Washington Picks the Bottlenecks

If You Only Read One Thing

The surprise is not that Washington wants more chips and more data centers; it is that Washington is starting to assign the bottlenecks. Apple Becomes Intel's Customer turns demand into industrial policy, while FERC Builds the Fast Lane shows grid access becoming national AI infrastructure, not ordinary utility queue management.

Apple Becomes Intel's Customer

The important part of the Apple-Intel story is not whether Apple suddenly loves Intel. It is that the U.S. government now owns part of Intel and is acting like a sales force.

Axios reported that Intel shares surged 11% after President Trump said Apple had agreed to work with Intel to design and build chips in the United States. Intel declined to comment, and Apple had not confirmed the arrangement. The market moved anyway because investors already knew the two companies had been discussing a foundry collaboration, and because the federal government holds roughly a 10% Intel stake from last year's CHIPS Act conversion.

Why it matters: This is industrial policy after the subsidy check. The first phase of CHIPS policy was capacity: give Intel money, keep fabs alive, and hope customers follow. The second phase is demand assignment. Apple does not need Intel-designed processors; it designs its own silicon. What it needs is optionality against a TSMC-centered supply chain, especially as AI demand crowds the advanced-node and advanced-packaging calendar. Intel needs the opposite: a customer credible enough to prove its foundry pivot is more than a government-backed turnaround story.

That is why the politics and the market reaction are inseparable. Axios noted that Intel's stock has risen roughly sixfold over the past year, making the government's stake far more valuable. Tom's Hardware wrote that any Apple work would likely be a foundry deal for lower-volume Apple-designed chips, not a return to Intel processors, and that speculation centers on Intel's 18A-P process for future MacBook Air and entry iPad Pro silicon. Intel said this week that 18A-P had entered risk production, with claimed performance and power improvements over 18A. A small Apple part would still matter because foundry credibility compounds: one marquee customer makes the next customer less crazy.

The counterargument is strong. This is not yet an Apple press release, a wafer-start schedule, or a guaranteed iPhone node. Intel still has to prove yield, cost, packaging, and delivery discipline. Apple also has every incentive to let Intel and Washington believe the relationship is larger than it is, because a second source improves its bargaining position with TSMC even if most flagship volume stays in Taiwan.

Room for disagreement: The charitable read is that government pressure is doing exactly what industrial policy is supposed to do: solve a coordination problem where no private buyer wants to be first on a risky domestic node. The skeptical read is that Washington is blurring ownership, procurement, and market signaling in a way that can reward Intel's stock before Intel has earned durable foundry economics.

What to watch: The test is not the next political post. It is whether Intel's next earnings call or foundry update names an external 18A/18A-P commitment with volume, timing, and product class attached. Without those three details, this is still option value, not supply-chain conversion.

FERC Builds the Fast Lane

FERC did not just make life easier for AI data centers. It told the electricity system that some loads are now strategic loads.

The Federal Energy Regulatory Commission said it issued show-cause orders under Section 206 of the Federal Power Act to all six regional grid operators under its jurisdiction, directing them to justify or reform the rules governing how data centers, factories, and other large users connect to the grid. AP reported that commissioners voted unanimously, that data centers would pay the full cost of needed grid upgrades, and that grid operators must respond within 30 days on power adequacy and within 60 days on integration plans.

Why it matters: The AI infrastructure fight has moved from chips and buildings to queues. A data center can raise billions, buy GPUs, sign a cloud customer, and still fail if it cannot get firm power on a useful timeline. FERC is trying to turn that bottleneck from a yearslong regional negotiation into a standardized federal pressure point. The phrase to remember is speed-to-power: not power generation itself, but the administrative and tariff machinery that determines who can use scarce transmission capacity first.

The order is more constrained than a pure hyperscaler gift. FERC's public framing says large users must connect faster while protecting ratepayers; AP says states retain control over retail rates and that large customers must pay their own upgrade costs. TechCrunch noted the catch: the directives can speed interconnection but do not create new generating capacity. Data centers already account for roughly 5% of U.S. electricity demand, according to EPRI figures cited by AP, and could triple by 2035. More than 4,000 U.S. data centers are operating, with another 3,000 planned or under construction. The queue is no longer a back-office utility process. It is a national allocation mechanism.

That makes FERC's move politically explosive. If hyperscalers get faster service and households see higher bills, the backlash will not distinguish between tariff design and electricity supply. If data centers bring their own behind-the-meter generation, accept curtailment, or fund dedicated upgrades, they become more like industrial campuses with negotiated grid rights. The likely result is a new divide inside AI infrastructure: companies with flexible load, private generation, and interconnection lawyers get treated like strategic customers; everyone else waits.

Room for disagreement: The strongest objection is that FERC is moving the line, not shortening it. Faster paperwork can make the queue more legible, but it cannot solve turbine delays, transformer shortages, local opposition, or fuel constraints. A federal fast lane may even intensify local fights if communities conclude that Washington is prioritizing data-center load over ordinary affordability.

What to watch: The 30- and 60-day grid-operator responses are the real policy. If they require dedicated upgrade funding, curtailment terms, or stronger interconnection security, FERC will have converted data-center speed into a new cost-allocation bargain. If they are vague, the order will become a litigation invitation.

The Contrarian Take

Everyone says: Washington is picking winners: Intel gets Apple, and data centers get a grid fast lane.

Here's why that's wrong (or at least incomplete): The better frame is that Washington is picking bottlenecks. Intel is not being handed a market; it is being handed the chance to prove a domestic foundry node with a customer that can still walk away. Data centers are not being handed free power; they are being pushed into a tariff structure where speed comes with cost, flexibility, and ratepayer politics. The policy tool is not subsidy alone anymore. It is controlled access to scarce infrastructure.

Under the Radar

• Jio's IPO is India monetizing the data layer. Jio Platforms filed draft IPO papers for a fresh issue of up to 27 crore shares, with proceeds largely earmarked for debt repayment, while Reliance says Jio has crossed 524 million subscribers and 268 million 5G users. The story is not just India's biggest potential listing. It is the public-market conversion of telecom scale, fixed wireless broadband, enterprise digitization, AI services, and a planned satellite broadband stack into one national digital platform. (Source)
• ASML is being asked to prove a negative. Bloomberg, republished by The Straits Times, reported that Commerce Secretary Howard Lutnick pressed ASML over concerns that an EUV machine reached China; ASML says no such system is in China and has catalogued 314 operating EUV machines with none there. The structural point is that ASML's monopoly is now a diplomatic liability as much as a commercial asset. (Source)

Quick Takes

• Apple's Brazil concession copies the DMA playbook. Apple will let Brazilian developers use alternative iOS app marketplaces and outside payment processing under its CADE agreement, while keeping notarization, marketplace authorization, and new commissions. The important part is geographic replication: once one regulator forces an App Store template, other regulators can demand local variants rather than start from scratch. (Source)
• The UK is turning search ranking into a regulated business input. The CMA imposed conduct requirements forcing Google to improve transparency and fairness in organic rankings, including AI Overviews, and to support search-data portability to authorized third parties. That is a quiet but important shift: search is being treated less like a private product and more like market infrastructure for businesses that depend on discovery. (Source)
• Accenture bought its way into industrial cyber software. Accenture will acquire a $3.25 billion majority stake in Dragos and buy runZero and NetRise, valuing the combined push at $4.18 billion. This is consulting moving upstream into operational technology software because power grids, factories, pipelines, and data centers are becoming the attack surface that boards cannot delegate to ordinary IT security. (Source)

The Thread

The day's stories are not all about AI, but they rhyme because they are all about controlled access. Apple may get Intel as a domestic second source because Washington wants a credible U.S. foundry. Data centers may get faster grid review because FERC is treating compute load as strategic infrastructure. Jio is turning 524 million subscribers into public capital. Apple is accepting Brazil-specific app distribution rules. Google is being told its search ranking is a business dependency. Accenture is buying software for the physical systems beneath the digital economy. The scarce asset is not only technology. It is permission to use the bottleneck.

Predictions

New predictions:

• I predict: By 2026-07-23, Intel will confirm at least one external 18A or 18A-P foundry customer commitment, but Apple will still avoid naming itself as a volume production customer. (Confidence: medium; Check by: 2026-07-23)

Generated on 2026-06-18 at 03:39 ET.