Semiconductors up 78% annually, software down 12% annually: The 'Liquidity Siphon' is playing out within tech stocks

marsbitPublicado a 2026-05-26Actualizado a 2026-05-26

Resumen

Semiconductor ETFs like SOXX have surged 78.5% year-to-date, while software ETFs like IGV have dropped 12.5%, creating a record performance gap exceeding 90 percentage points. This reflects a major "liquidity suction" within tech stocks, with capital flooding into semiconductors as software faces selling pressure. Driving the semiconductor boom are staggering capital expenditure plans from hyperscalers like Microsoft, Alphabet, Amazon, and Meta, whose combined 2026 capex is projected near $700 billion. This fuels demand for chips, with companies like SanDisk (up 426%), Intel (up 222%), and Micron (up 154%) leading the S&P 500. In contrast, major software firms like Microsoft, Adobe, and Salesforce are all down over 17% year-to-date. The software sector faces a dual challenge: capital is being redirected to semiconductors, and the rise of AI agents like Claude Code threatens traditional SaaS business models, triggering a narrative of AI displacement. Key unanswered questions remain: How long can hyperscalers sustain their massive capex, given potential free cash flow pressures? And will capital eventually rotate back into the deeply oversold software sector? While some analysts warn of a potential semiconductor bubble akin to the dot-com era, the sector's powerful momentum continues, making market timing exceptionally difficult.

Author: Claude, Deep Tide TechFlow

Deep Tide Introduction: The Semiconductor ETF (SOXX) has surged 78.5% year-to-date, while the Software ETF (IGV) has fallen 12.5% over the same period, creating a performance gap of over 90 percentage points, a historically extreme level.

SanDisk leads the S&P 500 with a 426% annual gain, Intel has tripled, Micron is up 154%, while Microsoft, Adobe, and Salesforce have all dropped more than 17% year-to-date. The four hyperscale compute companies' combined capital expenditures for 2026 are nearing $700 billion. Capital is pouring into the chip supply chain like a black hole, while the software sector faces a double squeeze from AI displacement narratives and fund outflows.

Recently, a hot post on the Reddit investment forum overseas stated that semiconductor stocks are "basically a black hole sucking everything else in," which resonated widely.

Data confirms this intuition. As of the May 22nd close, the iShares Semiconductor ETF (SOXX) has returned 78.5% year-to-date, while the iShares Expanded Tech-Software ETF (IGV) has returned -12.5%. Two ETFs belonging to the same broad technology category show a year-to-date performance gap exceeding 90 percentage points.

According to Tickeron statistics, all software stocks in the S&P 500 are currently trading below their 200-day moving averages, while about 89% of semiconductor stocks remain above their 200-day moving averages. Both sectors fell in sync to the zero line during the 2022 bear market, after which their trajectories completely diverged. This split was not gradual but explosive.

SanDisk up 426% leads S&P 500, Intel's triple gain crushes AMD

The numbers are even more dramatic at the individual stock level.

According to Benzinga Pro data, SanDisk (SNDK) has risen about 426% year-to-date, making it the best-performing stock in the S&P 500 in 2026, following a 559% surge in 2025. This storage chip company, spun off from Western Digital, has seen NAND flash prices rise over 200% year-over-year driven by AI, with March quarter revenue growing 250% to $5.95 billion and non-GAAP gross margins reaching a high 78.4%.

According to 24/7 Wall Street, Intel (INTC) has risen about 222% year-to-date to $225, doubling AMD's gain. Intel's rebound comes from an extremely low base, coupled with progress on its 18A process node, rumors of Apple foundry orders, and yield improvement data disclosed by CEO Patrick Gelsinger in a CNBC interview. Short sellers have been severely squeezed; according to S3 Partners data, since the March 30th low point, Intel's market cap has increased over $440 billion, resulting in paper losses exceeding $12 billion for short sellers.

Micron (MU) has risen about 154% year-to-date, with a cumulative gain of 661% over the past 12 months. Earnings support this move; for the fiscal 2026 second quarter, revenue was $23.9 billion, up 196% year-over-year, with adjusted earnings per share of $12.20, far exceeding market expectations of $9.21. DRAM accounted for 79% of total revenue, with High Bandwidth Memory (HBM) being a key driver. SK Hynix Chairman Chey Tae-won even predicted that the memory chip shortage could last until 2030.

In contrast, NVIDIA (NVDA), the true "money printer" of AI compute, has risen about 8% to $15 year-to-date, performing far worse than the aforementioned second-tier semiconductor companies. According to The Motley Fool, NVIDIA's current forward P/E ratio is about 21.5x, almost on par with the S&P 500's 20.3x. This means the market is no longer paying a growth premium for NVIDIA; capital is instead flowing towards chip companies with lower valuations and greater potential upside.

$700 Billion in Capex: The Hyperscale Compute Companies' 'Arms Race'

Behind the semiconductor surge is real money.

According to data compiled by the Financial Times and multiple institutions, the combined capital expenditures for 2026 by the four hyperscale compute companies—Microsoft, Alphabet (Google's parent), Amazon, and Meta—are projected to be between $650 billion and $725 billion, nearly doubling from about $410 billion in 2025. This represents the largest concentrated infrastructure investment cycle in tech history.

According to Tom's Hardware, Jefferies analyst Brent Thill stated bluntly: "The AI economy is healthy. The bear thesis is garbage."

Specifically: Amazon leads with a single-quarter capex of $44.2 billion, with AWS growing 28%; Alphabet's Q1 capex was $35.67 billion, doubling year-over-year, with Google Cloud backlog jumping above $460 billion; Microsoft's calendar 2026 capex is projected at $190 billion, with about $25 billion attributed to price increases in memory chips and components; Meta raised its full-year capex guidance to $125-$145 billion.

According to statistics from Om Malik's blog, three of the hyperscale compute companies recorded significant non-cash investment gains in their Q1 earnings: Alphabet recorded $36.8 billion (primarily from Anthropic equity appreciation), Amazon recorded $16.8 billion, and Microsoft recorded $5.9 billion (from OpenAI). While capital expenditures are burning cash fiercely, the AI investment targets themselves are also appreciating.

Software Stocks Slaughtered by AI Displacement Narrative, IGV Posts Worst Drop Since 2008

The other side of the coin is the brutal collapse of software stocks.

According to The Motley Fool, after Anthropic released Claude Code in early 2026, the software sector experienced a sharp decline—the market's logic was not to reward AI innovation but to punish those SaaS companies potentially displaced by AI. IGV recorded its largest drop since 2008 at one point.

As of late May, Microsoft is down about 17% year-to-date, Adobe down about 32%, Salesforce down about 31%, and Shopify down about 26%. The S&P 500 Software & Services Index is about 21% below its 200-day moving average, the largest such deviation since June 2022. According to Goldman Sachs and other institutional data, short positions in mid-to-large software companies have surged sharply over the past three months, with cybersecurity and SaaS companies being the most heavily targeted areas for shorts.

Two layers of logic underlie this divergence. The first is direct capital siphoning: market liquidity is finite. When $700 billion in capex pushes chip stocks into parabolic moves, capital must be withdrawn from somewhere. As the author of that Reddit post said: "Fundamentally decent software companies just sit there or bleed slowly, while the semiconductor index goes vertical."

The second is a revaluation of the narrative. The rapid evolution of AI agents is prompting the market to re-evaluate the moats of SaaS business models: when AI can automate programming, form-filling, and customer service, how long can subscription models based on per-seat fees last? The Motley Fool points out that software companies likely to survive will need features like real data, proprietary workflows, and deep customer integration that are difficult for AI to replicate.

Cycle Peak or Structural Shift? Two Key Questions Remain Unanswered

That Reddit user ended the post with two questions, representing investors' ultimate doubts about whether the semiconductor rally can persist.

However, these two questions remain unanswered.

First: How long can the hyperscale compute companies' capital expenditures be sustained?

According to CNBC, Pivotal Research projects Alphabet's 2026 free cash flow will plunge nearly 90% from $73.3 billion in 2025 to $8.2 billion. Of Microsoft's $190 billion annual capex, $25 billion is consumed just by price increases in memory chips and components. These companies are betting future profits on AI revenue that has not yet fully materialized.

Second: Is software the next rotation target?

According to Bank of America's Chief Investment Officer Hartnett's previous judgment in the Flow Show report, software is one of the best contrarian long positions for Q2 2026, given the extreme deviation of the sector relative to its 50-day and 200-day moving averages.

However, this does not mean the semiconductor rally is over. The Philadelphia Semiconductor Index (SOX) recorded a historic streak of 18 consecutive daily gains on April 25th, rising about 45% during that period. According to Intellectia analysis, some veteran analysts are beginning to compare the current move to the 1999-2000 internet bubble, warning of a potential 25% to 30% correction. But SOX has been up in 22 of the past 23 trading days, setting 15 intraday record highs; this momentum is itself a signal.

As that Reddit user said: "I don't want to call a top because I've been slapped in the face too many times before. But the extreme concentration of gains in a single sector is starting to smell like the late stages of a cycle."

Preguntas relacionadas

QWhat is the central theme of the article regarding the performance of tech stocks in 2026?

AThe central theme is a 'liquidity siphon' or 'black hole' effect within tech stocks, where capital is being aggressively pulled from software stocks and funneled into the semiconductor sector, creating a historic performance gap.

QWhich two ETFs are highlighted to demonstrate the extreme divergence between the semiconductor and software sectors?

AThe iShares Semiconductor ETF (SOXX) and the iShares Expanded Tech-Software ETF (IGV) are highlighted. SOXX was up 78.5% year-to-date, while IGV was down 12.5%, a gap exceeding 90 percentage points.

QWhat is driving the massive capital expenditure from the 'hyperscale' companies, and what is the estimated total for 2026?

AThe massive capital expenditure is driven by the AI infrastructure 'arms race'. The combined capex of Microsoft, Alphabet (Google), Amazon, and Meta is estimated to be between $650 billion and $725 billion for 2026, nearly double the 2025 figure.

QWhat are the two main reasons given for the severe underperformance of software stocks?

AThe two main reasons are: 1) Direct liquidity siphoning, where funds flow from software into the surging semiconductor sector. 2) A valuation narrative shift, where AI's capabilities (like automated coding) threaten the traditional SaaS subscription business model, raising questions about its long-term viability.

QWhat two critical unresolved questions about the market trend does the article present at the end?

AThe two unresolved questions are: 1) How long can the hyperscale companies sustain their enormous capital expenditures? 2) Is software poised to become the next sector for a rotational rally, or is the semiconductor momentum still strong?

Lecturas Relacionadas

China's AI Fronts: From Yan'an to Midway

This article analyzes the competitive landscape of China's AI industry through a dual-front war analogy: the "Eastern Front" of business model competition and the "Western Front" of global strategic positioning. **The Eastern Front: The Scramble for Supply Lines and Monetization** The "Eastern Front" examines the contrasting strategies of three Chinese tech giants—Tencent, Alibaba, and ByteDance—in the face of AI's high marginal costs. Tencent integrates AI as a catalyst within its existing ecosystems (advertising, gaming, cloud) for monetization, prioritizing high-value scenarios over user growth. Alibaba bets on a full-stack, self-developed approach from chips to applications, aiming to control costs and ecosystem, though this requires immense patience and resources. ByteDance, with Doubao as its flagship, pursues a traditional traffic-driven, "super app" strategy but faces severe monetization challenges as its massive user base incurs unsustainable operational costs. The central challenge for all is building a reliable "supply line" (sustainable funding/profit) and achieving efficient monetization, moving beyond being mere "token factories." **The Western Front: "Preserving Land" vs. "Preserving People"** The "Western Front" frames a global strategic divergence. The U.S. model ("preserving land") focuses on closed-source, high-premium models (e.g., Anthropic) targeting lucrative enterprise markets. China's strategy ("preserving people") leverages open-source models (e.g., Alibaba's Qwen, DeepSeek) and extremely low pricing to attract global developers and capture long-tail markets, akin to a "surround the cities from the countryside" approach. The goal is to make Chinese models the default infrastructure, locking in future ecosystem value. However, the critical test is whether this open-source ecosystem can achieve a commercial闭环, converting developer adoption into tangible revenue (e.g., via cloud services), and bridging the monetization gap with Western models that charge for value, not just tokens. **Conclusion: The Long March from Factory to Brand** The article concludes that China's AI industry possesses technology, users, and scenarios but must integrate them to create and capture value. Its ultimate success depends on navigating both fronts: companies must establish sustainable monetization on the Eastern Front, while the industry's Western strategy must evolve from simply "preserving people" (developer adoption) to truly "preserving both people and land" — transforming open-source ecosystem dominance into commercial success and premium brand value. This journey from being a "token factory" to a "value highland" will require strategic patience and the ability to outlast competitors in a prolonged contest.

marsbitHace 6 min(s)

China's AI Fronts: From Yan'an to Midway

marsbitHace 6 min(s)

A History of Technological Evolution Powered by Electricity: Aluminum, Bitcoin, and AI

The journey from the Rockdale aluminum smelter in Texas to space-based data centers illustrates a core economic principle: whoever controls the cheapest electricity dictates the use of computing power. The evolution is clear. Old industrial sites with pre-existing, high-capacity power grids are being repurposed. In Rockdale, a former Alcoa plant now houses vast Bitcoin mining rigs, which are increasingly being replaced by AMD chips for AI training. The logic is purely financial: while smelting aluminum yields $0.17–0.27 per kWh and Bitcoin mining $0.05–0.11, AI inference on H100 GPUs generates $1.27–3.67 per kWh. Recent deals confirm the rush for power infrastructure. Riot Platforms leases space to AMD; TeraWulf bought an old Kentucky aluminum plant for its grid; NYDIG secured a New York site for its cheap hydropower to mine Bitcoin. As AI giants like Anthropic, Microsoft, Google, and Amazon aggressively expand, they now directly compete with crypto miners for the same industrial power resources, often outbidding them. This has led to a decline in Bitcoin's global hash rate and a wave of miner conversions to AI data centers. This "digital resource curse" extends globally. Gulf nations, long offering subsidized power to attract heavy industry like aluminum, are now pivoting to become AI and cloud computing hubs—exporting computational power instead of physical commodities. Similarly, Bhutan halted its sovereign Bitcoin mining to sell hydropower directly to India for a steadier return. The frontier is space. Projects like Starcloud plan orbital solar-powered data centers, leveraging constant sunlight and natural cooling, with Bitcoin mining as a secondary use for surplus power. Even consumer brands are transforming; Allbirds shifted from footwear to AI infrastructure, causing its stock to surge. Meanwhile, crypto projects like Bittensor, Render, and Akash propose a decentralized alternative, creating markets to aggregate distributed, idle computing resources from individual hardware. The underlying infrastructure—the power grid—remains constant. As profit margins shift, the facilities built upon it will continue to evolve, from aluminum to Bitcoin to AI and beyond, always chasing the highest yield per kilowatt-hour, whether in Texas, Abu Dhabi, or low Earth orbit.

marsbitHace 15 min(s)

A History of Technological Evolution Powered by Electricity: Aluminum, Bitcoin, and AI

marsbitHace 15 min(s)

Conquering is easy, governing is hard: Polymarket must bow to regulations to plant its flag globally

Polymarket, a decentralized prediction market platform, faces significant regulatory hurdles in its global expansion. Its "permissionless" model, which bypasses traditional identity and financial controls, has led to widespread crackdowns. India recently blocked the site, categorizing it as illegal online gambling under new 2025 laws. Brazil also banned it and similar platforms, though it simultaneously authorized a regulated, investor-only version on its national exchange. Across Europe, countries like France, Portugal, and the Netherlands are enforcing bans based on existing gambling and financial regulations. To enter key markets, Polymarket is adopting a pragmatic, compliant approach. In the U.S., it paid a $1.12 million fine, acquired a CFTC-licensed exchange, and now operates a regulated, KYC-mandatory platform for American users. It also secured a major investment from Intercontinental Exchange (ICE), which will distribute its prediction data to institutional investors. In Japan, where gambling laws are strict, Polymarket has begun a long-term lobbying effort, aiming for legalization by 2030 through building institutional partnerships and community presence. Despite these challenges, the prediction market industry is booming, with global volume projected to surge from $51 billion to potentially $1 trillion by 2030. Polymarket's core dilemma remains: adapting its decentralized, anonymous model to fit within sovereign regulatory frameworks focused on licensing, consumer protection, and anti-money laundering rules. Its survival in each market depends on navigating this complex political and legal landscape.

marsbitHace 18 min(s)

Conquering is easy, governing is hard: Polymarket must bow to regulations to plant its flag globally

marsbitHace 18 min(s)

It's Easier to Conquer than to Govern: Polymarket Must Bend to Every Rule to Plant Its Flag Globally

Polymarket, a decentralized prediction market platform, is facing significant regulatory hurdles as it expands globally, illustrating the tension between permissionless, crypto-native platforms and national legal frameworks. The platform, which allows users to bet on event outcomes, was recently blocked in India under new online gambling laws and faces similar outright bans in Brazil and Ukraine, the latter citing moral objections to wagering on active war events. In Europe, countries like France, the Netherlands, and the UK are restricting access by enforcing existing gambling and financial derivatives regulations, forcing Polymarket to geo-block users or operate in view-only modes. To navigate this complex landscape, Polymarket is adopting a market-by-market, compliant strategy. In the U.S., it paid a $1.4 million CFTC fine, acquired a licensed exchange (QCEX) for $112 million, and now operates a regulated U.S. entity with strict KYC, abandoning anonymity. It also secured a major investment from Intercontinental Exchange (ICE), which will distribute its prediction data to institutional investors. In Japan, a high-potential market, it has begun a long-term lobbying effort aiming for legalization by 2030, acknowledging the country's strict anti-gambling laws and slow regulatory processes. The article concludes that while the global prediction market is growing rapidly—projected to reach $2.4 trillion by 2030—Polymarket's core challenge is transforming its decentralized model to fit sovereign regulatory systems built on licensing, consumer protection, and anti-money laundering rules. Its survival depends on proving its legitimacy in each jurisdiction.

链捕手Hace 23 min(s)

It's Easier to Conquer than to Govern: Polymarket Must Bend to Every Rule to Plant Its Flag Globally

链捕手Hace 23 min(s)

Trading

Spot
Futuros
活动图片