The New Front in the A.I. Race

The contest over artificial intelligence is widening

The race to dominate artificial intelligence is no longer confined to headline-grabbing model launches and soaring valuations. It is increasingly being fought on several fronts at once: in cyberespionage campaigns, in the guarded release of more powerful systems, in the jockeying for future stock market debuts and in a renewed push by investors to recast software as an A.I. winner rather than a casualty.

That broader shift came into view this week as CrowdStrike, the cybersecurity company, said China-linked groups were intensifying efforts to steal A.I.-related intellectual property; Anthropic moved to publicly release a model it had only recently kept under tighter control; Perplexity said it remained on track for a 2028 initial public offering; and the private equity giant Thoma Bravo declared that fears of an A.I.-driven collapse in software had been overdone.

Together, the developments suggest that the industry is entering a new phase, one in which the central questions are not only which company has the strongest model, but also who can protect valuable know-how, commercialize advanced systems responsibly, convince public investors of sustainable economics and turn A.I. spending into durable software profits.

Espionage concerns sharpen around A.I.

CrowdStrike said on June 9 that China-based entities were responsible for more than half of state-sponsored cyberattacks on technology companies aimed at obtaining A.I. assets, underscoring how valuable model research, training methods and related intellectual property have become.

The company’s warning adds to a growing view in Washington and across the tech industry that A.I. capabilities now sit at the center of strategic competition between the United States and China. The targets are no longer only government agencies or defense contractors, but also A.I. labs, chip companies and software firms whose data, code and design decisions may offer a shortcut in an intensely expensive technological race.

Beijing rejected the accusation and called instead for U.S.-China cooperation on A.I., a response in keeping with its long-running denials of state-backed cybertheft allegations. Still, the claims reflect a deepening anxiety in the American technology sector that, as frontier A.I. models become more commercially and militarily relevant, they will also become more attractive targets for espionage.

The allegations have not been independently verified beyond CrowdStrike’s reporting, and that remains an important caveat. But even the warning itself is significant: it points to a world in which the most prized A.I. assets may be not only chips and talent, but also the tacit knowledge embedded inside laboratories and software companies.

Anthropic tests a faster path from restricted access to public release

At the same time, Anthropic is signaling that some once-sensitive systems may move into wider use more quickly than before.

The company said it was broadening access to a “Mythos”-class model roughly two months after a more limited release through its Project Glasswing program, which had drawn notice in part because Anthropic said the system possessed advanced cyber capabilities. At the time, the company suggested that peer firms could reach similar levels within six to 12 months, a timeline that highlighted how quickly capabilities were converging across the sector.

Anthropic now says it can release a public version because of new safeguards designed to block or reroute high-risk requests. The move reflects a strategy that is becoming more central to leading A.I. labs: rather than withholding stronger systems indefinitely, they are attempting to deploy them with narrower, use-based controls.

That approach may prove essential in an industry under relentless pressure to ship. But it also carries risk. A recurring question for A.I. companies is whether such guardrails will hold up once models are exposed to millions of users probing for weaknesses, jailbreaks and unintended uses. Another is whether safety-minded release strategies can endure if rivals offer comparable power with fewer restrictions.

Still, the decision is notable for what it says about the state of the market. Companies appear increasingly unwilling to keep advanced models behind closed doors for long, especially when customers are demanding access and competitors are closing in.

The public markets loom in the background

The scramble to build and release stronger systems is also increasingly bound up with a financial endgame: who will be ready to go public, and when.

Perplexity’s chief executive, Aravind Srinivas, said the company still planned to pursue an I.P.O. in 2028 regardless of how public listings by Anthropic or OpenAI are ultimately received. The comment came after Anthropic confidentially filed to go public, reinforcing the idea that the next several years could produce a pipeline of closely watched A.I. listings.

That matters because public investors are likely to impose a different discipline on the sector than private capital has. In venture markets, companies have been able to raise vast sums on the promise of future dominance. In public markets, they will face harder questions about margins, infrastructure spending, customer retention and whether rapid usage growth can translate into dependable profits.

Perplexity’s insistence on a long-term timeline suggests that some A.I. companies are trying to present themselves not as speculative one-cycle winners, but as businesses preparing for a more traditional maturation path. Yet by 2028, much could change: the competitive landscape may be reshaped by open-source models, regulatory scrutiny, cloud costs or consolidation among larger platform companies.

Software investors sense a turn

For established software investors, however, the mood has begun to brighten.

Orlando Bravo, the founder of Thoma Bravo, said that the so-called “SaaSpocalypse” — the fear that A.I. would hollow out software economics by commoditizing products and reducing seat-based revenue — is over. Instead, he argued, A.I. is becoming an “enormous tailwind” for software companies.

That view marks a notable reversal from the anxiety that gripped much of the software sector after generative A.I. exploded into public view. Investors had worried that customers would bypass traditional applications, that chatbot interfaces would weaken incumbent software franchises and that expensive A.I. features would erode margins before they generated meaningful returns.

Now a different thesis is taking hold. If software companies can successfully embed A.I. into products, automate more tasks and build “agentic” tools that do work on behalf of users, they may be able to charge more, deepen customer reliance and revive growth rates that had slowed after the pandemic-era boom.

Private equity firms like Thoma Bravo have a particular interest in that outcome, because they have long been major buyers of enterprise software companies. A sustained A.I.-driven rebound could lift valuations across a sector that had been battered by higher interest rates and doubts about future growth.

But the optimism is not without hazards. Customers are still scrutinizing whether A.I. features actually save time or money, and many companies are only beginning to understand the infrastructure and support costs that come with broad deployment. The sector’s recovery will depend not just on excitement, but on proving return on investment.

Why this moment matters

What ties these threads together is a simple reality: as A.I. becomes more consequential, it is becoming harder to separate the technology story from questions of security, capital markets and the future of the software industry.

The theft of A.I. know-how, if it is occurring at the scale described, would underscore that leading models are now strategic assets. The public release of increasingly capable systems shows how quickly the boundary between experimental research and mass-market product is shrinking. The push toward eventual stock listings suggests the industry is already looking beyond breakthrough demos to financial durability. And renewed confidence in software hints that A.I. may strengthen incumbents as much as it disrupts them.

The next phase of the A.I. race, in other words, may be decided not only by who builds the smartest model, but by who can defend it, ship it, monetize it and convince investors that the economics will last.