A government stake in AI lands on private labs like OpenAI and Anthropic, not on the public chip layer they all depend on. The two cleanest ways to own the trade are Broadcom and TSMC, the toll booths every lab pays no matter who sits on its cap table.1
Key takeaways
- The insulated, highest-upside AI exposure is the silicon-and-packaging layer the labs cannot bypass. Broadcom (AVGO, 19.5x forward earnings, 67% FY27 EPS growth) and TSMC (TSM, 21.7x forward earnings, 25% FY+1 EPS growth) sell to every frontier lab regardless of who owns it.
- Washington already took a 9.9% stake in Intel and claims 15% of Nvidia's and AMD's China AI-chip sales. The stake mechanism now floated for OpenAI and Anthropic hits private equity holders, not public suppliers.
- H100 rental rates rose 31% over the past year while compute stayed sold out, so the binding constraint is fabrication and packaging, not model ownership. Whoever owns the labs still buys chips from the same two companies.
The theme
Washington has spent the past ten months turning industrial policy into equity ownership. On August 22, 2025, the US took a 9.9% stake in Intel, converting $8.9 billion in federal funding—$5.7 billion of remaining CHIPS grants plus $3.2 billion from the Secure Enclave program—into 433.3 million shares at $20.47 apiece.6 The position is passive and non-voting, with a warrant for another 5% if Intel drops below 51% ownership of its foundry.
That was the template. Golden shares in US Steel, an equity position in MP Materials, and a 15% cut of Nvidia and AMD China chip revenue followed.
By June 2026 the conversation reached the AI labs themselves. Senior officials opened preliminary talks with OpenAI about a federal equity stake ahead of expected IPOs.7 Bernie Sanders introduced a bill proposing a one-time 50% equity tax on OpenAI, Anthropic, and xAI.8
The question for a public-market investor is narrow and answerable. If the government takes a piece of the AI labs, where does that risk actually land, and what stays clean?
It lands on the private labs and their future shareholders. OpenAI and Anthropic are not publicly traded, and xAI trades only inside SpaceX, so the dilution and governance friction fall on private or pre-IPO cap tables, not on the listed companies that supply them.
The cleanest public exposure is therefore the layer that gets paid regardless of who owns the labs. That layer is custom AI silicon and advanced-node manufacturing.
Broadcom designs the chips. TSMC builds them.
The consensus and where it breaks
Consensus treats the AI trade as one position that rises and falls together. The data says it already fractured.
Over the trailing year, Intel returned 463% and AMD 299% as the government-stake and compute-hardware narratives ran. Microsoft fell 17%, Meta fell 15%, and Palantir slipped 2.5% over the same window. The S&P 500 returned 26%.1
The second consensus error is the assumption that government involvement is uniformly bearish for the complex. It is bearish for the entity that gets diluted and conflicted. It is close to neutral for the supplier that sells into the buildout no matter who owns the buyer.
The third error is treating compute as a commoditizing cost that falls over time. H100 rental rates rose from $1.90 to $2.48 per hour over the past year, a 31% increase, with B200 capacity still scarce near $4.50.3 Compute is getting more expensive because demand outruns supply, and that supply constraint is fabrication and packaging.
The names that express it
The silicon every frontier lab designs around
Broadcom is the connective tissue across every frontier lab. It designs the custom XPU silicon behind Google's TPU program through 2031, Meta's MTIA accelerators, and OpenAI's first-generation custom chip targeting 1 gigawatt of deployment in 2027.
Q2 FY26 revenue hit $22.2 billion, up 48% year over year, with AI semiconductor revenue of $10.8 billion, up 143%.4 The stock fell roughly 14% on June 4 because guidance reiterated rather than raised the FY27 target of more than $100 billion in AI revenue. That reset the entry to 19.5x forward earnings against 67% projected FY27 EPS growth.2
The one foundry the whole buildout funnels through
TSMC manufactures essentially every leading-edge AI chip on earth, including Nvidia GPUs, AMD parts, Broadcom XPUs, Google TPUs, and Amazon Trainium. Its CoWoS advanced-packaging capacity is the single binding constraint on accelerator volume, and it is sold out through 2027.
The 2025 financials show $122.42 billion in revenue and $55.13 billion in net income at a 59.9% gross margin.5 TSMC trades at 21.7x forward earnings against 25% projected EPS growth, a PEG ratio below 1.0, and it is only now starting to use its pricing power.2
The counter-argument
The bear case against this framework has three real legs, and none of them are generic.
First, customer concentration. Broadcom AI revenue is heavily weighted toward Google and Meta today.
If Google accelerates its shift toward MediaTek-designed silicon for future TPU generations, Broadcom loses share at its single largest AI customer, and the $100 billion FY27 target depends on new programs from OpenAI and others ramping on schedule. A slip in any one of the six named XPU customers compresses the growth rate that justifies the multiple.
Second, the supplier-insulation thesis assumes AI capex holds. Broadcom and TSMC are insulated from who owns the labs, but not from whether the labs keep spending. If a government stake in OpenAI triggers funding discipline, or if the hyperscalers pull back capex after three years of vertical buildout, the order book that supports both names contracts.
TSMC is cyclical, and a single quarter of capex guidance cuts from Microsoft, Amazon, Google, and Meta would hit the foundry utilization directly.
Third, and most serious, is geopolitical tail risk. TSMC insulation from government equity policy is real, but it is concentrated in Taiwan, where roughly 90% of its leading-edge capacity sits.
A cross-strait disruption breaks the thesis outright in a way no balance sheet can hedge. Arizona fabs and CoWoS qualification in the United States during 2026 partially de-risk this, but the bulk of the constraint remains an island the United States does not control.
There is also a scenario where the stake mechanism extends beyond the private labs. If a future Congress applies a Sanders-style equity tax to public chip suppliers rather than just to OpenAI, Anthropic, and xAI, the insulation argument weakens. That bill is unlikely to pass with the current vote math, but the Intel precedent means the probability is no longer zero, and a 15% revenue cut already applies to Nvidia and AMD China sales.
Key data
| Company | Ticker | Price | Fwd P/E | FY+1 EPS gr. | Tgt upside | 1Y ret. |
|---|---|---|---|---|---|---|
| Broadcom | AVGO | $376.71 | 19.5x | +67% | +39% | +52% |
| TSMC | TSM | $425.83 | 21.7x | +25% | +11% | +101% |
| Nvidia | NVDA | $207.41 | 16.3x | +42% | +44% | +44% |
| Meta | META | $600.21 | 17.2x | +6% | +38% | -15% |
| Microsoft | MSFT | $393.83 | 20.4x | +16% | +43% | -17% |
Source. Prices and returns as of June 16, 2026. Forward P/E computed from price divided by FY+1 consensus EPS. Sources listed in endnotes.
Catalyst map
Q3–Q4 2026
OpenAI stake decision
A formal federal equity stake in OpenAI confirms the mechanism targets private labs and validates supplier insulation directly. Watch for terms that touch suppliers rather than just the lab cap table.
2026 quarterly prints
Broadcom FY27 AI ramp
The $100 billion FY27 target hinges on OpenAI and new XPU customers ramping on schedule. Each AI semiconductor number either confirms the trajectory or exposes the concentration risk.
Q2–Q3 2026 earnings
Hyperscaler capex guidance
Microsoft, Amazon, Google, and Meta capex guidance is the demand signal underneath both names. A coordinated pullback is the clearest thesis-breaker short of a Taiwan event.
The bottom line
A government stake in AI dilutes the labs, not the toll booths every lab pays to build, and those toll booths trade at 19.5x and 21.7x forward earnings with the buildout sold out through 2027. The thesis breaks on a Taiwan disruption or a coordinated hyperscaler capex cut, not on who ends up owning OpenAI. The evidence points to Broadcom and TSMC as the two names that capture AI upside while sidestepping the ownership question Washington just put on the table.
Sources
- yfinance (Yahoo Finance), daily price, market-cap, and one-year total-return data for AVGO, TSM, NVDA, META, MSFT, INTC, AMD and the S&P 500; accessed June 16–17, 2026. Supports all prices, 1-year returns, and the Exhibit 1 bar values. https://finance.yahoo.com
- yfinance (Yahoo Finance), FY+1 consensus EPS estimates, forward P/E, and mean analyst price targets; accessed June 17, 2026. Supports the forward P/E, EPS-growth, and target-upside columns and Exhibit 3. https://finance.yahoo.com
- Ornn Compute Price Index (OCPI), Ornn AI Inc., H100 SXM and B200 GPU-hour rental series (Bloomberg Terminal); accessed June 16, 2026. Supports the H100/B200 rental rates and Exhibit 2. https://ornn.com
- Broadcom Inc., Second Quarter Fiscal Year 2026 Financial Results, Form 8-K (Exhibit 99.1), filed June 3, 2026. Supports Q2 revenue of $22.2B (+48%), AI semiconductor revenue of $10.8B (+143%), and the reiterated >$100B FY27 AI target. https://www.sec.gov/Archives/edgar/analytics/0001730168/000173016826000051/avgo-05032026x8kxex99.htm
- Taiwan Semiconductor Manufacturing Company, FY2025 audited results (TWD), filed April 16, 2026. Supports revenue of ~$122.42B, net income of ~$55.13B, and the 59.9% full-year gross margin. https://investor.tsmc.com
- Intel Corporation, Form 8-K (Exhibit 99.1), filed August 22, 2025. Supports the 9.9% passive U.S. stake, 433.3M shares at $20.47, $8.9B in federal funding (CHIPS + Secure Enclave), and the 5% foundry warrant. https://www.sec.gov/Archives/edgar/analytics/0000050863/000005086325000129/a08222025form8-kex991.htm
- NOTUS, Senior U.S. Officials Eye Government Shares in AI Giants, June 4, 2026, and CNBC, June 5, 2026. Supports the preliminary federal equity-stake talks with OpenAI. https://www.notus.org/technology/trump-ai-stake-openai
- Sen. Bernie Sanders, American A.I. Sovereign Wealth Fund Act, introduced June 1, 2026. Supports the proposed one-time 50% public equity transfer covering OpenAI, Anthropic, and xAI. https://www.congress.gov
ProCap Insights is a research division of ProCap Financial. This report is for informational and analytical purposes only. It does not constitute investment advice and does not make buy, sell, or hold recommendations on any security. Nothing in this report should be construed as a solicitation or recommendation to buy or sell any financial instrument. Readers should conduct their own due diligence and consult a qualified financial advisor before making any investment decision.