Preparing For the AI/Agentic Wars
Preparing For the AI/Agentic Wars
Why the lessons of Microeconomics 101 and the Internet Search Wars may determine who wins the next great technology battle - and where investors should place their bets.
In this week’s newsletter, we explore an increasingly important question for investors focused on the coming Autonomous Digital Economy and opportunities to invest in the future leaders.
Who will win the coming AI and Agentic Wars, and why?
Much of the current debate focuses on which company has the most intelligent large language model (LLM), the best researchers, or the most compelling AI applications. These are important questions, but we believe they overlook another potentially decisive competitive advantage.
The ability to build, power, finance, and operate the physical infrastructure required to meet exploding global demand for artificial intelligence.
We have seen this movie before.
During the early Internet Search Wars, several companies developed compelling technologies and competed to become the world’s preferred search engine. Yet software alone did not determine the eventual winner. Google’s ability to deploy enormous computing capacity, combine superior technology with an attractive business model, and secure powerful distribution partnerships helped it establish a dominant market position.
We believe something similar may be unfolding today, but on a substantially larger scale.
The implications for investors are profound.
What we learned in Stanford GSB Microeconomics 101
One of the first principles taught in economics is that the interaction of supply and demand determines prices:
When demand increases faster than supply, prices tend to rise.
When supply expands faster than demand, prices tend to fall.
Simple enough. But what happens when demand is growing exponentially, and supply requires enormous investments in expensive physical infrastructure that can take years to build?
Two things can happen:
First, companies that can increase supply most rapidly are positioned to capture disproportionate market share. They can serve customers that their competitors cannot, build relationships, establish distribution advantages, and potentially benefit from economies of scale.
Second, if aggregate supply cannot keep pace with demand, prices can rise. This creates the potential for unusually attractive economics for companies that control scarce capacity, provided competition and operating costs do not absorb those benefits.
Importantly, these outcomes are not mutually exclusive. The companies building capacity fastest can potentially enjoy both rising market share and stronger pricing. Over time, the advantages may become self-reinforcing:
More capacity enables more customers.
More customers generate more revenues and operating data.
More revenues support further investment.
Greater scale can lower unit costs, enabling the market leader to reinvest even more aggressively.
If one company can consistently expand supply faster than its competitors, it may establish a commanding industry position. Of course, physical infrastructure alone does not guarantee success. The company must also have competitive technology, an attractive business model, and effective distribution.
But when infrastructure is the binding constraint, the ability to build it rapidly can become the decisive competitive advantage.
We believe this is one of the most important economic principles investors should apply to the emerging AI and Agentic Wars.
We only have to go back to the Internet Search Wars to see why.
What we learned from the Internet Search Wars
Those of us who lived through the early years of the Internet remember that Google was not always synonymous with search.
During the 1990s, numerous companies competed to help users navigate the rapidly expanding World Wide Web. AltaVista, Yahoo!, Lycos, Excite, Infoseek, and others competed for users, advertisers, and distribution. Back then, Alison and Matthew were investors in @Home, AOL, Excite, and Inktomi, and we remember those days and what played out only too painfully:
The market was expanding extraordinarily rapidly. More people were coming online, more websites were being created, and the volume of searches was growing dramatically.
The technology was improving, but the economics were challenging.
Search required servers, storage, networking equipment, bandwidth, and substantial engineering resources.
Companies needed to crawl and index an ever-expanding Internet while responding quickly and reliably to growing user demand.
Many early search providers faced significant infrastructure and capital constraints.
Meanwhile, the industry was experimenting with competing business models.
Some services were free, supported by advertising. Others experimented with subscriptions, paid services, licensing, or portal-based commercial models.
There was no universal agreement on how companies would ultimately monetize Internet search.
Then Google emerged. Google combined three particularly important advantages:
First, superior search technology. Larry Page and Sergey Brin developed BackRub, the research project that evolved into Google. Its PageRank approach used the link structure of the Internet to help determine the relative importance of web pages. This produced highly relevant results and differentiated Google from many early competitors.
Second, an extraordinary commitment to physical infrastructure. Google developed a distinctive approach to building large-scale computing infrastructure. Rather than relying exclusively on expensive proprietary enterprise systems, it assembled enormous clusters of relatively inexpensive commodity servers, supported by proprietary software and increasingly sophisticated distributed computing architecture. Over time, Google built massive data centers containing millions of servers. Its ability to deploy computing capacity efficiently and at enormous scale became a core competitive advantage. Google could index more of the Internet, process growing numbers of searches, and deliver consistently fast responses. This infrastructure advantage was difficult for smaller competitors to replicate.
Third, an exceptionally powerful business model. Google made search free for users and monetized commercial intent through advertising. AdWords, launched in 2000, evolved into an increasingly sophisticated auction-based advertising marketplace. In the early years, many search advertising keywords were relatively inexpensive. As more advertisers recognized the commercial value of reaching customers precisely when they were expressing purchase intent, competition for valuable keywords intensified. Prices for commercially attractive keywords increased, although pricing varied substantially across categories. This created a powerful economic engine. Consumers received free search. Advertisers paid for access to high-intent customers. Google used the resulting revenues to improve its technology and expand its infrastructure.
Distribution completed the picture. Google also understood the importance of partnerships. Agreements with companies such as AOL, browser providers, and eventually Apple helped Google secure valuable distribution and become the default search experience for enormous numbers of users. These relationships reinforced its technological and infrastructure advantages.
The result was a virtuous cycle. Better search attracted more users. More users attracted advertisers. Advertising revenues financed greater computing capacity, further product improvements, and distribution agreements. Google ultimately emerged as the dominant global search provider, although outside markets like China.
It did not win simply because it had the best algorithm.
Google won because it combined strong technology, enormous infrastructure investment, an effective business model, and powerful distribution into a self-reinforcing competitive system.
We believe this historical precedent is highly relevant to the AI and Agentic Wars that are now beginning.
Past is prologue: The coming AI and Agentic Wars
Today, we are witnessing another extraordinary technology competition.
Anthropic, Google’s Gemini, Meta, OpenAI, SpaceX’s xAI/Grok, and several leading Chinese AI developers are competing to develop increasingly capable foundation models.
Their technologies are advancing rapidly, and their relative competitive positions can change from one generation of models to the next. Yet the most consequential development may not be the improvement in model intelligence alone.
It may be the extraordinary expansion in demand for intelligence itself:
From human AI usage to Autonomous Agentic demand
Today, much of the world’s interaction with artificial intelligence remains human-initiated. A person asks a question, requests an analysis, generates an image, writes software, or asks an AI system to complete a task. This already represents enormous and rapidly growing demand for computing capacity. But we believe human AI usage is only the beginning. As we have discussed in previous Fifth Era newsletters, we expect the emergence of an Autonomous Digital Economy in which intelligent agents increasingly act on behalf of individuals, businesses, and other agents. These agents will search for information, analyze alternatives, negotiate agreements, write software, manage workflows, execute transactions, and coordinate with other agents. Unlike human users, autonomous agents will not necessarily sleep, take holidays, or limit themselves to a handful of queries each day.
An agent may initiate hundreds or thousands of computational tasks in the course of completing a single objective.
And agents will increasingly communicate with other agents, potentially creating additional layers of machine-generated demand.
A single human request could trigger extensive chains of planning, research, reasoning, verification, and execution.
The relevant market may therefore evolve from billions of humans using AI to billions, and potentially trillions, of AI-agent tasks being executed across the global economy.
The physical infrastructure constraint
Unfortunately, intelligence is not produced without physical inputs.
AI requires specialized semiconductors, advanced networking, high-performance memory, data centers, cooling systems, electrical infrastructure, and reliable power.
Expanding capacity requires solving multiple difficult problems simultaneously.
Advanced chips must be designed, fabricated, packaged, and delivered. Data centers must be financed, permitted, constructed, and equipped. Electricity must be generated, transmitted, and distributed. Energy storage and backup systems must be installed. Suitable sites must be secured, often in competition with other industrial users.
In many jurisdictions, obtaining an electrical grid connection can take years.
Even where capital is available, equipment, construction labor, transmission infrastructure, and regulatory approvals may remain bottlenecks.
And as of this week, we may be about to see the arrival on the scene of an anti-innovation movement led by the same actors who tried to kill digital finance during the Biden administration.
These constraints cannot necessarily be overcome simply by writing a larger check. We therefore believe there is a meaningful possibility that global demand for AI inference and Agentic computing will grow faster than usable supply over the next several years.
Competitive advantage from building more and faster
Just as with Google during the first innings of the Internet Search Engine wars, it appears that competitive advantage may go to the LLM best at solving this demand/supply mismatch. The LLM best able to rapidly scale up compute and physical infrastructure may, in practice, be the only player users can turn to satisfy their demand.
Conversely, if supply remains heavily constrained, don’t expect free to last forever.
Could AI prices rise?
One of the striking features of the AI market today is how inexpensive many services appear relative to the value they provide. Consumers can access remarkably capable models for free or through relatively modest monthly subscriptions.
Developers can access powerful models through APIs at prices that have generally declined substantially as technology has improved. But this does not mean prices must continue falling indefinitely. If demand expands faster than effective capacity, providers may gain greater pricing power, particularly for premium reasoning, low-latency inference, reliable agent execution, and guaranteed enterprise capacity.
We could see a bifurcated market:
Basic intelligence may become increasingly commoditized and inexpensive as models and hardware improve.
At the same time, access to the most capable models, scarce compute resources, and high-priority Agentic execution may command significant premiums.
The relevant economic distinction is between the price of an individual unit of computation and the total cost of completing a valuable task.
Even if the cost per token falls, an autonomous agent completing a complex assignment may consume far more tokens and computational resources than a conventional search query.
The economic value of the completed task may also be much greater.
We believe investors should be prepared for a world in which the unit cost of intelligence continues to decline while total spending on intelligence increases dramatically - and scarce premium capacity periodically becomes more expensive.
That brings us to the most interesting strategic question: Which companies are best positioned to supply that intelligence?
What are the leading LLM companies doing about this?
We believe the major AI competitors are beginning to pursue materially different strategies for meeting the infrastructure challenge.
Three approaches are particularly instructive.
A. SpaceX/xAI: Elon Musk’s industrial infrastructure strategy
Elon Musk has demonstrated an unusual ability to organize capital, engineering talent, and industrial resources around extraordinarily ambitious physical infrastructure projects:
Tesla built manufacturing capabilities for electric vehicles, batteries, and energy storage.
SpaceX developed reusable rockets, satellite manufacturing, and global communications infrastructure.
Meanwhile, xAI pursued an aggressive approach to building the computing capacity needed to train and operate its Grok models.Following the combination of SpaceX and xAI earlier in 2026, Musk has an increasingly integrated platform spanning artificial intelligence, communications, industrial engineering, and infrastructure deployment.
The proposed combination of SpaceX and Tesla could take this strategy considerably further. In October 2026, Musk confirmed that the companies had formed independent committees to evaluate a potential merger. The companies have not agreed to any transaction, and completion remains uncertain. But the strategic possibilities are significant.
Consider what Tesla could contribute to an integrated AI infrastructure business:
Energy generation and storage. Tesla has capabilities in solar energy, battery systems, power electronics, and grid-scale Megapack storage. These technologies could help support the enormous electricity requirements of AI data centers, particularly where grid availability is constrained.
Industrial engineering and construction. Tesla and SpaceX have developed expertise in factory design, automation, manufacturing scale-up, supply-chain management, and rapid engineering iteration.
Electrical systems and power management. Tesla’s experience with batteries, inverters, energy management, and industrial electrical systems could complement the requirements of large-scale AI infrastructure.
Capital and organizational resources. A combination could potentially coordinate investment decisions, engineering resources, and procurement across multiple businesses.
Connectivity and distribution. SpaceX’s Starlink network could provide additional opportunities for distributing AI services and supporting geographically dispersed infrastructure.
One possible strategic direction would be to redirect a meaningful portion of Tesla’s engineering, energy, and industrial capabilities toward accelerating SpaceX’s AI capacity expansion. Instead of primarily serving external customers, a combined company could increasingly deploy some of these resources to meet the internal infrastructure requirements of a rapidly growing AI business.
We are not suggesting that such a redirection has been announced or approved. It is a strategic possibility worth considering.
Musk could potentially assemble a vertically integrated AI infrastructure platform that combines models, computing facilities, energy storage, manufacturing, connectivity, and industrial execution. The potential advantage would not simply be developing a better LLM. It would be the ability to bring enormous quantities of usable computing capacity online faster than competitors.
For investors, the key question is whether infrastructure becomes the decisive bottleneck in the Agentic Wars: Can Musk’s industrial capabilities prove as important as his AI research capabilities?
B. Anthropic: Competing through models, partnerships, and safety
Anthropic has established itself as one of the leading developers of frontier AI models. Its Claude family has developed a strong position in enterprise AI, coding, research, and increasingly sophisticated Agentic workflows. Anthropic’s strategy, however, differs from Musk’s vertically integrated industrial approach. The company relies heavily on partnerships with major cloud infrastructure providers and technology companies, including Amazon and Google, to secure computing capacity. This has important advantages.
Anthropic can access the capital, data centers, semiconductor capabilities, and operating expertise of some of the world’s largest technology companies without independently building every component of the infrastructure stack.
But the approach also creates dependencies.
Anthropic must compete for scarce computing capacity, manage complex commercial relationships, and ensure that its infrastructure partners can expand supply rapidly enough to meet demand.
The company has acknowledged the importance of electrical power availability, grid limitations, and long infrastructure lead times.
At the same time, Anthropic has been one of the most prominent advocates for stronger AI safety measures.
CEO Dario Amodei has called for greater independent evaluation, coordination between frontier developers, and a more measured approach to the development of increasingly powerful systems when safety capabilities are insufficient. These concerns should not be dismissed.
As AI agents become capable of executing increasingly consequential actions, cybersecurity, misuse, reliability, and control become critical issues.
Nevertheless, from a competitive strategy perspective, calls to slow frontier development create an interesting tension.
If the industry collectively slows capability development, competitive dynamics may change. If some companies slow while others continue accelerating, market share could shift.
Importantly, slowing the development of new model capabilities is not the same as slowing the deployment of existing models or the construction of inference infrastructure.
Anthropic is also continuing to advance its technology rapidly.
We do not interpret its safety position as evidence that it cannot scale, or that safety concerns are merely a response to infrastructure constraints. However, Anthropic illustrates the challenge of simultaneously pursuing frontier capabilities, infrastructure expansion, commercial growth, and responsible AI development. A slowdown would perhaps benefit Anthropic more than its competitors.
For investors, the key question is whether Anthropic’s infrastructure partnerships and enterprise market position can enable it to scale as quickly as demand requires.
C. OpenAI: Building a global computing ecosystem
OpenAI is pursuing a third approach:
Rather than relying exclusively on existing cloud capacity or attempting to own every part of the physical infrastructure stack, OpenAI has sought to mobilize a broad ecosystem of capital, infrastructure developers, semiconductor suppliers, and technology partners. Its Stargate initiative is central to this strategy.
Originally announced with an ambition to develop 10 gigawatts of AI infrastructure in the United States by 2029, Stargate represents an effort to build computing capacity at a scale previously associated with major industrial and energy infrastructure programs. OpenAI reported in April 2026 that it had already secured commitments exceeding that original capacity target.
There is an important distinction between securing capacity commitments and bringing fully operational infrastructure online. The latter still depends on construction, equipment delivery, power availability, and commissioning. Nevertheless, the ambition is unmistakable.
OpenAI is attempting to expand computing capacity ahead of expected demand. Its strategy involves working with infrastructure partners, securing access to advanced chips, expanding data-center capacity, and raising the enormous amounts of capital required to support these investments.
OpenAI also benefits from a broad distribution ecosystem:
ChatGPT has established a substantial global consumer presence, while OpenAI’s enterprise products and developer APIs provide channels through which businesses and software developers can incorporate AI into their operations.
This distribution matters because infrastructure is only valuable if it can be monetized.
OpenAI’s challenge is to convert its user relationships, developer ecosystem, and model capabilities into sustainable revenues that justify extraordinary infrastructure spending.
Recent statements from OpenAI leadership indicate that the company intends to continue expanding data-center capacity despite growing political and community scrutiny of large AI infrastructure projects.
For investors, the key question is whether OpenAI can orchestrate its network of partners quickly and efficiently enough to keep pace with demand.
Three strategies - which one wins?
The contrast is instructive:
Musk is emphasizing industrial integration,
Anthropic is leveraging strategic infrastructure partnerships, and
OpenAI is mobilizing a broad computing and capital ecosystem.
We should also remember Google.
Unlike many of its newer competitors, Google already possesses enormous global data-center infrastructure, custom AI accelerators, energy procurement capabilities, consumer distribution, and enterprise relationships.
It also has Gemini and DeepMind.
In many respects, Google enters the AI and Agentic Wars with the same categories of advantages that helped it win the original Search Wars. Whether those advantages translate into leadership in the next generation of AI remains to be seen.
But investors would be unwise to underestimate them.
Implications for Investors
What does all this mean for investors seeking exposure to the AI and Agentic Economy?
We believe there are several important implications:
Follow the infrastructure bottlenecks. If demand for AI intelligence grows faster than supply, companies that control or relieve critical bottlenecks may capture substantial economic value. This includes businesses developing advanced semiconductors, high-bandwidth memory, networking technology, data-center equipment, power generation, grid infrastructure, energy storage, cooling systems, and related technologies. However, investors should distinguish between companies benefiting from temporary scarcity and those developing sustainable competitive advantages. Scarcity can produce attractive margins, but it can also attract enormous capital investment that eventually creates oversupply.
The fastest builders may win disproportionately. The Search Wars suggest that the ability to deploy computing capacity rapidly can become a strategic advantage. Investors should therefore assess AI companies not simply on their model benchmarks or user growth, but also on their ability to secure and commission usable infrastructure. Relevant measures include power availability, time to deployment, capital efficiency, utilization, operating costs, and the reliability of capacity expansion plans. The distinction between announced gigawatts and operational gigawatts will become increasingly important.
Expect different economics for different types of intelligence. Not all AI inference will have the same economic value. Basic search, summarization, and routine language tasks may become increasingly commoditized. Advanced reasoning, specialized enterprise workflows, real-time decision-making, and reliable autonomous execution may command greater economic value. We expect pricing models to evolve accordingly. Subscription fees and token-based pricing may increasingly coexist with priority access charges, enterprise capacity commitments, and pricing based on tasks completed or outcomes delivered. Investors should focus on where pricing power can be sustained rather than assuming that all AI services will experience the same economics.
Look beyond the foundation models. The leading LLM providers will attract enormous attention and capital. But history suggests that significant value may also accrue to businesses enabling, distributing, and applying the underlying technology. We see opportunities across AI infrastructure, specialized software, agent orchestration, cybersecurity, identity, payments, and industry-specific applications. In particular, as autonomous agents become economically active, they will require trusted mechanisms for identification, authorization, coordination, settlement, and recordkeeping. This reinforces a central Fifth Era investment thesis. The convergence of artificial intelligence, the Internet, and blockchain technologies will help create an Autonomous Digital Economy. The companies that enable agents to transact securely and reliably may become as important as those supplying the underlying intelligence.
Distribution and customer relationships will matter. Google’s history demonstrates that superior technology and infrastructure are not sufficient by themselves. Distribution can determine which products reach customers and which companies capture the resulting economics. In the Agentic Economy, distribution may come through operating systems, enterprise software, developer platforms, devices, communications networks, and established customer relationships. We expect powerful competition for the position of default AI assistant, enterprise agent platform, and trusted transaction intermediary. The companies controlling these relationships may enjoy significant advantages even when the underlying models become more interchangeable.
Diversification is essential. It is far too early to know which foundation model companies will ultimately dominate. Technology leadership can change rapidly. Infrastructure strategies involve substantial execution risk. Capital requirements are extraordinary, and today’s most successful models may not retain their advantages indefinitely. There is also a meaningful risk that excessive infrastructure investment will produce periods of overcapacity and disappointing returns. We therefore favor diversified exposure across the AI value chain, with particular emphasis on specialist venture managers capable of identifying emerging leaders before they become obvious. This includes exposure to enabling infrastructure, foundational technologies, enterprise applications, and the businesses that will support autonomous digital commerce.
That is a lot to think about.
For our part, we stick to our investment strategy:
Fund VIII, our AI Access Fund, provides you with diversified early-stage access through partnerships with leading AI pure-play venture capital firms.
Direct Investments in Mid and Late-Stage. We share selected direct investments with you once emerging category leaders become evident.
Please let us know at ir@fifthera.com if you would like to know more about our investment strategies, including participating before year-end.
Conclusion: The winners may be determined by who can build fastest
So to summarize this week’s newsletter:
The original Internet Search Wars provide a valuable lesson. Google’s success was not simply the result of developing a better search algorithm. It combined technological excellence with enormous infrastructure investment, effective monetization, and powerful distribution.
We believe the AI and Agentic Wars may follow a similar pattern. Demand for intelligence is likely to grow enormously as billions of people and businesses adopt AI, and as autonomous agents increasingly undertake economic activity on their behalf. But supplying that intelligence requires physical infrastructure on a scale that few companies can finance, build, and operate. The ability to secure chips, electricity, data centers, capital, and distribution may therefore prove just as important as the ability to develop the most capable models.
SpaceX, Anthropic, OpenAI, and Google are pursuing different strategies to meet this challenge. Some will execute better than others. And as the Search Wars demonstrated, early technological leadership does not guarantee ultimate market leadership.
For investors, the opportunity is not simply to identify today’s best LLM. It is to identify the companies that will control the critical infrastructure, technologies, customer relationships, and transaction systems of tomorrow’s Autonomous Digital Economy.
We believe the AI and Agentic Wars are only beginning. The winners may be determined as much by who can build and deploy intelligence at scale as by who can create it.
Thank you for reading,
The Fifth Era Partners
ABOUT FIFTH ERA
'Focused on Innovation'
Our firm believes we are entering a period of unprecedented innovation, which we call the Fifth Era, and every industry and business will be dramatically impacted. We focus on investing in these new innovations, especially private companies focused on the convergence of Internet, AI/Agentic and Blockchain technologies to shape an Autonomous Digital Economy. Our investment strategies construct portfolios of hard-to-access early stage funds and make concentrated direct investments into emerging category leaders during their mid and late stage investment phases. Fifth Era's investment strategies are now in their 13th year and to date we have invested in a combined portfolio of 1,500+ companies including 80+ unicorns. In the US we are a SEC registered investment advisor, in the UK a FCA appointed representative and our funds are registered in Switzerland.
Visit us at www.FifthEra.com to learn more.
Note: SEC Registration does not imply a certain level of skill or training.