Hello, this is Hamamoto from TIMEWELL.
When the news broke on May 28, 2026 that Anthropic had raised funding at a valuation of $965 billion, my honest first reaction was, "Did I misread a digit?"[^1] $965 billion is somewhere around 150 trillion yen. It far exceeds Toyota's market capitalization, and as the valuation of a single private company it sits almost in the territory of a national budget. And the fact that this is the number for a company barely five years old—one without a flashy consumer product like ChatGPT—tells you everything about what this event really represents.
This news is not simply another "big money flows into AI again" story. It was the moment when the tectonic shift toward what people call the agentic economy—AI changing its role from "a convenient tool that humans use" to "a workforce that works in place of humans"—became visible as a movement of investor capital. In this article, rather than the fine technical details, I want to focus on what that shift brings to management, to markets, and to the front lines of Japanese companies.
What the $965 Billion Figure Actually Means
Let me first lay out the facts. Anthropic's Series H raised $65 billion at a post-money valuation of $965 billion, announced on May 28, 2026[^1]. The lead investors were Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, and the round also folded in $15 billion of hyperscaler investment, including $5 billion from Amazon[^2]. This came just three months after the Series G in February 2026.
What makes the number startling is the reversal it marks against its rival. At the end of March 2026, OpenAI raised $122 billion at an $852 billion valuation and had begun preparing for an IPO (Initial Public Offering—listing shares on a stock exchange to raise capital from public investors)[^3]. Anthropic has now, for the first time, overtaken OpenAI, which had long held the top valuation in the AI industry[^4]. Lining up where the major AI labs stand makes the scale of the shift clear.
| AI lab | Valuation | Latest raise | Date | Notes |
|---|---|---|---|---|
| Anthropic | $965 billion | $65 billion (Series H) | May 2026 | Most valuable AI startup in the world[^1] |
| OpenAI | $852 billion | $122 billion | March 2026 | Targeting a Q4 IPO[^3] |
Another basis for the valuation that you can't overlook is revenue growth. Anthropic disclosed that its run-rate revenue (the most recent revenue pace annualized) crossed $47 billion in May 2026[^1]. That works out to more than a tripling—from roughly $14 billion at the time of the Series G—in just a few months[^4]. Investors are justifying this valuation not with a "dream" but by looking at a revenue curve that already exists. This, I feel, is the decisive difference from the AI-bubble arguments of the past.
One more symbolic element is the share of the raise devoted to securing compute. Amazon is adding 5 gigawatts of new capacity; a partnership with Google and Broadcom brings 5 gigawatts of next-generation TPUs (semiconductors specialized for AI processing); and SpaceX facilities supply 10 gigawatts of GPUs. Together, around 20 gigawatts of computing capacity were locked up in this round[^2]. More than the money itself, it is the scramble for the electricity and semiconductors that power AI that has become the real contest. Behind the valuation, what is happening is less a pure technology race than a vast land grab over infrastructure. For Japanese companies, going head-to-head on this turf is not realistic. That is exactly why the decisive battleground becomes how to make the completed foundation actually work in your own operations.
To put 20 gigawatts in perspective, a single large nuclear reactor produces on the order of one gigawatt. Locking up twenty times that, just for one company's compute roadmap, signals that the limiting factor in AI is no longer "do we have a clever enough model" but "can we physically secure the power and the chips to run it at scale." This is a useful frame for executives outside the AI industry. When the scarcest resource shifts from algorithms to infrastructure, the advantage stops being about who can build the biggest model and starts being about who can apply an already-built model most precisely to a real problem. The frontier labs will keep fighting over gigawatts. Everyone else competes on fit—on how cleanly the technology slots into a specific workflow and produces a measurable result. For the vast majority of companies, including in Japan, that second game is the only one worth playing, and it happens to be the one where domain knowledge and process discipline matter more than capital.
Interested in leveraging AI?
Download our service materials. Feel free to reach out for a consultation.
How Claude Opus 4.8 Changed the Range of "Work You Can Hand to AI"
On the same day as the funding, Anthropic also announced a new model, Claude Opus 4.8[^5]. I don't want to drag too much technology into a management discussion, but this one point has to be touched on—otherwise it's impossible to explain why $47 billion in revenue gets generated. The key is a feature called Dynamic Workflows.
AI up to now has been, in effect, an excellent assistant: you give one instruction, it returns one piece of work. Dynamic Workflows, given a goal, plans the work itself and runs hundreds of sub-agents (small task-handling AIs carved out from the parent AI) in parallel within a single session[^5]. Picture work that would take a human several weeks—say, migrating an entire source base of hundreds of thousands of lines to a different specification—where AI takes on everything from the planning to the execution. You could rephrase it as AI moving from "a tool waiting for instructions" to "a working unit that organizes the plan and acts."
Another point that carries weight in practice is the improvement in quality. Opus 4.8 is reported to have cut the probability of overlooking a flaw in code it wrote itself to roughly a quarter of the previous generation[^5]. When you hand a substantial job to AI, the scariest thing has always been that it confidently produces a deliverable that is "plausible but wrong." The improved precision of its self-checking reliably widens the range of work the front line can entrust with peace of mind. On top of that, there is Effort Control, which lets you choose how much computing resource to pour into the processing, so you can use it differently—quick for light lookups, deliberate for heavy analysis[^5].
It's easily missed, but the price was held flat. At $5 per million input tokens and $25 per million output tokens, the level was maintained while what it can do simply grew[^5]. A token is the smallest unit AI uses when processing text—roughly speaking, it's "metered billing in proportion to the volume of characters processed." Performance rises, but the unit price doesn't. This single point is quietly but surely lowering the economic hurdle for companies to deploy AI at scale.
It's worth pausing on why that matters more than it sounds. In most software, you pay roughly the same regardless of whether the result is good. With AI, the cost per token is fixed, but the value per token keeps climbing as the model gets better at doing useful work in fewer steps. A model that plans well and checks its own output finishes a task with less wasted back-and-forth, which means the effective cost of a completed job falls even when the headline token price doesn't move. For a finance team trying to forecast what large-scale AI use will cost, this is the difference between a budget line that swells unpredictably and one that holds steady while output grows. The combination of flat pricing, parallel sub-agents, and self-checking is what turns "AI is interesting" into "AI is something we can put on a spreadsheet and justify."
The "47% Growth" Market That Investment Capital Is Front-Running
The enormous investment in Anthropic is not an isolated phenomenon. The whole market is facing the same direction. On May 19, 2026, Gartner forecast that worldwide AI spending would reach $2.59 trillion in 2026, up 47% year on year[^6]. The largest segment of this is AI infrastructure, said to account for over 45% of the total. AI-optimized servers are projected to triple over the next five years[^6].
What's interesting here is Gartner's point that "2026 will be the inflection point for enterprise spending"[^6]. Until now, AI spending has been driven by tech companies and hyperscalers (the giant cloud operators). In other words, it was the stage where "the side selling AI" invested in infrastructure. The read going forward is that "the side using AI"—ordinary enterprises—will begin opening their wallets in earnest. Anthropic and OpenAI competing on valuation is precisely because they are reaching for the moment just before that enterprise spending takes off.
In fact, rival OpenAI is also shifting its strategic center of gravity. Alongside research and development, it has repositioned itself as a "deployment" company that delivers AI into real operations, raising the weight it places on enterprise[^3]. The main battlefield of the AI labs has moved from a numbers race on performance benchmarks to "how to be used on the front lines of companies and generate revenue." The $965 billion figure is, as I see it, the result of investors pricing that transition in ahead of everyone else.
To executives who have started wondering, "Could we use AI agents ourselves?" The first thing to check is not the technical specs but whether your own data and operations are in a state that can be entrusted to AI. With TIMEWELL's AI adoption consulting, WARP, we start by diagnosing where you stand and design together which operations to entrust, how far, and in what order. Steady your footing before being swept up by the trend. That, which looks like a detour, is in my view the fastest road.
The Reality of Adoption: The Heat of Investment vs. the Temperature on the Ground
So far I've kept up a story full of heat. But shift your eyes to the front lines and the scenery changes completely. McKinsey's 2026 survey confronts us with that gap in cool, hard numbers[^7].
According to the survey, 88% of organizations are using AI in some form of work. Yet only 23% of companies can run AI agents at scale across the whole company, with 39% still at the experimental stage. Looking at individual business functions, the share able to run agents in production tops out at 10% or less[^7]. In other words, the overwhelming majority are companies that "have touched it but can't fully entrust to it."
When it comes to ROI (return on investment), an even harsher reality comes into view. A meta-analysis by Stanford HAI confirms productivity gains on the front lines where AI has actually been deployed—improvements of 14–15% in customer support, 26% in software development, and up to 50% in marketing production output[^7]. On the other hand, in PwC's global CEO survey, 56% of executives answered that "AI has not yet led to either revenue growth or cost reduction"[^7]. It works on the front line, yet it doesn't show up in the management numbers. Only 39% of companies can feel an impact on EBIT (earnings before interest and taxes)[^7]. This divergence is, I think, the single biggest obstacle of the agentic economy.
Why can't they fully entrust it? McKinsey reports that as the biggest barrier to running agents at full scale, about two-thirds of respondents cited concerns over security and risk. That ranked above regulatory uncertainty and technical limitations[^7]. Anxiety about handing important internal data and operations to autonomously acting AI is what stops the front line in its tracks. The harder investment capital steps on the accelerator, the larger this "wall of trust" looms by comparison.
There's a logic to this that I think is healthy rather than timid. An assistant that drafts a paragraph is low-stakes: a human reads it and decides what to keep. An agent that, on its own initiative, queries internal systems, moves data between them, and takes actions across hundreds of parallel steps is a different category of risk. The failure mode is no longer "the output was a bit off" but "it did something at scale before anyone noticed." That is exactly the kind of risk a security or compliance team is paid to refuse until the controls are in place. So the bottleneck in the agentic era is not model capability—the models are already capable enough for most operations—but governance: who can see what, what an agent is allowed to do, and how every action is logged and reversible. Companies that treat this as an afterthought stall at the experiment stage. Companies that build it in from the start are the ones that reach production, and the data backs that up.
Japanese Companies' Next Move: Steadying the Footing, Not Stretching
Turning to Japan, the picture gets even more concrete. According to the Ministry of Internal Affairs and Communications' FY2025 White Paper on Information and Communications, companies that answered they had already adopted, or were preparing to adopt, language-based generative AI came to 41.2%, a large jump from 26.9% the previous year[^8]. A Nikkei xTECH survey put generative AI adoption at 64.4%, while AI agent adoption stayed at 29.7%[^9]. Japan shares the same pattern as the rest of the world: companies have cleared the entrance to adoption, but the stage of entrusting work to agents is still ahead.
It's also worth noting that the leading challenge cited by adopting companies is "we don't know how to use it effectively," followed by "security risks such as leaks of internal information"[^8]. It maps cleanly onto the shared global worry McKinsey identified. That is precisely why I believe what Japanese companies should be doing now is not stretching themselves, overwhelmed by the digits of a valuation, but steadying their footing. Concretely, there are three things.
The first is to put in place a form you can use without sending data outside the company. The core anxiety about generative AI is information leakage. With a configuration that keeps confidential information flowing to overseas clouds, the front line can't genuinely entrust work to it. A mechanism that runs on domestic servers and lets you control which data AI sees becomes the foundation of trust. The reason our enterprise AI, ZEROCK, insists on AWS domestic servers and Knowledge Control (a mechanism for finely controlling the scope of internal information shown to AI) is precisely to get over this wall.
The second is to make AI reference your own knowledge accurately. No matter how clever a general-purpose model is, if it doesn't know your internal rules and past cases, all it returns is plausible-sounding generalities. The GraphRAG that ZEROCK adopts (a technology that captures the relationships among internal documents as a graph structure so AI can draw out accurate answers with the context in mind) is exactly the mechanism for "answering in your own context." The premise of entrusting work to an agent is creating a state where it is looking at the right knowledge.
The third is to start small and carry it into operations. As McKinsey's survey shows, the majority of companies stall at experimentation. Adopt it for one focused operation, confirm the effect in numbers, settle it into operations, and only then expand to the next. It's unglamorous, but on the front line I've come to feel this is the only road out of "merely touching it." The reason Anthropic could triple its revenue in three months is precisely that this kind of "adoption that has reached operations" is happening simultaneously all over the world.
These three are deliberately unglamorous, and that is the point. None of them require chasing the next valuation headline or betting on which lab wins the gigawatt race. They are about putting your own house in order so that, when the models keep improving—and they will—you are positioned to absorb the gains rather than read about them in someone else's case study. The companies that benefit most from the agentic economy will not be the ones that adopted earliest or spent the most. They will be the ones whose data was clean, whose knowledge was referenceable, and whose first pilot had a number attached to it.
The agentic economy will reach Japan whether we do anything or not. The question is whether, when that wave arrives, you hold the footing that lets you entrust work to it. Whether you end the $965 billion news with "wow, amazing story," or turn it into a trigger to re-examine your own footing—that difference, I believe, will divide competitiveness a few years from now. It's fine to start simply by writing down on paper which of your operations today's AI could plausibly handle. The design beyond that is something people who know the front line of implementation, like us, will think through with you.
To those who want to consider AI agent adoption not as a trend but as work for their own company.
On a related note, if you'd like to know a bit more about the technical details of Dynamic Workflows, see A Comprehensive Guide to Claude Dynamic Workflows, and for the context of Anthropic's enterprise strategy, Anthropic and Blackstone's Enterprise AI Services is worth a look as well.
Footnotes
[^1]: Anthropic, "Anthropic raises Series H at a $965 billion post-money valuation" (May 28, 2026) https://www.anthropic.com/news/series-h [^2]: Axios, "Anthropic tops OpenAI as most valuable AI startup" (May 28, 2026) https://www.axios.com/2026/05/28/anthropic-ai-fundraising-openai [^3]: CNBC, "OpenAI closes funding round at an $852 billion valuation" (March 31, 2026) https://www.cnbc.com/2026/03/31/openai-funding-round-ipo.html [^4]: CNBC, "Anthropic tops OpenAI as most valuable AI startup, nears $1 trillion valuation in latest round" (May 28, 2026) https://www.cnbc.com/2026/05/28/anthropic-open-ai-startup-value.html [^5]: Anthropic, "Claude Opus 4.8" (May 28, 2026) https://www.anthropic.com/news/claude-opus-4-8 [^6]: Gartner, "Gartner Forecasts Worldwide AI Spending to Grow 47% in 2026" (May 19, 2026) https://www.gartner.com/en/newsroom/press-releases/2026-05-19-gartner-forecasts-worldwide-ai-spending-to-grow-47-percent-in-2026 [^7]: McKinsey, "State of AI trust in 2026: Shifting to the agentic era" (2026) https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-era [^8]: Ministry of Internal Affairs and Communications, "FY2025 White Paper on Information and Communications: The Current State of AI Use in Enterprises" https://www.soumu.go.jp/johotsusintokei/whitepaper/ja/r07/html/nd112220.html [^9]: Nikkei xTECH, "Generative AI tool adoption among Japanese companies is 64.4%, AI agents 29.7%" https://xtech.nikkei.com/atcl/nxt/column/18/03314/090800004/ [^10]: OpenAI, "News" https://openai.com/news/