What’s next for Anthropic, Perplexity and Character.ai

For the past year, Anthropic, Perplexity and Character.ai have been discussed as if they represent the future of AI.

They don’t.

They represent a holding pattern.

Usage across large language models has stalled. Revenue models remain unproven. And none of these companies has solved the three problems that determine whether an AI business survives at scale: memory, governance and money.

That doesn’t mean these companies are failing. It means the window for improvisation is closing.

The next phase of AI will not be won by better demos, clever prompts or incremental model improvements. It will be won by systems that can remember, systems that users can control and systems that can be billed in a way that maps to real cost.

So the question is no longer “who has the best model?”

It’s this: which of these companies can make the transition before the market forces it on them?

This is not a question about technology. It is a question about survival.

A distinct class is forming

Anthropic, Perplexity and Character.ai share a striking set of characteristics that increasingly place them in the same risk category:

• Privately held, with valuations set by narrative rather than public-market price discovery
• Orders of magnitude smaller than hyperscalers that control infrastructure, distribution and capital
• Near-total exposure to AI — no diversified revenue base to absorb a slowdown
• No durable distribution advantage (no OS, no browser, no default search, no social graph at scale)
• Incremental product differentiation, not structural economic advantage
• Growth that has flattened relative to expectations baked into prior valuations

This is a very different position from Meta’s LLaMA or xAI’s Grok, which are embedded inside large, publicly traded social platforms with independent cash flows and built-in distribution. Those companies can afford AI to be inefficient for longer. This class cannot.

Why the timing suddenly matters

For months, these companies benefited from a market conversation focused on capability: benchmarks, model quality, safety posture and brand perception. That conversation is now being displaced by one focused on economics.

Oracle’s capex pullback, followed by broader market weakness in AI-adjacent stocks, has shifted attention from what models can do to what they cost — and whether pricing models actually map to compute.

That shift is asymmetric. Hyperscalers can absorb it. This class cannot.

The investor problem

Anthropic’s primary backers include Google and Amazon, alongside institutional investors. Perplexity has raised capital from a mix of venture firms and strategic partners. Character.ai’s largest backer is Andreessen Horowitz.

These are sophisticated investors. They understand that in a repricing environment, the most dangerous position is not “unprofitable,” but structurally exposed with no lever to pull.

The uncomfortable questions now being asked are not about survival in an absolute sense, but about optionality:

• What happens if public-market comps continue to reprice AI risk?
• What is the path to margin control if token-based pricing remains misaligned with compute cost?
• Who owns distribution if the hyperscalers decide to compete more aggressively?
• Is independence still the optimal outcome, or merely the default one?

Why weekend meetings are plausible

When market narratives shift quickly, boards and major investors do not wait for quarterly cycles. They convene. Quietly. Especially on weekends.

That does not imply panic. It implies triage.

The menu of options is familiar:

• strategic partnerships that trade independence for distribution or cost relief
• deeper alignment with hyperscalers that already control infrastructure
• reframing as acquisition candidates rather than category leaders
• or, in some cases, simply waiting to see whether the market’s new skepticism deepens

What has changed is not the existence of these options, but the urgency of evaluating them.

The larger implication

This moment is not about Anthropic, Perplexity or Character.ai individually. It is about what happens when AI moves from a capability-first narrative to a revenue- and governance- and adoption-first one.

In that world, scale, distribution, pricing models and economic discipline matter more than marginal model quality. Companies that lack those attributes are not doomed — but they are exposed.

And exposure, once visible, has a way of concentrating attention very quickly.

The market has begun asking harder questions.
This class of companies now has to decide how it wants to answer them.

SWOC

Strengths

• Organizational focus and nimbleness: Unlike hyperscalers, these companies are not burdened by legacy product lines, internal politics or conflicting incentives. They retain the ability to pivot pricing, feature sets, architecture and — critically — narrative.
• Brand credibility disproportionate to size: They are widely perceived as serious AI players: thoughtful, high-quality and comparatively trustworthy. That perception still carries weight with enterprises, regulators and sophisticated users.
• Strategic proximity without embedded inertia: They already operate inside the gravitational field of major infrastructure providers without being fully subsumed by their constraints, timelines or risk tolerance.
• High signal-to-noise products: Their offerings are narrowly defined and easy to understand. This makes it possible to carve out defensible niches through customization and audience alignment (education, under-18 users, mental health, professional domains) that are too small or too specific for hyperscalers to prioritize.

Weaknesses

• Structural economic exposure: Near-total dependence on AI workloads with pricing models misaligned to compute cost and margin control. This is existential: there is no diversified revenue base to absorb error.
• No owned distribution: No operating system, browser, default search position or scaled social graph. Growth depends entirely on product pull rather than embedded reach.
• Private-market valuation lag:
  Narrative-driven private pricing delays necessary recalibration and increases repricing shock. That said, narrative control remains one of the few levers still available.
• Minimal leverage over inputs
  Infrastructure, chips, cloud pricing and traffic sources are controlled by counterparties that can outwait them. Cost mitigation is only possible through aggressive optimization — an area the industry has largely avoided.

Opportunities

• Asymmetric differentiation: Governance, memory, pricing models, specialization, optimization and user agency offer paths that hyperscalers cannot pursue quickly without destabilizing existing businesses.

Current conditions

• Direct competition with hyperscalers: This is not a future risk. It is the current operating environment. Absent differentiation, these companies are already competing head-to-head with firms that possess orders of magnitude more capital, distribution and tolerance for inefficiency.
• Narrative exhaustion: “AI magic” is now table stakes. Without structural advantages, offerings collapse into interchangeability. This is already happening.

Strategic jujitsu

With asymmetric resources, survival depends on asymmetric strategy. Direct competition favors incumbents by definition.

1. Persistence without deviation is failure:
   Continuing on the current path is not prudence; it is inertia. Large incumbents struggle to pivot quickly even under existential pressure. Smaller organizations do not have that luxury — or that excuse.
2. Narrative control is not optional:
   If the prevailing story is unfavorable, changing the narrative is the first move, not the last. A resonant end-user-focused story — distinct from model-focused capability or benchmarks — is itself a differentiator.
3. Game-changing levers exist — but require immediate action. Acting first changes the narrative and differentiates immediately:

• Memory: Move to loss-less architectures that eliminate hallucinations, rebuild trust and materially improve adoption. No major platform is publicly claiming this today.
• User-level governance: Give endusers agency over their own experience — controls, constraints and outcomes. This is absent across mainstream offerings.
• Pricing via optimization, not abstraction: Reduce cost structurally and pass that advantage through pricing. Undercutting competitors via efficiency, not subsidies, changes the economics of adoption.
• Deliberate specialization
  Serve audiences too specific, regulated or nuanced for hyperscalers to tailor for. Focus creates leverage where scale does not.

Each of these paths is available today. None require new technology. All exploit asymmetries that still favor smaller actors. No one else appears to be doing any of these things, which means competitive advantages remain within runway length — for now.