# semiconductor guidance — X 热门讨论 (2026-10-09 15:09 UTC)
## @rickyho_1989 (Ricky Ho) · 10-09 10:28 · ♥61 ↻9 💬5 The more I look at these charts, the more I think investors are still using an outdated mental model when they describe technology as simply one sector of the economy, because what is happening now is much larger than another strong earnings cycle for software and semiconductors, with technology increasingly becoming the dominant source of capital formation, corporate earnings growth, margin expansion, market capitalization and even incremental industrial activity across the United States, which is why I think the phrase “Tech Is the Everything Cycle” is actually much closer to reality than it initially sounds.
The first chart may be the most important one because technology-related investment now represents roughly 55% of all U.S. capital spending, up from only around 15% in 1960, and that tells us something fundamental about how the structure of the economy has changed, because technology is no longer simply selling products into the economy but is increasingly determining where the economy itself invests its next dollar of capital, whether through semiconductors, software, R&D, data centers, networking equipment, servers, power infrastructure or the enormous physical buildout required to support AI.
This becomes especially important in the current AI cycle because the distinction between “technology spending” and “industrial spending” is beginning to disappear, since building an AI model requires GPUs, but deploying those GPUs requires servers, networking, data centers, transformers, transmission lines, cooling systems, backup generation, gas turbines, copper, construction workers, financing and enormous amounts of electricity, which means one dollar of AI spending can create a chain of activity that extends far beyond the traditional technology sector and into industrials, utilities, commodities, construction and capital markets.
That is why I increasingly think the AI boom should not be viewed merely as another software cycle, because it is becoming an industrial investment cycle built around computing, one in which digital intelligence is creating demand for an enormous amount of physical infrastructure and pulling capital into parts of the economy that would otherwise appear completely separate from technology.
And this helps explain why an economy can remain surprisingly resilient even with the U.S. 10-year Treasury yield above 5%, because while higher rates are clearly tightening conditions for housing, leveraged private equity, commercial real estate and weaker companies dependent on cheap refinancing, the largest technology companies are simultaneously deploying hundreds of billions of dollars from internally generated cash flow into physical infrastructure, effectively creating a private-sector fiscal impulse that partially offsets monetary tightening elsewhere.
That is a very unusual configuration, because normally higher rates suppress capital investment as companies become more reluctant to borrow and projects become less attractive as hurdle rates increase, but the AI hyperscalers are not behaving like ordinary leveraged companies because Microsoft, Alphabet, Meta, Amazon and the broader technology complex possess extraordinary balance sheets, enormous operating cash flow and increasingly urgent strategic reasons to invest, which means their capital spending is being driven less by the cost of external financing and more by the perceived opportunity cost of not having enough compute.
This is one reason I think people asking why the Nasdaq can trade at all-time highs while the 10-year yield sits above 5% are sometimes looking at only half of the equation, because the discount rate has unquestionably risen, but the earnings, cash flows and addressable markets of the companies dominating the index have also changed dramatically, and if the numerator in an equity valuation equation grows faster than the denominator, stocks can continue rising even as rates remain uncomfortable.
The second chart reinforces the same point from another angle because technology represents only roughly 35% of the world’s 100 largest publicly traded companies by company count but approximately 60% by market capitalization, which means tech companies are not merely becoming more numerous among the global corporate elite but are becoming disproportionately valuable once they reach scale, and that is not an accident because the economics of the best technology businesses are fundamentally different from those of many traditional industries.
Software and digital platforms can scale revenue much faster than physical capital and labor requirements, network effects can strengthen competitive advantages as businesses grow, incremental distribution costs can approach zero, global markets can be accessed almost instantly and successful platforms can reinvest enormous amounts of cash into adjacent markets while maintaining extraordinarily high returns on capital, which means that once a technology company achieves sufficient scale, its economic value can expand far more rapidly than its physical footprint.
Now AI is adding another layer to that advantage because the world’s largest technology companies already possess the distribution, customer relationships, data, cloud infrastructure, balance sheets and installed user bases necessary to deploy AI faster than almost anyone else, meaning that the companies that benefited most from the previous computing era may also possess structural advantages in the next one, particularly because they can fund development internally, distribute new products through existing platforms and absorb early losses while smaller competitors are still trying to build the necessary infrastructure.
This is why market-cap concentration by itself does not automatically mean bubble, because concentration becomes dangerous when market capitalization separates from economic value creation, whereas these charts show that technology’s growing market weight has occurred alongside a growing share of profits, earnings growth and margins, which means at least part of the concentration is being driven by fundamentals rather than simply multiple expansion.
That distinction matters enormously because the third set of charts may be the strongest evidence of all, with technology now accounting for close to half of S&P 500 profits, technology earnings having increased by roughly 3.8 times more than non-tech earnings over the measured period and tech appearing to have contributed approximately three quarters of total earnings growth, which tells us that the market’s increasing concentration in technology is not happening in isolation from the income statement but is instead reflecting where the profits are actually being generated.
The index is becoming more technology-heavy because the profits are becoming more technology-heavy, and this is the part of the discussion that often gets lost when investors compare today’s market concentration with previous bubbles, because concentration in 2000 was accompanied by enormous speculation around businesses that frequently had limited revenue, negative cash flow and uncertain business models, whereas today’s dominant technology companies include some of the highest-margin, highest-cash-generating and most financially powerful corporations ever created.
That does not mean valuations cannot become excessive, because of course they can, and NVIDIA can become overvalued, Microsoft can become overvalued, Meta can become overvalued and Alphabet can become overvalued, since even the greatest business in the world can still be a terrible investment at the wrong price, but that is fundamentally different from arguing that the entire technology leadership cycle must reverse simply because technology has become a large percentage of the market.
Size alone is not a bear thesis, because the relevant question is whether the earnings supporting that size continue to grow, and so far, they are.
The margin charts make the same argument even more clearly because S&P 500 margins themselves are historically elevated, but the expansion in technology margins has been particularly dramatic, with tech approaching roughly 20% plus profit margins versus something closer to high-single digits or low-double digits for much of the rest of corporate America, which is one of the reasons technology can support far larger market capitalizations without necessarily requiring absurd valuation assumptions.
The market ultimately values cash flows, and a business producing US$100 of revenue at a 20% margin is economically very different from one producing the same revenue at a 5% margin, especially if the higher-margin company also grows faster, requires less incremental capital and can reinvest profits at superior returns, which is why the combination of growth, scale, margins and capital efficiency can justify a much larger share of total market value.
And AI could widen this difference further because the conventional AI discussion has focused heavily on how much technology companies are spending, while I think the more important medium-term question is what happens to corporate margins when AI begins reducing the cost of knowledge work, given that software development, customer service, marketing, legal work, financial analysis, administration, recruiting and countless other white-collar processes contain enormous labor costs that can increasingly be augmented or partially automated by AI agents.
The first stage of the AI cycle is therefore capex intensive, but the second stage could become margin expansive, because companies are spending heavily to build intelligence infrastructure today, while that infrastructure may eventually allow them to produce more output with fewer incremental workers, shorten development cycles, automate customer support, improve advertising conversion, optimize logistics and increase employee productivity, meaning that the payoff could appear not only in AI revenue but also in lower operating intensity across the entire economy.
That is why I think AI may ultimately matter more for margins than people currently appreciate, because the market is obsessed with the question of whether hyperscalers can monetize their US$100 billion-plus capex programs directly through AI subscriptions and cloud revenue, while some of the return may instead appear indirectly through higher productivity, better monetization, lower unit costs and stronger margins inside existing businesses.
Meta is a good example because AI does not need to become a standalone US$100 billion subscription business for Meta to earn a return on AI investment if recommendation quality improves, engagement rises, advertising conversion increases, creative generation makes advertisers more effective and personal agents eventually create new commerce opportunities.
Alphabet does not need every Gemini interaction to generate subscription revenue if AI improves Search monetization, Cloud consumption, Workspace retention and advertising relevance, while Amazon does not need every dollar of AI infrastructure to be monetized through AWS alone if AI improves fulfillment, inventory management, advertising and retail efficiency, which is why measuring the AI return purely through direct AI revenue may substantially underestimate the economic impact.
The final chart is also important because technology and energy have been leading S&P 500 earnings growth while the strength is simultaneously broadening, with roughly 94% of companies meeting or beating EPS expectations, which tells us that this is not currently a situation where a handful of technology companies are growing while the rest of corporate America is collapsing.
That does not guarantee the breadth will persist, and earnings beats partly depend on how conservative analyst expectations were before reporting season, but it does weaken the argument that the equity rally is being held together entirely by five or six mega-cap technology companies, because the AI cycle may actually be creating a broader multiplier in which technology buys semiconductors, semiconductor fabs buy equipment, data centers require construction, construction requires steel and cement, power demand requires turbines, transformers and transmission, higher electricity demand benefits utilities and energy producers, networking expansion benefits optical suppliers, cooling requirements benefit industrial equipment manufacturers, data-center financing generates business for banks and private credit and, eventually, AI productivity benefits companies outside technology that adopt the tools.
That is why the AI ecosystem keeps broadening even as the largest share of the economic value continues accruing to technology, because in some ways AI is beginning to function like the railroad, electricity and internet buildouts simultaneously, requiring huge physical infrastructure today while creating a general-purpose productivity layer that can eventually affect nearly every industry.
This is also where I think investors need to distinguish market concentration from economic concentration, because if technology were 60% of market capitalization but only 20% of earnings and 15% of investment, I would be extremely uncomfortable, but when technology simultaneously represents a growing share of capital expenditure, profits, earnings growth, margins, R&D and market value, then the concentration is telling us something about the changing structure of the economy rather than merely reflecting speculative enthusiasm.
The economy itself is becoming more technological, with a bank increasingly functioning as a software company with a balance sheet, an automaker increasingly operating as a software, battery and semiconductor company wrapped around a manufacturing operation, retailers increasingly competing through algorithms, logistics automation and digital advertising, drug discovery increasingly depending on computational biology, manufacturing increasingly relying on robotics, digital twins and AI optimization and media increasingly depending on recommendation algorithms and generative tools.
Every company is gradually becoming a technology company because technology is becoming embedded inside the production function itself, and AI accelerates that transition dramatically, which is why I think saying “technology is too big” is becoming increasingly similar to saying electricity was too important during electrification or the internet was becoming too central during digitization, because at some point something stops being a sector and starts becoming infrastructure, and technology is increasingly reaching that point.
There is, however, an important bearish interpretation of these charts that should not be ignored, because if technology now accounts for 55% of capital spending and the overwhelming majority of incremental investment, then the U.S. economy is becoming increasingly dependent on the continuation of the AI capex cycle, which means a significant slowdown in hyperscaler investment could have much larger macroeconomic consequences than it would have five years ago.
This is the reflexivity investors need to watch, because the stronger AI investment becomes, the more industries expand capacity around it, the more data centers are built, the more power plants, transformers, networking equipment and semiconductor capacity are justified, and the more financing is raised, the more capital becomes dependent on continued utilization.
Eventually, if AI monetization disappoints and hyperscalers simultaneously decide they have enough infrastructure, the slowdown would propagate far beyond NVIDIA, which is the genuine bear case, not simply that technology stocks are expensive but that AI capex eventually outruns economically useful AI demand.
That is why I continue watching cloud backlog, GPU utilization, token consumption, inference volumes, agent adoption, data-center lease rates, GPU rental prices, residual values, hyperscaler capex guidance and the revenue generated per dollar of AI infrastructure, because these indicators should reveal whether the investment cycle is creating durable economic value or merely building capacity ahead of demand.
So far, however, the evidence still points in the opposite direction, with GPU rental economics remaining resilient, older GPU residual values holding up, hyperscalers continuing to discuss insufficient capacity rather than excess capacity, cloud AI demand continuing to grow, token costs continuing to fall and making more use cases economic, agents dramatically increasing compute intensity and the largest technology companies continuing to produce extraordinary earnings growth despite spending unprecedented amounts on infrastructure.
That does not look like the end of the cycle, because it looks like the cycle is broadening.
This is why I remain LONG technology, particularly the companies controlling the bottlenecks, infrastructure, distribution and monetization layers of AI, because I think the biggest analytical mistake investors can make today is assuming that technology is simply repeating another historical sector boom when the evidence increasingly suggests something structurally different is happening.
Technology is becoming the largest source of U.S. capital formation, producing an increasingly disproportionate share of global corporate value, generating close to half of S&P 500 profits, driving the majority of incremental earnings growth and operating with structurally superior margins, while AI is now pulling semiconductors, power, industrials, commodities, construction and finance into the same investment cycle.
So perhaps the reason technology keeps outperforming despite high rates, enormous capex and constant bubble warnings is simpler than we think, because the market is not merely assigning a higher multiple to technology; the economy itself is becoming more technological.
And if AI turns intelligence into an increasingly abundant, cheap and scalable input to production, then this transition may still be much closer to the beginning than the end, because tech used to be a sector inside the economic cycle, while increasingly, tech is becoming the cycle itself.
Long tech. https://x.com/rickyho_1989/status/2108504790150127705