Korea surpassed China as ASML's largest market in Q1, hitting 45% of net system sales at €2.84 billion, up from 22% in the prior quarter, a clean readout of where memory capex is now flowing. (CNBC)

The mix shift inside the business is notable. Memory applications rose to 51% of system sales from 30% in the prior quarter, while logic fell to 49% from 70%, consistent with HBM and DDR5 capacity buildouts tied to AI servers. ASML said it expects to ship 60 low-NA EUV tools in 2026, up roughly 25% from 2025, with capacity to ship 80 in 2027.

Why it matters: ASML remains the cleanest single read on AI capex at the tool level, and the raised full-year guide confirms the cycle is not peaking. But a light Q2 guide and the China drag are a reminder that the AI pull-through is uneven quarter-to-quarter, and the geopolitical restrictions are starting to show up in the customer mix rather than just in rhetoric. Memory vendors (Samsung, SK Hynix, Micron) and Korean memory capex look like the cleanest derivative beneficiaries, while any Chinese foundry thesis needs another haircut.

Meta and Broadcom Lock In a Multi-Gigawatt Custom Chip Deal Through 2029

• Meta and Broadcom committed to an initial 1 gigawatt of Meta's MTIA accelerators co-designed with Broadcom, running through 2029, with scope for multi-gigawatt deployments beyond that

• Broadcom described the collaboration as producing "the industry's first 2-nanometer AI accelerator," anchoring multiple generations of Meta's custom stack

• Analysts frame the initial 1GW commitment as a $12 to $15 billion revenue opportunity for Broadcom over the next 24 months

• Broadcom CEO Hock Tan will step off the Meta board and transition to an advisor role, clearing the conflict-of-interest path for the expanded commercial relationship

• Why it matters: Broadcom now has two multi-year anchor customers (Google and Meta) at multi-gigawatt scale, validating the custom-silicon revenue engine, while Nvidia faces a second large hyperscaler meaningfully diverting AI workloads to custom silicon at a 2nm node

Meta and Broadcom announced a sweeping extension of their custom silicon partnership, committing to an initial 1 gigawatt of Meta's MTIA training and inference accelerators co-designed with Broadcom and running through 2029, with scope for multi-gigawatt deployments beyond that. Broadcom described the collaboration as producing "the industry's first 2-nanometer AI accelerator," which will anchor multiple generations of Meta's custom stack across accelerators and networking. (CNBC, Bloomberg, Meta newsroom)

Analysts frame the initial 1GW commitment as a $12 to $15 billion revenue opportunity for Broadcom over the next 24 months, sitting on top of the existing ramp. Alongside the deal, Broadcom CEO Hock Tan will step off the Meta board, transitioning to a formal advisor role on Meta's custom silicon roadmap; Meta had disclosed on April 8 that Tan would not stand for re-election at the 2026 annual meeting, and the announcement now makes clear the move was timed to clear the conflict-of-interest path for a far larger commercial relationship. (SiliconANGLE, Quartz)

The deal is the clearest signal yet that the hyperscaler custom silicon story is not a hedge but a first-class architecture. Meta said earlier this quarter its 2026 AI capex would land between $115 and $135 billion, nearly double 2025, and the Broadcom commitment locks in a material share of that spend away from merchant GPU suppliers.

Why it matters: Broadcom's custom ASIC business now has two multi-year anchor customers (Google and Meta) at multi-gigawatt scale, which validates the custom-silicon revenue engine the stock has been priced on and extends visibility. For Nvidia, this is the second large hyperscaler willing to meaningfully divert AI workloads to custom silicon, and the first at a 2nm node, which matters for how investors model the merchant GPU TAM through 2029. The Meta corporate governance signal (senior CEO moving from board to advisor explicitly to expand a commercial deal) is also a clean tell about where Meta sees its strategic priorities.

Jane Street Puts $7B Behind CoreWeave, Anchoring the Neocloud's First Big Financial Services Customer

• Jane Street signed a roughly $6 billion multi-year AI cloud agreement with CoreWeave and separately took a $1 billion equity stake at $109 per share, making it one of CoreWeave's five largest shareholders

• The deal includes access to Nvidia's Vera Rubin systems and the software stack to deploy them across multiple CoreWeave facilities

• Jane Street is the first financial services customer at this scale; CoreWeave's pipeline has been dominated by frontier AI labs and hyperscalers until now

• Why it matters: This is the first concrete data point that non-AI-lab enterprises are contracting AI capacity at a scale rivaling hyperscaler deals, diversifying CoreWeave's revenue and signaling that internal compute buildouts are no longer sufficient for the quant trading industry

CoreWeave announced that Jane Street, the quantitative trading firm, has signed a roughly $6 billion multi-year AI cloud agreement and separately taken a $1 billion equity stake at $109 per share, making Jane Street one of CoreWeave's five largest shareholders. Under the deal, Jane Street gets access to next-generation compute across multiple CoreWeave facilities, including Nvidia's Vera Rubin systems and the software stack to deploy them. (Bloomberg, CoreWeave press release, Yahoo Finance)

The contract adds to a book of business that now includes roughly $35 billion from Meta, $12 billion from OpenAI, and $6.3 billion in capacity commitments tied to Nvidia, all signed in the last year. Jane Street is notable for being the first financial services customer at this scale; until now CoreWeave's pipeline has been dominated by frontier AI labs and hyperscalers. Jane Street said the CoreWeave capacity will support both its own trading infrastructure and its push into deploying large AI models internally. (24/7 Wall St.)

Why it matters: This is the first concrete data point that non-AI-lab enterprises are now contracting AI capacity at a scale that rivals the hyperscaler deals. For CoreWeave, it diversifies revenue away from a handful of frontier labs and meaningfully extends the customer concentration story that has weighed on the multiple. For Nvidia, Vera Rubin has a visible next-customer pipeline beyond the obvious names. And for the quant and systematic trading industry, a $6 billion contract to a third-party AI cloud is a signal that internal compute buildouts are no longer sufficient to keep pace with model scaling, which has read-throughs to Citadel, Two Sigma, DE Shaw, and the proprietary trading arms of the largest banks.

TSMC Reports Q1 at the Top of Its Guide, With the Margin and Capex Prints Coming on Today's Call

• TSMC confirmed Q1 revenue of roughly $35.7 billion, landing at the very top of its January guide, up about 35% year over year, with March alone hitting a record NT$415.2 billion

• Wall Street consensus for today's call is EPS of roughly $3.30, a 55% year-over-year jump driven by AI accelerator demand and the early 2nm N2 ramp

• Investors are focused on the Q2 guide against a $37 to $38 billion consensus, whether management raises its "close to 30%" full-year growth framing, and whether 2026 capex moves above the $52 to $56 billion range

• Analysts expect Nvidia to overtake Apple this year as TSMC's largest customer by revenue, reflecting how aggressively the AI cycle has restructured the foundry's book of business

• Why it matters: TSMC's guidance on today's call will set the tone for the rest of semiconductor earnings season, and a raise to full-year growth or capex would validate ASML's raised guide and support the narrative that the AI cycle is re-accelerating

TSMC confirmed first-quarter revenue of NT$1,134.1 billion, or roughly $35.7 billion, landing at the very top of the $34.6 to $35.8 billion range it guided in January and up about 35% year over year. March alone hit NT$415.2 billion, up 45% year over year and 31% month over month, the strongest March on record. Wall Street consensus going into today's call is for EPS of roughly $3.30, a 55% year-over-year jump, driven by AI accelerator demand and the early commercial ramp of the 2nm N2 node that began mass production at Fab 20 and Fab 22 in January. (Reuters via Investing.com, TrendForce, Techi)

The full Q1 earnings call is scheduled for today at 14:00 Taipei time (02:00 ET). Investors are focused on three things: the Q2 revenue guide against a roughly $37 to $38 billion consensus; whether management holds the "close to 30% in USD terms" full-year 2026 growth framing or raises it; and whether 2026 capex stays inside the $52 to $56 billion range TSMC signaled in January, versus a potential upward revision given the AI-driven demand commentary from ASML yesterday. Options markets are pricing a 5 to 7% post-print move. (FinancialContent)

The strategic subplot is customer mix. Analysts expect Nvidia to overtake Apple this year as TSMC's largest customer by revenue, reflecting how aggressively the AI cycle has restructured the foundry's book of business. Apple remains the anchor for the upcoming A20 process but no longer commands exclusive pricing or queue priority at the leading node, a shift that has direct implications for Apple's silicon cost curve and for how Samsung Foundry, still fighting to get 2nm yields out of the mid-50s, fits into the supply picture. (TradingKey)

Why it matters: TSMC is the cleanest single indicator of AI infrastructure health, and its guidance on today's call will set the tone for the rest of the semiconductor earnings season. A raise to full-year growth or capex would validate ASML's raised 2026 guide and support the narrative that the AI cycle is re-accelerating into a second leg. A hold at prior ranges would lean the other way and likely pressure the AI-beneficiary complex (Nvidia, Broadcom, Marvell, and the AI datacenter infrastructure names) near term. Either way, the 2nm ramp commentary will shape how investors model Apple's silicon economics and Samsung Foundry's competitive position for the back half of the year.