Key Events of the Week: What Happened

From July 12 through July 18, 2026, five themes shaped the most important developments affecting the U.S. technology industry. AI infrastructure continued to expand. Technology companies explored ways to turn computing capacity into an external source of revenue. Enterprise technology budgets shifted toward AI-related hardware and cybersecurity. China’s AI and memory industries faced new tests of their commercial and financing capabilities. At the same time, AI development encountered clearer social and institutional limits.

Meta, TSMC, and ASML Continued to Expand AI Infrastructure

Meta announced plans to expand its Hyperion data center in Louisiana to 5 GW, well above its earlier target of more than 2 GW. The revised plan brings projected investment in the facility to more than $50 billion. The company also said it would invest an additional $1 billion in local roads, water systems, and wastewater infrastructure. Since construction began in late 2024, Meta has awarded more than $1.6 billion in contracts to Louisiana businesses.

TSMC reported a 77 percent increase in second-quarter profit to a record NT$706.6 billion. It raised its 2026 capital expenditure forecast to between $60 billion and $64 billion and announced an additional $100 billion investment in Arizona. TSMC said demand signals from its largest cloud customers remained strong. The company now expects revenue in U.S. dollar terms to grow by slightly more than 40 percent in 2026, up from its previous forecast of more than 30 percent. Demand also remains strong for its 3-nanometer and 2-nanometer processes and CoWoS advanced packaging.

ASML reported second-quarter revenue of €9.33 billion, exceeding market expectations, and raised its 2026 revenue forecast to between €43 billion and €45 billion. The company also plans to increase equipment capacity over the next two years to meet demand for advanced chip production and AI processors.

During the same week, NVIDIA announced partnerships with FANUC, Yaskawa Electric, and other Japanese companies to develop robotics and physical AI. Noetra, a company backed by the Japanese government, also plans to purchase 27,500 NVIDIA Rubin chips for AI infrastructure supporting manufacturing and robotics.

Taken together, these developments confirm that the physical buildout of AI infrastructure remains strong. Major technology platforms, chip foundries, semiconductor equipment makers, and computing platform providers continue to increase investment. Demand is also expanding beyond model training into agentic AI, CPUs, robotics, and other real-world applications.

Meta and SpaceX Sought to Sell Compute Externally as Markets Examined AI Asset Returns

Meta is reportedly discussing a two-year computing capacity agreement with Anthropic worth as much as $10 billion. The talks remain at an early stage and may not result in a transaction. Under the proposed arrangement, Anthropic would make monthly payments for access to Meta’s computing capacity. The deal would mark Meta’s first major effort to generate external revenue from AI infrastructure that has primarily supported its own models, advertising operations, and consumer products.

SpaceX is also reportedly in talks to provide the U.S. Department of Defense with billions of dollars in data center capacity. The company previously entered into computing arrangements with Google and Anthropic and is considering more direct competition with neocloud providers such as CoreWeave. The reported Pentagon discussions have not been independently confirmed.

Databricks signed a term sheet during the week for a strategic funding round that would value the company at $188 billion. The proposed financing suggests that private investors remain willing to support platforms with strong positions in enterprise data, analytics, and AI applications, even as public markets reassess some AI-related valuations.

Public markets, however, turned more cautious. The Philadelphia Semiconductor Index fell roughly 10 percent during the week and ended more than 20 percent below its late-June high. SpaceX shares traded below their $135 IPO price for the first time on July 15, touching an intraday low of $132.15 before recovering. By the end of the week, the shares were roughly 45 percent below their post-IPO peak.

Apple briefly overtook NVIDIA on July 17 to become the world’s most valuable company before NVIDIA regained the lead. This did not necessarily reflect a sudden deterioration in NVIDIA’s competitive position. It suggested that some investors were shifting toward companies with lower capital intensity, steadier cash flows, and established device and services ecosystems through which they could monetize AI.

Investors are still willing to fund AI infrastructure, but they are drawing a clearer distinction between the scale of a company’s buildout and the returns those assets can generate. Owning large numbers of GPUs and data centers no longer supports a premium valuation on its own.

IBM Plunged as AI Began to Reallocate Enterprise Technology Budgets

IBM shares fell about 25 percent on July 14 after the company released preliminary second-quarter results ahead of schedule. IBM said it expected to report revenue of $17.2 billion, up only 1 percent from a year earlier and below market expectations. Infrastructure revenue declined 7 percent. Software revenue rose 5 percent, slowing considerably from 11 percent growth in the first quarter. IBM expected GAAP earnings per share to fall 2 percent to $2.27. Adjusted earnings per share were expected to rise 5 percent to $2.93 but still fall below market expectations.

IBM said some enterprise customers had redirected capital toward servers, storage, and memory to secure supply ahead of expected price increases. The company acknowledged that it had underestimated the scale of the shift and responded too slowly. Several large transactions also failed to close within the expected period. The weakness was concentrated in the mainframe business and its related software portfolio.

The results reflected more than an execution problem at IBM. They also offered a clear example of how enterprise technology budgets are being reallocated. Higher spending on AI infrastructure does not necessarily lift every technology category. As servers, memory, storage, and cybersecurity receive higher priority, companies may postpone software purchases, consulting projects, mainframe upgrades, and other IT investments.

Netflix reported second-quarter revenue and earnings broadly in line with market expectations. Its third-quarter guidance fell slightly short of analysts’ forecasts, however, and the shares declined in after-hours trading. The company said generative AI had been used in about 300 titles, primarily in post-production.

Netflix and IBM faced different business conditions, but both illustrated the same market requirement. Using AI does not by itself prove that AI has created new economic value. Investors still need evidence of higher revenue, utilization, margins, or cash flow.

Kimi K3, DeepSeek, and CXMT Advanced China’s AI Industrial System

China’s Moonshot AI released Kimi K3, an open-weight model with 2.8 trillion parameters and a context window of one million tokens. Moonshot said Kimi K3 could compete with leading U.S. frontier models in selected tests of software development, complex-task performance, and GPU optimization. Several third-party evaluations also found it highly competitive in web interface development and multistep tasks. These results remain benchmark evidence rather than proof of equal performance across all workloads.

According to The Information, DeepSeek’s annualized revenue has reached between $400 million and $500 million, primarily through paid API access for enterprises and developers. The company is also reportedly discussing another funding round and preparing for a possible listing on Shanghai’s STAR Market. The Information reported that DeepSeek has maintained a gross margin of more than 50 percent while charging less for model access than some U.S. competitors. These figures have not been formally disclosed by DeepSeek.

ChangXin Memory Technologies, China’s leading DRAM producer, also moved forward with its STAR Market IPO during the week. CXMT expects to raise at least RMB 57.9 billion, or approximately $8.55 billion. Proceeds could increase to RMB 66.6 billion if the over-allotment option is fully exercised. CXMT is now the world’s fourth-largest DRAM supplier, with an estimated 7.7 percent share of the global market in 2025. It continues to trail SK Hynix, Samsung, and Micron in advanced DDR5 products, manufacturing yields, and HBM technology. Major Chinese technology companies, however, are increasing their purchases of domestically produced memory.

Apple had previously sought assurances from the U.S. government that restrictions would not suddenly prevent it from purchasing CXMT memory, according to the Financial Times. That request did not occur during this reporting period. Two U.S. lawmakers did write to the Commerce Department during the week, urging the administration not to facilitate U.S. purchases of Chinese memory and asking it to consider adding CXMT to the Entity List.

Kimi K3 provides evidence of improving model capability. DeepSeek is testing whether low-cost models can support a profitable API business and attract large-scale financing. CXMT is extending the same competition into memory production and domestic capital markets. Together, they suggest that China’s AI industry is moving beyond individual low-cost models. A broader system is beginning to form around models, paying customers, memory suppliers, and domestic sources of capital.

Data Center Expansion Faced Greater Local and Government Constraints

New York imposed a statewide pause on large new data centers that use at least 50 MW of power. The measure can remain in place for up to one year, making New York the first U.S. state to adopt a full moratorium of this kind. During the pause, the state Department of Environmental Conservation will stop issuing discretionary permits for projects whose applications have not already been deemed complete. State officials will use the period to develop consistent environmental standards for future data centers. Governor Kathy Hochul said large data centers could raise household utility bills, strain natural resources, and create uncertainty for local communities. She also plans to pursue legislation that would repeal sales tax exemptions for large data centers.

A Reuters investigation found that xAI had installed 59 natural gas turbines for its Colossus 2 project without securing federal clean-air permits. At least 57 of the turbines are located in Mississippi, across the state line from the Memphis data center they support. xAI and Mississippi environmental regulators have argued that the turbines are temporary and mobile and therefore do not require permits. The U.S. Environmental Protection Agency has taken a different position, stating that temporary turbines must obtain permits when their potential emissions exceed federal thresholds. The dispute is now part of ongoing litigation.

The controversy illustrates how some AI companies are building private, off-grid power systems to bring new computing capacity online more quickly. In some cases, the speed of development is exceeding the pace of environmental review and public participation.

On July 18, opponents of rapid data center expansion held 142 demonstrations across 42 states. Protesters called for greater transparency in project approvals, stronger protection of local resources, clearer community benefits, and greater accountability for developers.

During the same period, the White House formally launched Gold Eagle, an AI cybersecurity coordination initiative. The program brings together federal agencies, critical infrastructure companies, and open-source software partners to identify, prioritize, verify, and address software vulnerabilities discovered with AI.

These developments reflect two related forms of institutional pressure. Local governments and communities are scrutinizing how data centers use electricity, water, land, and public infrastructure. The federal government is establishing a system for faster vulnerability detection and remediation across critical infrastructure.

AI expansion now depends on more than capital, chips, and engineering. It also requires lasting support from local communities and credible governance at the state and federal levels.

Narrative Observation: What It Means

Taken together, these developments show that the questions investors are asking have changed.

AI Demand Remains Strong, but Markets Are Beginning to Separate Demand from Price

TSMC and ASML provided strong evidence that industry demand remains healthy. Meta’s decision to expand Hyperion to 5 GW also showed that major platforms still expect to need substantial computing capacity. The sharp correction in semiconductor stocks demonstrated, however, that strong underlying demand and further share price gains are separate questions.

When valuations already assume nearly perfect demand, even results that exceed expectations may not support another increase in share prices. AI infrastructure is therefore moving from supply validation toward price and return validation.

Investors still expect demand for AI chips to grow. They are now asking whether current valuations already incorporate several more years of rapid expansion and whether major cloud providers can continue increasing capital spending at the same pace.

Compute Is Moving from an Internal Cost to an Income-Producing Asset

The potential Meta and Anthropic agreement, along with the reported discussions between SpaceX and the Department of Defense, suggests that an external market for GPUs and data center capacity is beginning to form.

When a data center serves only one company’s internal models, its returns depend largely on the success of that company’s products. Leasing capacity to other model developers, cloud platforms, or government customers can improve utilization and produce a more diversified stream of contract revenue. Long-term customer agreements may also help technology companies obtain debt, private equity, and other forms of infrastructure financing.

Turning compute into a revenue-producing asset does not remove the underlying risks. Financing capacity will depend on early termination provisions, customer credit quality, the useful life of the equipment, and continued access to reliable, low-cost electricity.

Compute is beginning to resemble a conventional infrastructure asset, but it still carries the rapid obsolescence and changing demand associated with technology equipment.

AI Spending Is Redistributing Economic Value Across the Technology Industry

IBM’s preliminary results showed that AI investment does not increase every category of technology spending at the same time. Enterprise budgets are limited. As servers, storage, memory, and cybersecurity move higher on the priority list, companies may postpone software purchases, consulting projects, mainframe upgrades, and other IT investments.

AI is therefore affecting the technology industry in two ways. It is creating new demand while also redirecting spending that previously went elsewhere. Companies that capture a direct share of AI infrastructure budgets are likely to benefit. Software companies that can help customers increase revenue or reduce costs may also retain a high priority in enterprise budgets.

Companies that add AI to their product positioning without showing that customers will pay more may face greater pressure on both revenue and valuation. The central question is no longer whether a technology company has adopted AI. It is whether AI has made that company a higher priority in its customers’ budgets.

U.S. and Chinese Competition Is Moving from Individual Products toward Two Industrial Systems

Kimi K3 suggests that the capability gap between Chinese open-weight models and leading U.S. models may continue to narrow. DeepSeek’s reported revenue and financing plans also indicate that lower-priced models do not have to remain technical demonstrations. If inference costs are sufficiently low, they may support API businesses with attractive gross margins.

CXMT’s planned IPO extends the competition into memory and semiconductor supply chains. Model capability, inference revenue, domestic customers, memory supply, and local capital markets are beginning to reinforce one another. Purchases of CXMT memory by large Chinese technology companies can help the manufacturer expand production and improve its technology. Additional capital can then allow CXMT to provide more domestic supply to China’s AI industry.

U.S. restrictions may limit CXMT’s access to global customers and advanced manufacturing equipment. They may also encourage China to develop more independent systems for models, chips, memory, and capital formation.

Future competition will therefore involve more than OpenAI against DeepSeek or NVIDIA against individual Chinese chipmakers. It will increasingly take place between two AI industrial systems.

AI Expansion Requires Institutional Capacity as Well as Engineering Capability

Meta’s expansion of Hyperion shows that major platforms can still secure land, energy, chips, and capital.

New York’s moratorium, the dispute over xAI’s natural gas turbines, and nationwide protests show that the ability to build does not guarantee lasting permission to operate. Local communities want technology companies to explain how much electricity and water their facilities will use, whether projects will raise household utility bills, how many permanent jobs they will create, and who will pay for roads, electrical grids, and water infrastructure.

Gold Eagle also shows that the federal government increasingly views AI as a cybersecurity capability requiring formal coordination and oversight. Data center operators and model developers will need to do more than improve their technology. They must also demonstrate that they can operate within local resource limits, environmental regulations, critical infrastructure requirements, and national security frameworks.

The Momentum of Trust: Why It Matters

This week’s shifts in trust fell into four broad categories.

1. Confidence in underlying AI demand remained high, while confidence in some semiconductor and infrastructure valuations began to weaken.
2. Investors continued to believe that major platforms could build data centers and secure computing capacity. Local communities, however, became less confident in how these companies manage resource use, environmental costs, and public decision-making.
3. Commercial confidence in China’s AI models and memory industry increased. At the same time, the U.S. government’s security concerns about Chinese supply chains continued to rise.
4. Enterprises gave greater priority to AI and cybersecurity. Markets became less confident that traditional software providers, consulting firms, and mature technology companies could retain their previous share of enterprise budgets.

Confidence in AI Demand Continues to Rise

Investment plans from TSMC, ASML, Meta, and NVIDIA show that underlying demand for AI chips, data centers, and emerging applications remains strong. Investors have not broadly rejected AI infrastructure or concluded that the current buildout cycle is over. Enterprise customers and cloud platforms continue to require more advanced manufacturing capacity, packaging, CPUs, GPUs, memory, and data center capacity.

The market correction during the week should therefore not be interpreted as evidence that AI demand has suddenly disappeared.

Confidence in AI Valuations and Capital Returns Is Declining

The semiconductor selloff and SpaceX’s fall below its IPO price show that investors are beginning to question whether current valuations already assume too much future growth. They are also examining whether major cloud providers can sustain the current pace of capital spending and whether investments in data centers and GPUs will generate sufficient revenue, free cash flow, and returns on capital.

Capital remains available for AI development, but investors and lenders are beginning to demand clearer evidence of financial returns.

Commercial Confidence in China’s Alternative System Is Rising

Kimi K3, DeepSeek’s reported revenue, and CXMT’s planned IPO are strengthening the case that China can convert technical capability into revenue, supply capacity, and capital formation.

China has not caught up with the United States across every part of the AI stack, and CXMT has not reached the technological level of the leading HBM suppliers. Chinese companies are nevertheless showing that lower prices, domestic customers, policy support, and local capital markets can sustain an alternative system with room to expand.

Social and Institutional Trust in AI Expansion Remains Limited

New York’s moratorium, the dispute over xAI’s energy infrastructure, and nationwide protests show that local communities do not necessarily trust technology companies to bear a fair share of the costs created by data centers.

Meta’s commitment to invest in roads, water systems, and wastewater infrastructure suggests that large technology companies increasingly understand that data centers require more than regulatory approval. They also need lasting support from the communities in which they operate.

Gold Eagle is a federal effort to improve cybersecurity coordination and build confidence in the use of AI across critical infrastructure.

Governments and markets continue to support AI development. Technology companies will nevertheless need to provide greater transparency, clearer responsibility for costs and risks, and stronger evidence of effective governance.

The Coming Weeks: What to Watch

• Alphabet, Tesla, and Intel are scheduled to report earnings. Investors will focus on AI capital spending, cloud growth, chip demand, free cash flow, and returns on investment.
• IBM will release its full second-quarter results on July 22. Beyond confirming the preliminary figures, investors will look for management’s explanation of delayed large transactions, slower software growth, and how long the shift in enterprise budgets toward AI infrastructure may last.
• The outcome of the Meta and Anthropic discussions will also matter. If the companies reach an agreement, attention will turn to the amount of capacity involved, pricing, contract length, early termination provisions, and whether Meta plans to establish a broader external cloud business.
• Investors will also watch whether the reported talks between SpaceX and the Department of Defense produce a computing agreement. Further expansion of services for Google, Anthropic, government agencies, and other enterprise customers would help determine whether SpaceX’s data centers support a limited set of strategic partnerships or a new long-term business.
• CXMT is expected to list on Shanghai’s STAR Market on July 27. Investors will examine its valuation after trading begins, its use of proceeds, the pace of capacity expansion, and the market’s willingness to accept a high valuation relative to historical earnings.
• The U.S. Commerce Department may respond to lawmakers’ requests for tighter restrictions on CXMT and other Chinese memory suppliers. Any decision could affect Apple’s supply chain options as well as the pricing and bargaining power of Micron, SK Hynix, and Samsung.
• DeepSeek’s reported funding round will provide another important signal. The company will also need to offer further evidence that its lower-priced model APIs can produce sustainable revenue, gross margins, and customer retention.
• The White House may disclose more information about the companies participating in Gold Eagle, as well as its vulnerability management procedures and information-sharing rules. The June executive order also requires the government to develop a voluntary review framework for frontier models. The framework’s effect on model release schedules and the government’s treatment of open-weight models will be important to monitor.
• New York may turn its one-year pause into longer-term environmental, electricity, and tax policies for data centers. Other states may consider similar measures.
• TSMC, ASML, NVIDIA, and memory producers will need to continue providing strong evidence of demand while supporting their valuations. Investors will also need to determine whether the semiconductor correction reflects short-term deleveraging or the beginning of a longer period of return validation.

Conclusion

From July 12 through July 18, 2026, underlying demand for AI infrastructure and the pace of construction remained strong. Meta expanded its plans for Hyperion, TSMC raised its capital expenditure forecast, ASML prepared to increase equipment capacity, and NVIDIA extended AI into robotics and the physical world. The AI industry continued to expand.

The questions attracting investor attention, however, have changed. Markets began examining whether data centers and GPUs could maintain adequate utilization, whether computing capacity could produce stable contract revenue, and whether current semiconductor and infrastructure valuations already reflected too much future growth.

IBM’s preliminary results also showed that AI spending does not lift every technology company at the same time. As enterprises spend more on servers, memory, storage, and cybersecurity, they may postpone software purchases, consulting projects, and traditional IT upgrades.

Kimi K3, DeepSeek’s reported revenue, and CXMT’s planned IPO show that China’s AI competition is moving beyond individual model capabilities. An industrial system supported by model revenue, memory supply, domestic customers, and local capital markets is beginning to take shape.

New York’s data center moratorium, the dispute over xAI’s energy infrastructure, nationwide protests, and the launch of Gold Eagle further demonstrated that AI expansion requires more than chips, electricity, and capital. It also requires lasting support from local communities and credible government institutions.

Markets still believe in AI. That confidence is no longer based on scale alone. It increasingly depends on whether the entire system can produce sustainable returns and retain the support it needs to operate over the long term.

Note: AI tools were used both to refine clarity and flow in writing, and as part of the research methodology (semantic analysis). All interpretations and perspectives expressed are entirely my own.