CPI, PPI and AI Capex: Tech Retest or Lower High

The Market in One View

• CPI: Headline inflation rose 0.1% month over month and 3.4% year over year in July. Core CPI rose 0.2% for the month and 2.5% over the year.
• PPI: Final-demand producer prices were unchanged in July, below the 0.2% consensus shown by MarketWatch. The annual rate slowed to 4.7% from 5.5% in June.
• Employment and purchasing power: Payrolls fell by 23,000 in July, May and June were revised down by a combined 103,000, and real average hourly earnings fell 0.1% month over month.
• Federal Reserve: The July meeting held the federal-funds target at 3.50%–3.75% by a 9–3 vote. Softer inflation and employment data support a pause, but the internal policy debate is not finished.
• Market structure: The S&P 500 closed at a record 7,798.99 on August 13. The Nasdaq reached 26,803.03, but the Philadelphia Semiconductor Index remained about 15% below its June 22 record as of August 12.
• AI capital spending: AI infrastructure continues to support earnings, semiconductors, networking, construction and utilities, but it is also absorbing cash, increasing depreciation and adding to the supply of corporate bonds.

July CPI: A Better Monthly Number, Not an All-Clear Signal

The Bureau of Labor Statistics reported that headline CPI increased only 0.1% in July after falling 0.4% in June. The annual rate eased from 3.5% to 3.4%. Core CPI increased 0.2% month over month and slowed from 2.6% to 2.5% year over year.

The composition was constructive. Energy fell 1.5% for the month, gasoline declined 2.9%, and shelter increased only 0.1%. Shelter still accounted for roughly two-thirds of the monthly headline increase, but the rate was materially calmer than during the spring inflation shock.

However, the annual energy index remained 14.7% higher, gasoline was up 24.6% year over year, food increased 3.0%, and shelter increased 3.2%. In other words, July delivered disinflation at the margin, not price stability across the household budget.

The distinction matters because the BLS real-earnings report showed that real average hourly earnings fell 0.1% from June to July and were 0.2% lower than a year earlier. Real weekly earnings were unchanged for the month and only 0.1% higher year over year. Consumers therefore received inflation relief without a clear improvement in individual purchasing power.

July PPI: The Headline Cooled, but the Pipeline Is Mixed

The BLS Producer Price Index was unchanged in July, following a revised 0.1% decline in June. Final-demand goods fell 0.7%, helped by a 3.1% fall in energy and a 5.7% decline in gasoline. Final-demand services rose 0.2%.

The soft headline reduced immediate rate-hike pressure, but the internals were not uniformly disinflationary. Final demand excluding food, energy and trade services rose 0.4% for the month and 4.7% over the year. Portfolio-management prices increased 6.5%, while construction prices advanced 2.2%.

This is why CPI and PPI should be read together. CPI says the direct monthly burden on consumers eased. PPI says goods and energy pressure fell, but important service, construction and investment-related costs remain. The combination supports patience from the Fed, but it does not prove that the inflation problem has ended.

The Federal Reserve Has More Reason to Pause—and More Reason to Keep Watching

The Federal Reserve held rates at 3.50%–3.75% in July, with three dissenters. After the July payroll loss, softer CPI and flat PPI, another immediate rate increase would risk tightening into weaker labor income and reduced demand.

That does not mean a rate cut is close or that the debate has disappeared. Cleveland Fed President Beth Hammack continued to argue for tighter policy, while Richmond Fed President Tom Barkin described a future hike as an open question. The market’s base case has shifted toward a September hold, but the August inflation releases will arrive before that meeting.

The most important point is that the federal-funds rate is no longer the only source of tightening. Even if the Fed pauses, rising long-term real yields can increase mortgage rates, corporate borrowing costs and the discount rate applied to technology earnings. Financial conditions can therefore tighten without another 25-basis-point policy move.

For a deeper discussion of why a pause may be preferable to a hike in the current mix of jobs, inflation and oil, see Fed Rate Pause or Hike? July CPI, Jobs, PPI and Oil Risks Point to a Better Policy Path.

Technology Near the Highs: Retest in Motion or Lower High?

Thursday’s close gave the bulls an important result. According to Reuters, the S&P 500 gained 0.65% to a record 7,798.99, while the Nasdaq advanced 0.81% to 26,803.03. Advancing S&P 500 stocks outnumbered decliners by 1.7 to one, and seven of the eleven sectors rose.

The move was positive, but it was not complete confirmation. Trading volume remained below its recent average, and the semiconductor complex still carried visible damage from the June–July reversal. On August 12, Reuters reported that the Philadelphia Semiconductor Index was still about 15% below its June 22 record, even after a 2.5% daily advance.

This divergence creates the present technical question:

• A successful retest and breakout would require the Nasdaq to move through the prior high area and hold above it, with semiconductors, software, market breadth and volume confirming the move.
• A lower high would become more credible if the Nasdaq rejects the prior peak, semiconductor leadership stalls, new highs contract, volatility rises and long-term yields resume their advance.
• A range remains possible if inflation stays moderate but growth, oil and AI-spending concerns prevent investors from paying higher valuation multiples.

A lower high should not be declared before price rejects the resistance area. Equally, a retest is not a durable breakout merely because a headline index briefly trades above a previous level. Acceptance, participation and follow-through matter.

What AI Capex Is—and Why It Now Matters to the Whole Market

Capital expenditure, or capex, is money used to acquire or build long-lived productive assets. In the AI cycle, that includes GPUs and CPUs, servers, memory, networking equipment, data centers, cooling systems, land, power connections and supporting energy infrastructure.

The expense does not normally hit the income statement all at once. The asset is capitalized and then charged through depreciation over its estimated useful life. That accounting delay is central to the current debate: today’s cash expenditure can support tomorrow’s revenue, but it can also create years of depreciation, power and operating costs before the commercial return is proven.

Semiconductor PPI: Hidden Inflation Inside AI Capex

The flat July final-demand PPI concealed exceptional price increases in parts of the U.S. semiconductor and electronic-component industry. These detailed indexes provide a more direct view of the hardware inflation confronting the AI investment cycle.

An important technical distinction is required. The BLS Producer Price Index measures changes in the selling prices received by domestic producers for their output. It is not a direct accounting measure of manufacturers’ production costs. Nevertheless, these selling-price indexes are highly relevant to data-center purchasers because semiconductors, printed circuit boards and loaded assemblies are incorporated into servers, networking equipment, storage systems, cooling controls and other data-center hardware.

Source: U.S. Bureau of Labor Statistics PPI data via the Federal Reserve Bank of St. Louis. Industry indexes are not seasonally adjusted. Year-to-date changes compare July 2026 with December 2025 and may be revised.

The movement in these indexes creates three connected consequences:

1. Higher data-center hardware costs: Rising component selling prices can increase the acquisition cost of servers, networking systems and supporting equipment.
2. More capital required for the same capacity: Hyperscalers may need to increase capex simply to purchase the same number of components, rather than to create proportionally more computing capacity.
3. A larger future depreciation burden: Higher purchase prices increase the capitalized cost of new equipment. If useful lives and depreciation methods remain unchanged, annual depreciation on that new equipment also increases.

This means an increase in reported AI capex does not necessarily represent an equivalent increase in productive capacity. Part of the increase may compensate for component-price inflation. Cash flow is reduced when the equipment is purchased; depreciation reaches the income statement over subsequent years; and replacement capex may arrive before the original equipment has produced its expected return.

The largest component increases require careful interpretation. A 177% increase in a specialized printed-circuit-assembly index does not mean every AI server costs 177% more. These domestic-producer series can be influenced by product mix, technical specifications, contract repricing, reporting changes and comparative base effects. Data centers also use imported components that are not fully represented by domestic industry PPI. The direction of the signal is nevertheless important: electronic-component inflation is materially stronger than the flat final-demand headline suggests.

How AI Capex Feeds Through the Economy and Markets

1. Immediate demand: Orders flow to chipmakers, memory producers, networking suppliers, construction firms, utilities and data-center operators.
2. Corporate earnings: Cloud capacity, AI services, advertising tools and enterprise software can produce revenue and large contracted backlogs.
3. Cash-flow pressure: Capital spending reduces free cash flow before depreciation reaches the income statement.
4. Inflation and bottlenecks: Heavy simultaneous investment can raise prices for advanced components, skilled labor, construction capacity and electricity.
5. Bond supply and yields: Companies increasingly finance the build-out through debt. More corporate issuance competes with heavy government borrowing for investor capital.
6. Future supply and productivity: If the infrastructure produces useful, scalable services, greater capacity and productivity can eventually become disinflationary. If demand disappoints, the result is overcapacity, impairments and lower returns.

The scale is already large enough to affect macro markets. Reuters reported that Goldman Sachs expected 2026 capital spending by Alphabet, Amazon, Meta, Microsoft and Oracle to approach $800 billion. Microsoft reported $41 billion of quarterly capex, with roughly two-thirds directed to shorter-lived assets such as CPUs and GPUs, and specifically noted the impact of higher component pricing. Alphabet reported $44.9 billion of second-quarter capex, with approximately 60% of its technical-infrastructure investment directed to servers and 40% to data centers and networking equipment. Alphabet raised its 2026 capex range to $195–205 billion. Meta guided to $130–145 billion for the year while second-quarter free cash flow fell to $784 million.

There is genuine revenue support behind part of the spending: Microsoft’s cloud backlog reached $678 billion, while Alphabet reported strong Google Cloud growth and a $514 billion backlog. Nevertheless, returns, depreciation and financing must be monitored together. Revenue growth can validate the build-out, but it does not make the capital free.

The AI Economy Will Divide Into Two Main Commercial Layers

The developing AI economy is likely to separate into two fundamentally different business models: a capital-intensive infrastructure and platform layer, and a much broader AI-services distribution layer.

1. AI Data Centers and Infrastructure Platforms

The first group will own or control data centers, computing capacity, major foundation-model platforms and the supporting infrastructure. Only a limited number of financially powerful companies will be able to sustain the scale of investment required. These are likely to be the usual hyperscalers and major technology platforms.

They must absorb rising semiconductor and hardware prices, data-center construction costs, electricity demand, financing requirements, rapid technological obsolescence, replacement capex and increasing depreciation. Their scale creates a substantial competitive barrier, but their returns depend on keeping expensive computing capacity fully utilized and converting infrastructure investment into recurring revenue.

Infrastructure ownership may therefore produce the largest absolute revenue pools and cash profits, but it also carries the greatest balance-sheet risk. A company can report rapidly increasing AI capex while receiving much less additional computing capacity than the headline investment suggests if hardware prices are also rising.

2. AI-Related Service Distributors

The second group will consist of the much larger number of companies distributing AI-powered products and services to businesses and consumers. Rather than constructing their own data centers, these businesses can rent computing capacity, purchase access to AI models or incorporate AI into existing software and commercial services.

This layer can address the mass market through subscriptions, usage charges, advertising, financial services, professional tools, customer support, education, healthcare, media and industry-specific applications. Successful distributors may achieve higher returns on invested capital because they can scale revenue without carrying the full cost of data-center ownership, semiconductor replacement and infrastructure depreciation.

In economic terms, the infrastructure company owns the factory, while the service distributor owns the relationship with the customer. Ownership of the customer relationship can capture substantial value—but distribution alone does not guarantee profit.

There will be many AI-service distributors, creating intense competition. Companies offering easily replicated interfaces or simple model wrappers may face falling prices, low customer loyalty and limited margins. The strongest distributors will need durable advantages such as:

• An established customer base and low customer-acquisition costs;
• Proprietary data that improves the service;
• Integration into essential commercial or professional workflows;
• Recurring subscription or transaction revenue;
• A trusted brand, regulated position or specialist industry knowledge;
• The ability to use several AI providers rather than depending completely on one platform.

The infrastructure companies may continue to generate the largest absolute profits. However, differentiated AI-service distributors could earn superior margins and returns on capital because they sell AI to the mass market without financing the entire physical system behind it. The major technology companies may operate on both sides—owning infrastructure while distributing AI through cloud services, software, advertising and existing consumer platforms.

Component manufacturers will benefit from both protagonists in this new technological phase. Data-center owners require increasingly powerful semiconductors, memory, circuit boards and networking equipment, while the expansion of mass-market AI services creates the demand that keeps this infrastructure operating and growing. As AI changes how people live and work, component manufacturers become the essential bridge between physical computing capacity and the services distributed through it.

The newest market risk is circular. AI spending supports technology earnings and equity prices, but AI-related bond issuance may push real yields higher. Higher real yields then reduce the present value of long-duration technology profits and raise the financing cost of the next phase of AI investment. Reuters reported that Alphabet, Amazon and Meta had issued almost $220 billion of bonds in 2026, while the U.S. 30-year real yield was near 3%.

This directly extends the argument in AI Depreciation: The Hidden Cost Behind the Boom: the next stage of the AI trade will be judged not only by capex growth, but by revenue conversion, asset life, depreciation, free cash flow and return on invested capital.

Market Catalysts: This Week and Next Week

All times are Eastern Time. Release dates and times should be reconfirmed with the publishing agency before trading.

Source calendar: MarketWatch U.S. Economic Calendar. Monthly expiration date: Options Industry Council.

Catalysts for a Breakout—or a Lower High

Oil and Tariffs Remain the Exogenous Risks

Energy delivered much of July’s CPI and PPI relief, but that benefit can reverse quickly. Brent moved back toward $88 a barrel on August 14 as the United States threatened an indefinite blockade of Iran and shipping risks persisted around the Strait of Hormuz. A renewed oil spike would affect headline inflation, freight, consumer confidence, real income and bond yields at the same time.

Trade policy is another unresolved transmission channel. The reversal of broad tariffs and distribution of approximately $100 billion in refunds reduce part of the burden already paid by importers, but the closure of the de minimis exemption and new targeted duties preserve cost uncertainty. Tuesday’s import-price report will therefore be more useful than political claims: it will show whether external prices are easing or rebuilding pressure before goods reach the consumer.

For the wider trade-policy context, see Trump Tariffs Verdict 2026: Did the Trade War Work?.

Conclusion: The Retest Is Valid, but Confirmation Is Still Missing

July CPI and PPI were good enough to interrupt the rate-hike narrative. They were not strong enough to remove inflation risk, particularly while energy remains exposed to geopolitics, import costs remain uncertain and AI investment competes for capital.

The S&P 500 has already confirmed a new closing high. Technology is more nuanced. The Nasdaq is close to its earlier peak, but the semiconductor index remains below its June record. That makes the present move a legitimate retest in progress—not yet a confirmed lower high, but not yet a fully confirmed technology breakout either.

The decisive evidence should come from price participation and the macro transmission mechanism: semiconductors, breadth and volume on the equity side; retail sales, import prices, Fed minutes, oil and long-term real yields on the macro side; and revenue, free cash flow, depreciation and bond issuance on the AI side.

The AI sector must also be evaluated as two different economic models. A small group of infrastructure owners will carry the hardware, financing and depreciation burden, while a much larger distribution layer will sell AI services to the mass market. The infrastructure companies may generate the largest absolute profits, but differentiated service distributors may ultimately achieve the stronger returns on capital.

For traders, the practical rule is simple: price action must confirm the narrative. A clean break with participation would favor continuation. A rejection led by semiconductors and reinforced by rising yields would make the lower-high scenario substantially more credible.