Meta's Shift to Closed Models and Anthropic Surpassing $30 Billion in Annual Revenue—Reading the 'Next Investment Opportunity' from AI Money Flows
The flow of money surrounding AI is entering a new phase.Meta has announced its first closed model, 'NewSpark,' signaling a clear departure from the open-source path it has long maintained.Anthropic has surpassed an annual revenue run rate of $30 billion, and a secondary sale at a $350 billion valuation saw an unusual 'seller shortage.' Based on Bloomberg Tech's reporting, we break down the front lines of AI investment.
1. Meta's AI Model Strategy Shift—NewSpark and the Pivot to Closed Models
1-1. A 'Major Pivot' from Open Source
Meta has long positioned its open-source Llama series as its flagship. However, according to Bloomberg Tech, since early 2025, Llama has been criticized for its performance gap compared to frontier models like OpenAI, Anthropic, and Google Gemini.
The new model NewSpark is the first closed model developed by Meta's 'Super Intelligence Labs,' designed so that developers cannot access the backend code. This is seen as a clear strategic shift focused on monetization.
Reporter Riley Griffin stated on the program:
'When you listen to analyst assessments, they are saying, "Meta is back in the game." This is a very significant development.'
Alex Wang, who leads the lab, posted on Instagram, positioning Meta's AI model as 'fourth in the industry.' Rather than claiming the top spot, it appears to be a self-recognition that they are 'back in the game.' Following the news, Meta stock rose by approximately 10% at one point.
1-2. Training with Chinese Models—What Using Alibaba's 'Qwen' Means
A notable point about NewSpark is that it used several third-party open-source models for training, including Alibaba's Qwen. As the AI supremacy race between the U.S. and China intensifies, many major U.S. tech companies are moving to prevent Chinese firms from training on their models.
Griffin pointed out:
'In the U.S.-China AI competition, multiple tech leaders have stated the need to protect their advantage from a national security perspective, and some are showing moves to block Chinese companies from training on their models. Meta's approach this time is the opposite—"learning from Chinese models"—which in itself is worth noting.'
Meta states that this decision was made after implementing security measures, but it may spark debate within the industry.
2. Anthropic—Surpassing $30 Billion in Annual Revenue and the 'Seller Shortage' Secondary Sale
2-1. The Arrival of the $350 Billion Valuation Era
According to Bloomberg reports, Anthropic has completed a secondary sale for employees. The valuation based on new funding was set at $350 billion.
However, the result was unexpected. Investors had prepared a purchase quota of approximately $6 billion, but employee willingness to sell was low, leaving the offering undersubscribed.
Bloomberg reporter Rebecca analyzed it this way:
'Anthropic's growth this year is phenomenal. The company announced this week that its annual revenue run rate has exceeded $30 billion. For employees, a $350 billion valuation likely didn't seem "high enough yet."
2-2. 'All About Anthropic' at the HumanX Conference
At the HumanX conference held in San Francisco (with approximately 7,000 attendees), Anthropic was effectively the center of attention. Rebecca stated the following:
"In almost every conversation, Anthropic's name came up without even being asked. For many investors and startups, it has become the benchmark against which they measure themselves."
Anthropic is establishing a dominant position in the AI coding space, and further expansion into other areas such as financial services is expected. Regarding the new model released this week, the judgment that it was 'too powerful to release to the public' has reportedly caused a sense of crisis among some founders.
3. Outlook for AI Infrastructure Investment—A J.P. Morgan Perspective
3-1. The 'Compute Shortage' Continues
Stephanie Aliaga of J.P. Morgan Asset Management spoke about the outlook for AI capital expenditure on Bloomberg Tech:
"Growth prospects are inextricably linked to demand. In this quarter's earnings, every hyperscaler stated, 'If we had more capacity today, our revenue would have been even higher.' We remain in a deeply compute-constrained environment."
3-2. How Geopolitical Risk Spills Over into AI Hardware Costs
Aliaga also touched upon the risks posed by the situation in the Middle East.
"The economies most dependent on oil—Taiwan, South Korea—are also the most critical suppliers of AI hardware. If high crude oil prices persist, those costs will be passed on to U.S. software companies."
Ceasefire negotiations in the Middle East are ongoing, but with both Iran and the U.S. continuing to accuse each other of violations, uncertainty remains for the market. However, Aliaga views a scenario where the conflict resolves in a matter of days or weeks as more probable.
3-3. Technology Stock Valuations are 'Historically Cheap'
On the other hand, she pointed out that tech stock valuations have contracted to levels cheaper than the consumer staples sector. Some major stocks are trading at lower multiples than when the AI boom began, describing a structure where 'investors are paying a lower price for robust earnings.'
"Despite geopolitical headwinds, earnings forecasts have actually been revised upward. The factor behind this is the strength of the AI narrative."
4. NVIDIA Technical Breakout and the Divergence Between Semiconductors and Software
4-1. Escaping Nine Months of Stagnation
According to Bloomberg market reporter Ryan Vlastelica, NVIDIA's stock price had been moving mostly sideways for about nine months, but it has recorded a gain of over 10% in the last six sessions, showing noteworthy technical movement.
"If the stock price maintains the $185 level, it shows that investors are returning to this stock. It can be seen as a sign of a breakout toward the $200 range and even all-time highs."
4-2. Semiconductors vs. Software—A Zero-Sum Composition
While chip stocks have seen gains of over 20% year-to-date, software stocks have fallen significantly. Aliaga pointed out regarding this disparity that 'the market is viewing the AI wave as a zero-sum game.'"
While concerns that AI will render existing software obsolete weigh heavily on software stocks, the view remains that demand for AI infrastructure—a broad range of semiconductors including not just GPUs, but also memory, CPUs, and custom silicon—will continue to expand.
However, Aliaga added, 'The fundamental fundamentals of software have not deteriorated. The market's cautious stance is simply being reflected in valuations, and there is still room for software to reinvent itself.'
5. Other Notable Topics
5-1. Amazon's Chip Business: $20 Billion Run Rate
Amazon CEO Andy Jassyrevealed in his annual shareholder letter that the company's chip business has reached an annual revenue run rate of over $20 billion, achieving triple-digit year-over-year growth. He also mentioned the potential to reach a $50 billion scale if external sales ramp up in earnest.
5-2. OpenAI Pauses UK Stargate Project
OpenAIhas paused its data center construction project in the UK, citing high energy costs and regulatory challenges. The UK faces issues with some of the highest energy costs in Europe, and obtaining planning permission is also difficult.
Bloomberg Tech reporter Shona Ghosh stated the following:
'We are seeing the reality that not all announced infrastructure investments will come to fruition. The issue of energy costs affects not only OpenAI but also other data center providers within the UK.'
5-3. Nico Rosberg on the Winning Strategy for AI Venture Investment
Former F1 World Champion and current venture capitalist Nico Rosberg spoke at the HumanX conference as follows:
'The speed of innovation right now feels faster than driving an F1 car. You cannot sit on the sidelines. You need to invest and get involved. However, diversification—both across sectors and across time horizons—is essential.'
He holds a portfolio with direct investments in about 35 companies and pointed out that in the era of AI, returns in venture capital are increasingly concentrating among the top 10 to 12 large multi-stage funds. He expressed the view that venture capital is currently the most effective means of exposure, as much value creation occurs in the private stage before an IPO.
