Global AI infrastructure spending will reach ~$1.5 trillion through 2030. Critics question whether this makes economic sense.

The answer becomes clear when we examine just one sector: pharmaceuticals.

Lifestyle vs. Life-Saving Drugs

Lifestyle drugs (obesity market): $100 billion by 2030. Helpful, but not life-or-death.

Current cancer drugs: $200+ billion annually for treatments that extend life by months.

Now imagine: A treatment that cures or manages cancer long-term. Not extending life by months, but by years or decades. What would that market be worth?

Conservative estimate: $500 billion-$1 trillion annually. Why?

• 20+ million new cancer diagnoses worldwide annually
• At $50,000-$100,000 per patient (for a cure or multi-year remission), that’s $1-2 trillion in addressable market
• Realistic market capture (50% of patients, averaging $75k): $750 billion annually
• Near-100% patient uptake (vs. lifestyle drug resistance)
• Universal insurance coverage (no debate when alternative is death)

The Pharmaceutical Industry Context

Global pharmaceutical companies generate ~$700 billion in combined revenue annually, spending ~$300 billion on R&D. They desperately need breakthroughs—current methods have 90% failure rates and take 10-15 years per drug.

The comparison that matters:

• AI infrastructure buildout (2024-2030): ~$1.5 trillion total
• One breakthrough cancer treatment: $500B-$1T annually = $5-10 trillion over patent life
• Just 2-3 such breakthroughs: Pays for entire AI infrastructure multiple times over

Why This Works

If AI improves pharma R&D success rates by just 15-20%, that generates $45-60 billion in additional annual value. Over a decade: $450-600 billion—equal to the entire AI buildout cost.

But the real value is in breakthrough discoveries. AI accelerating 2-3 major life-saving drugs by 3-5 years creates $1.5-3 trillion in value from pharmaceuticals alone.

And this is just one industry.

The right question isn’t “can we afford this AI buildout?” It’s “can we afford not to?”

---

Sovereign Capability Monopoly

SpaceX should not be valued as a conventional aerospace company or even as a high-growth technology firm. It should be priced as a sovereign capability monopoly—the only American company capable of industrial-scale space transport, constellation deployment, and deep-space logistics. Under this framework, an $800 billion valuation appears not aggressive, but potentially conservative.

---

1. The Competitive Reality: A Sovereign Capability Monopoly

SpaceX is the only American player with reusable heavy-lift capability. No other U.S. company can match the cadence, cost structure, or scale required for next-generation space infrastructure.

The only Western competitor of any note, based in France, is:

• 10–20 years behind
• not reusable
• operating at a cost per kilogram that is multiples of SpaceX
• unable to launch Starship-class payloads
• unable to support lunar cargo, Mars cargo, or large-scale constellation deployment
• not capable of meeting U.S. Department of Defense requirements

China is not an option for U.S. military, intelligence, or sovereign-infrastructure needs under any circumstances.

This leads to several unavoidable conclusions:

• SpaceX is the only entity capable of deploying and maintaining next-generation American space infrastructure.
• It is the only company able to deliver large-scale satellite constellations economically.
• It is the only path to lunar and deep-space logistics at industrial scale.

This is not a normal monopoly. It is a sovereign capability monopoly.

---

2. What a Sovereign Capability Monopoly Means for Valuation

Companies that become irreplaceable components of national power are not valued using traditional aerospace or telecom multiples. Their valuation is driven by:

• irreplaceability
• strategic dependence
• control over future economic infrastructure
• monopoly scale
• government-anchored demand and long-term certainty

This is the logic behind valuations for companies like Nvidia (compute acceleration), AWS (cloud infrastructure), and key defense primes. SpaceX’s dominance is even more complete at comparable stages. There is no plausible Western competitor. There is no acceptable foreign competitor. There is no substitute architecture.

---

3. Starlink Under the Correct Framework

Starlink is approaching ~$12 billion in annual revenue with strong growth across consumer, enterprise, mobile, aviation, maritime, and defense markets. As a strategic, monopoly-grade communications layer, it justifies 20–35× forward revenue multiples.

This yields:

• $240–420 billion based on current fundamentals
• $500 billion+ if direct-to-device mobile and secure defense networks scale

Starlink is not just another telecom service; it is the only deployable, global, American-controlled broadband constellation at meaningful scale.

---

4. Launch, Starship, and Industrial-Scale Space Transport

The remainder of SpaceX—launch, Starship, logistics, and payload capacity—is the only system capable of:

• reusable heavy-lift
• industrial-scale constellation deployment
• lunar and deep-space transport
• rapid deployment of national-security payloads
• future orbital construction and depot/refueling logistics

This portion of the company supports a valuation of $350–500 billion, reflecting both current economics and the massive option value tied to future infrastructure markets that no other American entity can serve.

---

5. Combined Valuation Under Sovereign-Monopoly Assumptions

Scenario	Starlink	Launch/Starship	Total Valuation
Low	$240B	$350B	$590B
Base	$300–350B	$400–450B	$700–800B
High	$400–500B	$450–500B	$850B–$1.0T

Under realistic strategic assumptions, $800 billion is not a stretch—it is the midpoint of the expected range.

---

6. Conclusion: The Lower Bound Argument

SpaceX is the only American institution capable of industrial-scale launch, deployment, and maintenance of the space infrastructure that will underpin communications, defense, logistics, and exploration for decades. Its monopoly is structural, not temporary. It is grounded in physics, engineering, cost advantages, and national-security necessity.

Viewed through this lens, $800 billion appears less like an upper estimate and more like a lower bound for one of the most strategically important companies in the world.

---

The challenge is a complex mix of politico-economics, But in the long range the Chinese controlled economy/exchange rate may win. China may well become the last man standing. Certainly after Europe and possibly after USA dollar

The West Spends an unsustainable much on Interest (Except for Scandinavia and Switzerland)

Europe

France now spends about 8–9% of its government budget on interest.
Italy and Greece are higher again.
For several euro countries, the interest bill is already close to the limits of what is politically tolerable

United States

The US spends roughly 13% of federal spending on interest payments.
That is a bigger share of the budget than most major European economies.

China

Officially, about 18–20% of China’s government spending goes on interest.

But that number sits on top of three key features:

• state-owned banks are pushed to lend to government at low rates

• state enterprises recycle profits back into state finances

• capital controls keep domestic savings trapped inside China

If you strip those supports out and apply more market-like funding costs, China’s “true” interest load would be well above 20%. The point is not the exact number, but that the system is designed to carry it.

Europe: First to Crack

Europe’s weakness is political, not arithmetic.

Debt stress has moved into the core: France now combines high deficits, rising interest costs and repeated government breakdowns. Germany will not write blank cheques for the south. France cannot impose deep adjustment. The eurozone cannot agree on a shared solution.

When that happens, markets take over. The euro weakens, borrowing costs rise, and austerity is forced rather than chosen.

Europe is therefore the first major bloc likely to be pushed into crisis by high debt plus limited political capacity to act.

The US: Lasting Longer, Still Sliding

The US interest bill is already heavy, but the dollar buys time.

Because the dollar is the world’s reserve currency, global demand for US assets lets Washington roll its debt on a scale nobody else can match. That delays the reckoning, but it does not remove it. Debt service is growing faster than tax receipts, so the long-term direction is still negative.

The US almost certainly outlives Europe in a debt squeeze, but it is still drifting toward harder choices.

China: Built To Endure

On paper, China’s numbers look worse. In practice, its institutional design gives it more staying power.

• State banks absorb government debt at policy-set yields, not market yields.

• Capital controls stop large outflows into foreign assets.

• State enterprises act as a hidden tax base, with profits indirectly supporting the budget.

That mix lets Beijing tolerate an interest burden that would trigger a funding crisis in an open democracy. China is not healthier; it is more insulated from market pressure.

Who Survives the Debt Endgame?

If high interest costs remain a structural feature, the likely order is:

• Europe fails first, blocked by fragmented politics and a rigid currency system.

• The US lasts longer, supported by the dollar but dragged down by compounding interest.

• China lasts the longest, using a controlled financial system to suppress the forces that topple open systems.

On current trends, China – not the US and certainly not Europe – is positioned to be the last major economy standing.

This Post Is Formatted for Landscape

scroll right on mobile >>

My take is that they are not and despite bleating by the worried commentators, the AI value march continues to pick up pace.

The PC era eliminated typing pools, filing rooms, and manual clerical labor. Tasks like these moved from human workflows into software. That shift did not reduce total spend on labor. Instead, corporate IT spend increased because coordination complexity increased when work scaled. Software became a fixed operating requirement, not an optional enhancement.

The winners were platform vendors:

• Microsoft controlled the work surface (Windows + Office).
• Intel captured compute demand
• Cisco captured networking demand .

The AI growth rhymes with this.

The key dynamic was that labor-cost reduction cycled back into software and compute spend, and the software layer that mediated daily work became the structural control point for enterprise productivity.

The same pattern applies to AI.

Organisations like NHS have deployed large initial use cases for Palantir – Its easy to see this being replicated globally

AI does not only replace repetitive tasks. It replaces coordination—the planning, routing, optimizing, deciding, and communicating that sits between organizational nodes.

Coordination shifts from humans to software, the operational execution layer becomes the new control surface.

Labor spend is not removed—it is converted into:

• Compute capacity
• Model inference
• Workflow orchestration platforms
• Integrated operational decision systems

If one platform becomes the default place where operational decisions are executed, it occupies the same structural position Microsoft held in the PC revolution: the layer through which work occurs. There will be a place for a Microsoft of the AI world

That platform becomes the operating system for the organization. the spend base is no longer defined by the size of the IT budget. It is defined by the scale of operational labor being replaced. (Just like Windows was once)

Western workforce ≈ 440M
Assume 7.5% roles eliminated → ~33M jobs
Average fully-loaded cost per role → ~$100k/yr
Annual labor cost removed → ~$3.3T/yr

Historically, automation cycles reallocate 40–60% of displaced labor cost into the enabling systems.
Use the midpoint: 50%.

New capital flowing into AI systems:
~$1.5T per year.

Allocation across the AI stack:

• Compute/datacenters/GPUs: ~45–55% → ~$700B/yr
• Model/API inference: ~10–15% → ~$150–$250B/yr
• Operational AI platforms: ~20–30% → ~$300–$450B/yr
• Integration + workflow transition: ~10–15% → ~$150–$250B/yr

This puts the operational AI platform market (the layer where decisions and workflows run) at:

~$300B–$450B per year steady-state.

This is the layer Palantir is architected to occupy.

It is the only part of the stack that controls execution, and execution is where persistent value accrues.

Valuation Implication

If Palantir becomes a mid-tier platform (5–10% share):

• Revenue: ~$15B–$40B/yr
• At 12×–20× sales: ~$200B–$800B valuation

If Palantir becomes a dominant operational layer (15–25% share):

• Revenue: ~$45B–$110B/yr
• At 12×–20× sales: ~$550B–$2.2T valuation

If Palantir becomes the default operating system for organizational execution (the Microsoft-on-operations outcome):

• Market share approaches 30%+
• Revenue exceeds ~$120B/yr
• Valuation aligns with Microsoft-scale: ~$1.5T–$3T depending on margin structure

Conclusion
The relevant comparison is not against current SaaS markets.
It is against the operational labor base being automated.
If AI shifts coordination and decision-making into software, the platform that becomes the default execution layer captures structural, recurring, control-layer economics—the same dynamic that made Microsoft the dominant winner of the PC era.

Under that model, Palantir’s valuation path ranges from $1T on modest penetration to $2T+ if it becomes the operational OS.

Why the next phase of AI won’t be free — and how it will evolve into a paid global $500bn service industry.

The era of free, unlimited access to advanced AI tools is ending. Systems like ChatGPT, Claude, and Gemini are shifting from experimental novelties to essential infrastructure — much like broadband and mobile networks once did. And nobody expects mobile data to be free.

Each major technology wave starts open, then matures into a paid utility once it becomes indispensable. AI is now reaching that stage. The underlying economics no longer support free use: billions are being spent on GPUs, data centers, and electricity, and every AI interaction consumes measurable compute power. Advertising can’t absorb that cost indefinitely.

AI will become a metered service industry — paid for in the same way as connectivity:

• Free or limited tiers for occasional and educational users
• Subscription or bundled access for individuals and businesses
• Enterprise contracts for high-volume or automated workloads

Over time, prices will stabilize and commoditize, just as mobile and broadband eventually did. Today’s premium subscriptions may one day resemble phone plans — ranging from low-cost basic access to all-inclusive “unlimited intelligence” bundles. But that stage lies years ahead; first, AI must stand as a self-funding global service industry.

If we look at the likely revenue scale, three broad user groups emerge: a free tier that stays minimal or ad-supported, a large consumer tier paying modest subscription rates, and a smaller professional tier willing to pay premium prices for high performance and integration. Even if only half of global internet users adopt AI services under this mix — say, an average of $10–15 per month — total consumer revenue could reach $400–600 billion a year. When added to enterprise and government demand, this level of spending is enough to sustain the data-center and GPU investments already underway.

Bottom up measure of the revenue available to support today’s datacentre build out.

The likely revenue from AI end users to the GPU/DataCenter is intact and realistic.

AI infrastructure economics work from the customer upward. End users pay for AI-enabled products, and that revenue flows to applications, middleware, hyperscalers, data centers, and GPU suppliers. Only end-user spending generates new revenue; the rest simply passes it upstream.

How much cost to cover

Today’s global AI build-out totals approximately $350–400 billion invested, or about 5–7 million GPUs. Annual operating and capital recovery costs are roughly $150–200 billion per year, which must be supported by downstream revenues.

Where does that $$ come from up the Stack

(Best viewed landscape)

Sector	Current Annual Revenue (USD B)	AI Revenue Growth (%)	Incremental Revenue (USD B)	Total Post-AI Revenue (USD B)
Retail & Advertising	~700	+35–65	250–450	950–1,150
Enterprise & Government Productivity	~9,000 (OPEX base)	+2–3 effective capture	180–280	180–280 (net benefit)
Defense & National AI Programs	~800	+8–12	60–120	60–120
Financial Services	~2,000	+2–3	40–60	40–60
Healthcare & Life Sciences	~9,000	+0.5–1	45–90	45–90
Industrial & Logistics	~4,000	+1–2	40–80	40–80
Media & Entertainment	~2,500	+2–3	50–75	50–75
Education & Training	~1,000	+2–4	20–40	20–40
Total Downstream Revenue Creation	—	—	≈ 685–1,195	≈ 685–1,195

Downstream revenues provide roughly three to six times coverage of the infrastructure’s annual cost ($150–200 billion per year).

Timing

2025–26: Utilization 50–65 percent, with strongest pull-through from retail and advertising.
2026–27: Enterprise, government, and defense spending lift utilization toward 80 percent.
2027–28: Equilibrium reached; downstream revenues of roughly $0.8–1.2 trillion per year meet or exceed the $150–200 billion annual infrastructure cost.

Conclusion

The AI data-center build-out is financially supportable. It is ahead of revenue, not in excess of it. As retail, enterprise, and government adoption matures, the system becomes self-funding within about two years.

Q&A

Q: Does this account for the current build-out pace?
A: Yes. It assumes GPU capacity roughly doubles by 2026, with capex rising 25–35 percent per year and AI-driven revenues expanding 30–40 percent per year. On that trajectory, revenue equilibrium is expected by 2027–2028. If build-outs outpace revenue growth, equilibrium could slip by about one year.

The AI Market HUGE – Bottom Up Analysis

Produced by Priory House Partners — PrioryHouse.com

The Insatiable Demand for AI Tools

This report examines the explosive and unrelenting growth in demand for AI-driven tools — a market already exceeding $ 350 bn in 2025 and projected to expand toward $ 700–800 bn by 2028. It explores how AI adoption across software, infrastructure, and operations is transforming business models, creating both new revenue layers and deep structural cost efficiencies. The following sections break down these effects by industry, monetization model, and their cumulative impact on GPU infrastructure demand.

Introduction & Market Scope

AI monetization is expanding faster than expected across industries. 2025 AI tools revenue of $ 350 bn is projected to reach $ 700–800 bn by 2028, with ~40% captured by GPU infrastructure (≈ $ 280–320 bn).

AI Revenue Split by Industry (2028)

Industry Totals (Incremental AI Tools Revenue, $B)

Industry	Incremental AI ($B)
Enterprise / SaaS	73
Transport / Energy / Infrastructure	65
Government / Military	53
Ecommerce / Retail	49
Finance / Insurance	32
Healthcare	29
Education	16

Monetization Models Across Segments

Software / SaaS Sector – Add-On Module Model

Vendors integrate AI into existing platforms as modular upgrades (e.g., copilots, anomaly detection, predictive reporting). Fast to deploy, scalable, and capital-light.

• AI tier or “Pro” subscription plan
• Usage-based charges (API calls, inference tokens)
• Feature-based upsells inside enterprise suites

Example: productivity SaaS adding summarization, design, or code-generation modules with 15–25% premium pricing.

Industrial / Infrastructure – AI as a Cost Option

Sold as efficiency enhancements integrated into existing control/data systems to improve reliability, energy use, and logistics.

• Predictive maintenance for fleets and energy grids
• Route and load optimization in logistics
• Smart forecasting in utilities or construction

Typical pricing: recurring license uplift or analytics subscription (~5–10%).

Cost-Saving Models – Human Replacement at Lower Cost

AI substitutes human analysis/processing at lower cost; monetization via throughput or per-transaction pricing.

• Insurance / Health claims: automated review, adjudication, fraud detection
• Credit scoring: instant AI risk modeling replacing manual underwriting
• Medical scanning & test results: automated radiology/lab interpretation
• Drug discovery: generative compound screening accelerating R&D
• Customer service: conversational agents replacing Tier-1 call centers

These substitutions expand compute demand and model training cycles, sustaining GPU utilization even in cost-cutting environments.

E-commerce – AI Uplift on Take Rates ($ 49 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Retail media & sponsored listings	240	8.5	20
Dynamic pricing & recommendation	120	7.5	9
Fulfilment & logistics	160	6.0	10
Customer support & personalisation	80	7.5	6
Fraud & trust management	45	10.0	4

Enterprise SaaS – Productivity & Analytics ($ 73 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Productivity & collaboration	230	7.5	17
Engineering / design software	170	7.5	13
Data / analytics / BI	230	7.5	17
CX / CRM / marketing automation	260	7.5	20
ERP / HR / ITSM	90	6.5	6

Conclusion

The market for AI is immense—far larger than broadly appreciated—because AI is becoming pervasive across tools, systems, and services. Each incremental capability compounds efficiency, decision quality, and revenue capture, driving durable GPU infrastructure demand.

Appendix – Other Industries

Transport / Energy / Infrastructure ($ 65 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Connected vehicle systems	90	8.5	8
Fleet & logistics	120	8.5	10
Aviation ops	140	7.0	10
Energy exploration & production	130	7.5	10
Grid management & forecasting	150	8.5	13
Mining & materials	60	8.5	5
Agriculture & food	100	9.0	9

Government / Military ($ 53 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Defense: ISR, simulation, autonomy	220	11.5	25
Tax & revenue services	70	10.5	7
Citizen services	55	9.0	5
Education IT & workforce	80	9.0	7
Public research & environment	95	9.0	9

Healthcare ($ 29 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Diagnostics / imaging	120	7.5	9
Clinical data mgmt	100	8.0	8
Drug discovery / R&D	80	9.0	7
Hospital operations	70	7.5	5

Finance / Insurance ($ 32 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Risk modelling & trading	130	8.0	10
Claims & underwriting	110	8.0	9
Fraud detection	90	8.0	7
Customer advisory / CX	80	7.0	6

Education ($ 16 bn)

On mobile, scroll left/right to view all columns.

Segment	Base Market ($B)	AI Adoption (%)	Incremental AI Rev ($B)
Digital learning / tutoring	80	9.0	7
Institutional admin	45	8.0	4
Research / analytics	60	9.0	5

Conclusion First: LLM Companies Will Own Inference—And With It, the Rights to the Silicon $$

OpenAI’s vast monopoly has turned the tables on Nvidia.

Processing does not own LLM’s – LLM’s own Silicon!

The most important investor takeaway is this: Large Language Model (LLM) companies will not just dominate inference—they will own the rights to the silicon that powers it. In the same way Intel and ARM controlled instruction sets for decades, model owners like OpenAI, Anthropic, and Google now control the “instruction set” of AI: the closed model weights. No chip can run their models without their approval. This creates a monopoly dynamic with far greater lock-in than past hardware eras.

NVIDIA’s Blackwell GPUs are world-class for training, but they are the wrong long-term solution for inference. Inference accounts for 99% of AI compute today, and it can be executed more cheaply and efficiently on custom chips tied directly to model architectures. Blackwell is overbuilt for this role, leaving an opening that the model owners themselves are best positioned to fill.

---

Blackwell vs. Inference Chips: Efficiency at the Core of AI Deployment

NVIDIA’s Blackwell architecture (B200/B100) is unmatched for training and flexible enough for both training and inference. But in inference-only use cases, it wastes silicon and power on unused capabilities. In contrast, dedicated inference chips—like Tesla’s FSD chip, Apple’s Neural Engine, Google’s TPUv4i, or AWS’s Inferentia—are optimized for low-bit precision, simpler interconnect, and streamlined memory pipelines. The result is 3–10× higher performance per watt and significantly lower cost per inference.

This efficiency gap demonstrates why inference will migrate away from general-purpose GPUs toward custom silicon.

---

Inference Monopolies: How Model Owners Will Control the Future of AI Compute

Why Inference Will Dominate

• Training is rare and centralized.
• Inference is constant and scales with usage.
• Economically, inference drives the overwhelming majority of AI market value.

Cost-Effective Silicon for Inference

• Inference chips can be built on 5nm or 7nm, avoiding the costs of bleeding-edge nodes.
• Apple, Tesla, and Samsung have proven custom inference silicon can be built for $150M–$300M.
• Quantization and model compression make inference even more efficient at lower geometries.

The Strategic Shift: Models as the New Instruction Set

• Historically, Intel and ARM monopolized by controlling instruction sets.
• In AI, model weights are the instruction set—closed, copyrighted, and protected by license agreements.
• This means the model owner alone controls the right to design inference hardware compatible with their models.
• Competitors cannot copy or replicate without permission, creating a legal and technical lock-out.

Deployment Path

1. Centralized inference: Run in lab-controlled data centers (today’s norm).
2. Enterprise inference: Licensed chips deployed inside corporate data centers.
3. Edge inference: Personal Intelligence Engines (PIEs) embedded in user devices.

Investor Implications

• Model owners (OpenAI, Anthropic, Google): Gain not only software monopoly but also a hardware monopoly, by controlling both the model and the silicon rights.
• Chip vendors (NVIDIA, AMD): Training remains important but is a smaller, less scalable market. Their dominance weakens as inference shifts to model-specific chips.
• Enterprises: Must license inference hardware, paying rent to model owners.
• Consumers: AI engines run locally but still locked to the model maker’s ecosystem.

---

Final Takeaway for Investors
Training may still generate headlines, but inference is where the money and the control lie. And inference belongs to the model makers, who own the weights and thus own the rights to the silicon. This is a stronger monopoly than Intel or ARM ever achieved—legal, architectural, and economic. For investors, the key is clear: back the companies that own the models, because they will own the hardware market that runs them.