AI Will "Fundamentally Change the Business Model of the Web"

Despite the U.S. Department of Justice insistence that Alphabet’s Google search business is a monopoly in need of remediation, lots of observers might note that technological disruption is likely going to happen, in any case, rendering all the potential activity moot. 

According to Cloudflare co-founder and CEO Matthew Prince, “AI is going to fundamentally change the business model of the web,” which “for the last fifteen years has been search.”

“One way or another, search drives everything that happens online,” he said. “Today, 75 percent of the queries that get put into Google get answered without you leaving Google, get answered on that page.”

For example, Google's AI Overview has, by one estimate,  led to a 60-percent decrease in organic search traffic for publishers, with possibly $2 billion in foregone ad revenue.

And that obviously disrupts the advertising monetization model underpinning content on the web, which has featured Google search sending traffic to third-party content sites. But AI changes all that by providing a direct answer without sending the user someplace else. 

Since most digital advertising models depend on high volumes of visitors to generate impressions and clicks, less traffic means fewer ad views, lower click-through rates, and ultimately, diminished income. 

“The consequence of that means that original content creators that are creating that content, if they were deriving value through selling subscriptions or putting up ads, or just the ego of knowing that someone is reading your stuff, that’s gone, right?” he adds. 

A decade ago, scrapped pages led to traffic about once for every two pages indexed, he said. Today, the ratio might be more like six to one.

For users on OpenAI the ratio is more like 250 to one, he argued. At Anthropic the ratio is perhaps 6,000 to one. 

“And if content creators can’t derive value from what they’re doing, then they’re not going to create original content,” he added. All of which suggests some new business model has to be created. 

The other point is that the search business model that supposedly drives Google’s monopoly arguably is going to be disrupted, in any case.

Is DoJ Alphabet Antitrust Unnecessary?

The U.S. government's antitrust actions against Alphabet (advertising technology in one case and search monopoly in another) might be a case of “fighting the last war,” as artificial intelligence models arguably are displacing search in any case. Apple argues that search volume fell in April for the first time ever, for example.

Some of us might argue that the monopoly value of traditional search and advertising is already under threat from AI, which could organically erode Google’s dominance without the need for heavy-handed intervention.

In fact, that is precisely the sort of disruption we see often in fast-moving technology markets.

In other words, regulators are pursuing remedies for a monopoly that may soon be undercut by AI-driven market forces. The whole effort is possibly quite unnecessary.

Data Center Capacity Demand to Grow 3.5X Between 2025 and 2030, McKinsey Estimates

Momentary concerns about enterprise or hyperscale data center demand aside, virtually all observers might agree that demand for data center capability is going to grow substantially, to support artificial intelligence and computing workloads  in general. 

Consultants at McKinsey and Co., for example, estimate that between 2025 and 2030 data center capacity demand is going to grow by 3.5 times. 

source: McKinsey and Co. 

That, in turn, will drive an estimated $3 trillion to $8 trillion in additional data center capacity by 2030.

source: McKinsey and Co.

Are Government Home Broadband Networks Facing Worse Business Cases?

Dr. George Ford, Phoenix Center for for Advanced Legal & Economic Public Policy Studies chief economist, notes in a recent study that three sales of municipal home broadband networks illustrates the financial issues such networks face. 

The Bardstown, Ky. network, for example, privatized in 2024, illustrates the revenue side of the problem. 

source: Phoenix Center 

Another study looked at the financial performance of every municipal fiber project (with published financial data) in the U.S. operating in 2010 through 2019. None of the 15 projects generated sufficient nominal cash flow in the short run to maintain solvency without infusions of additional cash from outside sources or debt relief. 

To be sure, 68 operating networks provide no public financial information, some observers note. 

Similarly, 87 percent have not actually generated sufficient nominal cash flow to put them on track to achieve long-run solvency. 

Some 73 percent generated negative nominal cash flow over the past three fiscal years, leaving them poorly positioned to make up their deficits and causing them to fall farther into debt, the authors note. 

Fully 53 percent of projects would not be on track to reach breakeven even assuming the theoretical best-case performance in terms of capital expenditures and debt service.

Business Model Issue	Impact	Sources
Short-Term Revenue Shortfalls	Most projects fail to cover operating costs with subscription fees, requiring taxpayer subsidies.	2,4,10
Long-Term Viability Concerns	Only 2/15 projects studied showed potential for self-sustaining cash flow over 20-25 years.	2,6,10
Network Upgrade Costs	Frequent tech advancements require reinvestment, straining budgets not designed for dynamic needs.	137
Cross-Subsidization Risks	Many rely on municipal utility funds or bonds, distorting competition and transparency.	7,10,11
Crowding Out Private Investment	Municipal entry reduces private sector incentives to build/upgrade networks in the same areas.	3,6,11
Project Management Complexity	Lack of expertise in broadband operations leads to cost overruns and service quality issues.	3,5,9
Political vs. Market Incentives	Prioritizing coverage over profitability results in unsustainable pricing and service models.	3,10,11
Financing Challenges	Securing loans/investment is harder due to incumbent opposition and uncertain ROI.	5,9,11

CoreWeave's Valuation in High-Performance Computing as a Service

Times of high volatility in equity markets tend to be even more volatile where it comes to assets in new or emerging fields where demand is hard to forecast. So it is with high-performance computing “as a service,” the business CoreWeave and others are in. 

So it might not come as a complete surprise that some observers are skeptical about CoreWeave’s valuation, at the very least. 

To evaluate, we would need to estimate the global market; the U.S. share of that market (CoreWeave will operate internationally as well as in the U.S. market); CoreWeave’s  enterprise value (debt plus equity) and then CoreWeave’s estimated revenue. 

So assume global 2023 HPCaaS revenues of about $36 billion; North America share is 40 percent and U.S. share of North America is 85 percent. That gives us a 2023 U.S. market of about $12.4 billion. 

U.S. High-Performance Computing "as a Service" Market
Year	Low Scenario (8% CAGR)	Baseline Scenario (13.4% CAGR)	High Scenario (18% CAGR)
2023	$12.42 billion	$12.42 billion	$12.42 billion
2024	$13.41 billion	$14.08 billion	$14.66 billion
2025	$14.49 billion	$15.97 billion	$17.30 billion
2026	$15.65 billion	$18.11 billion	$20.41 billion
2027	$16.90 billion	$20.54 billion	$24.09 billion
2028	$18.25 billion	$23.29 billion	$28.42 billion
2029	$19.71 billion	$26.41 billion	$33.54 billion
2030	$21.29 billion	$29.95 billion	$39.58 billion

So that establishes some expectations for the market opportunity CoreWeave and others are chasing. 

An immediate question is whether CoreWeave can hope to make a profit, given its debt and equity structure, both of which place claims on some of CoreWeave’s cash flow.

The company took on $7.5 billion in debt facility led by Blackstone and Magnetar in May 2024, with participation from Coatue, Carlyle, CDPQ, DigitalBridge Credit, BlackRock, Eldridge Industries, and Great Elm Capital Corp., following an earlier $2.3 billion debt facility led by Magnetar Capital and Blackstone in August 2023.

A $650 million credit facility was added in October 2024. So total debt is $10.45 billion.

Equity issued includes a $1.1 billion Series C round in May 2024, valuing CoreWeave at $19 billion.

A $642 million Series B round was completed in December 2023 with another $420 million Series B round in April 2023.

According to Crunchbase, CoreWeave raised $2.37 billion in equity across 13 rounds, including seed, Early-Stage, and Late-Stage rounds).

So assume $2.37 billion as the total equity raised. Add the $10.45 billion in debt and total investment  is $12.82 billion. 

Enterprise value includes the firm’s market capitalization plus debt plus cash. If CoreWeave’s market cap is about $18 billion, then enterprise value is about $28.55 billion. 

For the current EV of $28.55 billion and revenues of $1.92 billion, CoreWeave has an implied revenue multiple of 14.87 times. That might be considered conservative for a fast-growing technology firm.

At $1.9 billion in a U.S. market that might have generated $14 billion in annual revenues, CoreWeave might already have as much as 13,5 percent share in the high-performance computing as a service market. 

Can AI Replace Cloud Computing?

I saw an odd headline asking “when will AI replace cloud computing?” It’s a non-sequitur. Sort of like asking when a layer 7 function is replaced by a layer 1 function. In other words, AI exists at a different layer of the software stack.

OSI Model	AI/Cloud Layer
7. Application (user-facing apps)	AI Model Deployment & APIs4
6. Presentation (data formatting/encryption)	Data Preprocessing/Feature Engineering4
5. Session (connection management)	Cloud API Gateway Management2
4. Transport (data integrity)	Distributed Training Orchestration4
3. Network (routing)	Cloud Network Virtualization2
2. Data Link (node-to-node transfer)	GPU Cluster Interconnects2
1. Physical (hardware)	Cloud Infrastructure (GPUs/TPUs)

Saying AI replaces cloud computing is like saying "video games replace electricity." Cloud computing is infrastructure, supplying functions lower in the stack.