The AGI Race Is Leaving You Behind

Today, we’re diving into:

Dear Folks,

This briefing reflects publicly reported developments in AI spending, geopolitics, platform strategy, and security as of July 2026, and is intended for informational purposes to support enterprise technology planning.

AI Spending Is About To Explode. Are You Ready?

That is not a trend line. That is a market rewriting its own rules every quarter, and every enterprise leader is being forced to keep pace or fall behind competitors who move faster.

Total market revenue is expected to climb from roughly 39.3 billion dollars to 64.3 billion dollars in a single year. Few enterprise categories move at that velocity.

The Real Story Isn't The Number. It's Where The Money Is Going.

Generative AI model spending alone is forecast to grow 117 percent this year. Meanwhile, AI platform spending rises a comparatively modest 37 percent.

That gap tells you something important. Buyers are no longer paying for access to AI. They are paying for outcomes, and they expect proof.

Analysts tracking this shift note that spending is moving toward providers who can demonstrate clear value across cost, latency, performance, and reliability, not just raw capability.

Why Your Board Will Ask About "Domain-Specific" Models Sooner Than You Think

Domain-specific and specialised generative AI models are the fastest-growing category in the entire market, projected to surge 210 percent, from 1.6 billion to 4.9 billion dollars.

Foundation model spending is also more than doubling, climbing from 11.4 billion to 23.4 billion dollars. Generic capability is no longer the differentiator you think it is.

The advantage is shifting toward models trained for your industry, your workflows, and your regulatory environment, not toward the broadest general-purpose tool available.

If your roadmap still treats every model as interchangeable, you are already behind the leaders who are quietly locking in domain-specific advantages this year.

Your Budget Is Under More Scrutiny Than Ever Before

Enterprise AI budgets are entering a phase of intense oversight, and three things now decide every renewal conversation:

The Leaders Who Win Won't Be The Ones Who Spend Most

The pattern across every segment is consistent. Growth is rewarding the leaders who can govern AI usage across their organisation, not the ones who simply deploy the most tools.

As more vendors enter the field, the leaders who win will be those who help their organisations decide where AI belongs, and where it does not.

What This Means For You, Right Now

As usage-based pricing becomes harder to forecast, your edge depends on visibility into where AI spend actually goes:

The leaders who treat this shift as strategic, not operational, will be the ones setting the pace for their industry through 2026 and beyond.

China Just Gave Away The AI Playbook. Ready?

The model carries 2.8 trillion parameters, making it the largest open-source system released to date. It is set to become fully accessible to developers worldwide by the end of this month.

On several benchmarks, including coding and complex agent tasks, the new system outperformed some of the most capable proprietary models built by leading American labs.

Why Every CEO In This Room Should Be Paying Attention

On major developer marketplaces, open weight models built in China now occupy the top spots by weekly usage. That is not a fringe trend anymore.

Enterprise teams are increasingly experimenting with cheaper open source and open weight models as AI costs climb, and engineers are the ones driving that shift from the ground up.

Where AI adoption once happened regardless of model choice, leadership teams are now weighing cost against capability before committing to any single provider.

That single change in mindset is reshaping vendor negotiations across every industry currently scaling its AI footprint.

The Closed Versus Open Divide Is Splitting The Industry In Two

American labs have largely kept their most capable models closed, arguing that openness invites misuse and undermines security. Chinese labs have gone the opposite direction, releasing openly and often.

That divide is no longer theoretical. It is now shaping which vendors your engineering teams reach for by default when building new products.

Openness gives developers the ability to inspect, customise, and run models on their own infrastructure. Closed systems give the maker tighter control over security and pricing.

Your Enterprise Doesn't Need The Smartest Model. It Needs The Right One.

Industry investors now argue that open source models could eventually handle the vast majority of enterprise queries, leaving only the hardest problems for premium providers.

Routine coding, summarisation, data extraction, and customer service rarely require the most powerful model available. That reality is changing how leadership teams justify AI spend internally.

Reserving premium models for the hardest problems, while routing everyday work to cheaper systems, is quickly becoming standard practice across enterprise engineering teams.

The Truth Nobody Admits

What This Means For Your Next Board Conversation

Some American companies are already responding, launching their own open-weight models built for deep customisation, while chip makers expand open model families to drive further demand.

The real question for your leadership team is not which single model wins. It is whether your organisation has a deliberate strategy for mixing open and closed systems.

The labs and enterprises that build that strategy early will control their costs and their roadmap. The ones that wait will simply react to whatever comes next.

This shift is moving faster than most planning cycles account for, and the gap between fast movers and slow movers will only widen from here.

The AI Model Race Is A Distraction. Here's Why.

You have watched major AI labs give away some of their most valuable technology this year, handing key protocols to open foundations instead of guarding them behind closed doors.

That is not generosity. When a company gives away a successful piece of its stack, it is signalling that it plans to compete somewhere else entirely.

Other major labs made similar moves within months, donating their own protocols to the same kind of open governance structure.

Why Giving Something Away Is Actually A Power Move

This pattern has played out before, in cloud computing, in developer tools, in on-premises infrastructure. Platforms rarely release technology they still depend on for their edge.

They release it once competitive advantage has already shifted to a different layer of the stack, one they intend to own instead.

This has happened in containers, in version control, in cloud infrastructure. The free layer becomes the on-ramp to something the provider still charges for.

Your Model Choice Isn't The Moat You Think It Is

Frontier model leaderboards shift almost weekly. Betting your enterprise strategy on any single vendor staying permanently ahead on raw model quality is a fragile plan.

The real work, and the real value, sits in connecting models to enterprise data, workflows, permissions, and governance. That layer is far harder to replicate.

Every AI provider still invests billions in bigger models because that attracts talent and attention. But none of them expect that lead to hold for long.

The Gravity Well Your Competitors Are Already Building

Every new integration your teams build makes a platform harder to leave. Every workflow added increases its pull on daily operations.

Enterprise incumbents chase this same gravity from the opposite direction. They do not need to own the frontier model, only to connect it to systems already running the business.

Leading AI providers understand this. They keep training bigger models because that attracts developers and headlines, but they know benchmark leadership alone builds nothing durable.

The real prize is becoming the default place where AI-assisted work happens, where teams, tools, and accumulated history are hardest to walk away from.

Open Standards Won't Save You From Lock-In

Open protocols make it easier to swap one compatible model for another. That sounds like freedom, and in a narrow sense, it is.

But your organisation stays deeply tied to wherever your prompts, evaluations, security policies, and employee habits have already accumulated. Standards reduce friction. They rarely remove advantage.

Neutral governance still matters, since nobody wants foundational infrastructure controlled entirely by a direct competitor. But open interfaces are not the same as open markets.

Interoperability lowers switching costs at the surface while leaving the deeper dependencies fully intact underneath.

What Enterprise Leaders Should Actually Focus On

Comparable patterns already exist across your infrastructure, and they point to the same lesson for your AI strategy every leadership team keeps missing.

That is where lasting advantage gets built, and it is the question every leadership team should be asking before its next AI investment.

Your Frontier AI Model Might Refuse You During A Breach

Picture your team mid-breach, racing to analyse attacker logs, and the AI model you rely on every day suddenly refuses to help.

That is exactly what happened to a major AI platform company last week. Its own commercial frontier models blocked the analysis it urgently needed during an active attack.

An attacker had used an autonomous AI agent system to access internal datasets and credentials. The company detected the breach, cut off access, and moved straight into incident response.

Log analysis needed to happen fast, and it needed a model capable of processing raw exploit code without treating every line as a threat to itself.

When Safety Guardrails Become A Liability

The incident required sending large volumes of real attack data, including exploit payloads and command artifacts, straight into a model for rapid analysis.

Commercial frontier models blocked the requests. Their safety systems could not distinguish a defender racing to stop an attacker from an attacker building new exploits.

That distinction matters enormously during a live breach, when every hour spent negotiating with a model's refusal is an hour the attacker keeps moving.

No security team wants to discover that gap for the first time in the middle of an actual incident.

The Workaround Nobody Wanted To Need

With time working against them, the security team turned to an open-weight model instead, one large enough to run entirely inside their own infrastructure.

Running the model locally meant sensitive attack data never left their controlled environment. It also meant no guardrail stood between the team and the analysis it needed.

The model they turned to carries several hundred billion parameters and performs close to the level of the leading closed frontier system it replaced that day.

Open Weight Models Are Closing The Capability Gap Fast

Several open weight models built overseas now approach the capabilities of leading closed frontier systems, particularly in coding and vulnerability discovery.

They often run at a fraction of the inference cost too. But the price advantage means little if a closed model simply declines to do the work at all.

Guardrails that trigger cautiously reduce misuse risk, but they also make a model less useful for the real defensive work security teams do every day.

A Policy Fight Your Security Team Should Be Watching

This incident has landed in the middle of a wider policy debate. Officials have floated restricting American companies from using foreign open-source models.

Critics argue that limiting access to capable open models on security tasks only weakens American teams facing attackers who already use every available tool.

Some officials have floated the idea of steering companies away from foreign open models entirely, citing governance and security concerns rather than banning them outright.

Enterprise leaders now face a genuine tension between policy pressure and the practical reality their own security teams encountered under fire.

What This Means For Your Incident Response Plan

If your organisation depends entirely on one closed model for security analysis, this incident should prompt an honest review of that dependency.

The next breach will not wait for a policy debate to resolve. Your response plan should already account for that reality today.

Your opinion matters!

Hope you loved reading our piece of newsletter as much as we had fun writing it. 

Share your experience and feedback with us below ‘cause we take your critique very critically. 

Thank you for reading

-Shen & Towards AGI team