Microsoft eyes a cheap Chinese brain for Copilot as the AI bill comes due

When Microsoft starts shopping for a cheaper model, pay attention

On June 18, 2026, the Seoul Economic Daily — citing reporting by Axios — reported that Microsoft is reviewing whether to add DeepSeek, a Chinese AI model, to its enterprise tool Copilot Cowork. The product currently runs on models from OpenAI and Anthropic. Microsoft is considering offering a lower-cost option, built on a fine-tuned version of DeepSeek V4 or another open-source model, that customers could choose directly — priced below the ChatGPT- or Claude-powered versions, with a release reportedly planned within weeks.

Note the framing carefully: this is under review, not decided. But the reason behind it is what matters, and Microsoft was blunt about it. The reporting attributes the move primarily to cost. When a product runs on a closed model like ChatGPT or Claude, the provider pays more in usage fees as customers consume more. Run an open-source model on your own servers instead, and you mostly just pay for the servers.

Then came the line that should stop every business owner mid-scroll. Charles Lamanna, a corporate vice president in Microsoft's Copilot division, said the company's testing led it to conclude that offering the tool on an unlimited-use basis is impossible — noting that some users run hundreds of tasks a week, impressive productivity that drives costs very high. As a result, the report says, Microsoft is shifting Copilot Cowork from a flat-rate plan to usage-based pricing.

Read that again. The company that did as much as anyone to put AI in front of the world has just concluded that "unlimited AI for a flat fee" doesn't work — and is now choosing models by cost, and charging by consumption. If that's the reality at Microsoft's scale, it is the reality at yours too.

‍

The "token economy" has arrived — and your business is already in it

This isn't an isolated Microsoft decision. It's the leading edge of a shift the industry is now naming out loud.

The same report points to data from Ramp, which handles corporate cards and expense software for more than 70,000 US companies. Its analysis of May spending found DeepSeek ranked first among the fastest-growing software — and Ramp framed it bluntly: AI prices are rising across most providers, the era of subsidies and flat-rate plans is fading, and companies are increasingly choosing models based on cost-effective performance. Ramp called it "an early signal of the token economy" — the world in which AI is priced by the token, the smallest unit of data a model processes, and every business has to think about consumption the way it thinks about electricity.

The broader trend is real: the report notes that Chinese models like DeepSeek and Alibaba's Qwen, competing hard on price-for-performance, have surged in developer adoption. The plain takeaway for an owner-led firm is this: the comfortable phase — where AI felt like a cheap, fixed monthly subscription — is ending. AI is becoming a metered utility. And in a metered world, which model you use for which task stops being a technical footnote and becomes a line on your budget.

The trouble is that almost no small or mid-sized business is making that choice on purpose.

‍

Cheaper is a real option - but it's a trade off, not a free lunch

Microsoft's move also surfaces the tension that makes this a decision rather than a no-brainer: the cheapest capable model on the market right now is Chinese, and that carries baggage.

The report is candid about it. A growing number of US companies are adopting DeepSeek to cut costs, which has raised concerns about security and national-security exposure tied to Chinese AI. Microsoft's stated answer, per the reporting, is that even if it adopts DeepSeek, customer data would stay within Microsoft's cloud under Azure's enterprise security and compliance controls, cut off from Chinese servers so information doesn't flow to China. Separately, US authorities have been weighing how to treat DeepSeek at a policy level.

We take no position on DeepSeek's safety — that's a live, contested question, and not one a business should resolve from a headline. The point for you is the shape of the decision, not the verdict. Every model choice is a balance of cost, capability, privacy, and risk:

• The most capable frontier models are powerful, expensive, and the most likely to be restricted or to change.
• Cheaper open-source models can deliver strong value — but raise questions about where they run, who built them, and what data touches them.
• Where a model physically runs — a US cloud, your own servers, a box on your premises — changes your privacy and continuity exposure entirely.

There is no universally "right" model. There is only the right model for a given task, given your tolerance for cost, privacy, and risk. Microsoft is making that calculation deliberately, with a division of experts. The question is whether your business is making it at all.

‍

Most businesses aren't choosing. They're drifting.

Here's the uncomfortable reality inside the typical small or mid-sized firm. Nobody sat down and selected an AI model. A staff member started using whatever tool was in front of them, it worked, and it quietly became part of how the business runs — with no one weighing the cost as usage climbs, no one checking where the data goes, and no fallback if the price jumps or the tool changes.

That was survivable when AI felt free. In the token economy, it's expensive. The Microsoft story is the enterprise-scale version of a problem that's playing out, invisibly and at smaller dollar figures, inside firms like yours right now: usage no one is metering, model choices no one made on purpose, and spend creeping upward with nothing tying it to a result.

Deliberate beats accidental — and deliberate is a discipline, not a one-time setting.

This is exactly the work AgentsROI.ai does. We turn an accidental, drifted-into AI setup into a governed, cost-aware, and resilient one:

• A Shadow-AI Risk Assessment & AI Governance Audit shows you what AI your team is actually using, where your data is going, and what it's costing you — the visibility Microsoft has and most SMEs don't.
• Model Selection & Continuity Planning matches each workflow to the right model in the right place — premium where it earns its keep, cheaper where it makes sense, local where privacy or cost demand it — with a fallback for anything business-critical.
• ROI Measurement & Reporting keeps every dollar of AI spend tied to a result, so a usage-based future is a budget you control, not a bill you brace for.

And because we're vendor-neutral, we have no stake in which model you land on. Our only job is to make sure you chose it — on cost, privacy, and risk — instead of inheriting it by accident.

The flat-rate era is ending. Choose before the meter does.

Microsoft just told the market, in plain terms, that unlimited AI for a fixed fee doesn't add up — and that even the biggest players now pick their models by cost and charge by use. That future doesn't arrive someday. For your business, it's already here; you simply may not have looked yet.

The firms that come out ahead won't be the ones using the most AI, or the cheapest, or the most powerful. They'll be the ones who chose — deliberately, with their eyes open to cost, privacy, and risk — instead of drifting into whatever was easiest and discovering the price later.

Book a Shadow-AI Risk Assessment and find out exactly what AI your business is using, where your data goes, and what it's costing — before the meter makes the decision for you.

This article references reporting by the Seoul Economic Daily (June 18, 2026), citing Axios, and data attributed to Ramp. Microsoft's consideration of DeepSeek is described in those reports as under review and not finalized; statements attributed to Microsoft and others reflect their publicly reported positions. This article takes no position on the safety, security, or merits of any specific AI model or provider. It is for general informational purposes only and does not constitute legal, financial, security, or compliance advice. Consult a qualified professional regarding your specific obligations.

‍