Micron Technology (MU) stock investors had plenty of reason to expect another ‘DeepSeek moment’ following the launch of Kimi K3.
The concerns weren’t new, though, and pretty straightforward.
Cost-effective, more efficient AI could weaken demand for the expensive hardware powering the boom.
Wall Street was questioning the relentless AI spending cycle, but in a note shared with me, Bank of America moved in the opposite direction.
The bank reiterated its Buy rating and $1,550 price target on Micron, arguing the latest AI disruption doesn’t break the memory thesis.
Instead, BofA sees the market focusing on the wrong part of the equation.
Kyle Green/Bloomberg via Getty Images
Kimi K3 raises the stakes in the AI race
According to Wikipedia, Chinese AI startup Moonshot AI, which launched Kimi K3 on July 16, was viewed as a major escalation in the country’s push to narrow the gap with U.S. AI developers.
More AI:
- Workers just sent AI companies an ultimatum
- Palantir CEO has a blunt verdict on OpenAI and Anthropic
- Elon Musk pulls no punches with AI rivals as Grok 4.5 debuts
Reuters described Kimi K3 as the world’s largest open-weight AI model, with Moonshot claiming it has 2.8 trillion parameters, a 1 million-token context window, and capabilities designed for coding, knowledge work, and deep reasoning.
What stood out was Kimi efficiently combining frontier-level performance claims with open weights and considerably lower deployment costs than most of its Western alternatives.
Users can download, customize, and operate the model according to their specific requirements rather than relying on closed APIs.
On top of that, the release underscored how China’s open-model ecosystem is rapidly advancing, increasing competitive pressure on OpenAI and Anthropic while potentially broadening demand for the chips, memory, and storage needed to run powerful models.
Interestingly, as I recently covered, Microsoft CEO Satya Nadella has been sounding the alarm over customers effectively paying AI companies twice, indirectly supporting the rationale behind Kimi K3’s open-weight release.
Businesses can eventually deploy models privately while retaining greater control over prompts, corrections, and workflows that create institutional knowledge.
Why does BofA think Kimi’s efficiency still increases memory demand?
BofA’s argument hinges on separating compute efficiency and memory capacity.
For perspective, Kimi K3 is a 2.8 trillion-parameter mixture-of-experts model, but it activates roughly 50 billion parameters per token, or roughly 1.8% of the total. Naturally, that substantially reduces the amount of computation needed for inference.
However, it hardly has an impact on the entire expert pool needed in fast memory.
BofA estimates Kimi K3 still requires about 1.4 TB of HBM at native MXFP4 precision and at least 64 accelerators per serving instance. In comparison, OpenAI’s 120-billion-parameter open model needs around 63 GB (roughly 23 times lower).
The broader trend is perhaps even more important.
DeepSeek V3 had 671 billion parameters, Kimi K2 reached 1 trillion, DeepSeek V4 Pro rose to 1.6 trillion, and Kimi K3 reached 2.8 trillion.
Compression can reduce memory per parameter by 50%, but model sizes are growing almost in tandem. That means, even though efficiency per computation is rising, things are getting a lot more memory-intensive per deployment.
What happened to U.S. tech stocks?
DeepSeek and Kimi K3 sparked similar “China AI shock” moments, sending shockwaves through U.S. tech stocks.
When DeepSeek’s R1 gained attention in January 2025, AI poster childNvidia (NVDA) plummeted 17%, losing $589 billion in one session, according to Yahoo Finance, while AI-linked chip, power, and infrastructure stocks collectively lost over $1 trillion.
Kimi K3 stoked those fears in July 2026.
On July 17, according to Investopedia reporting, the Nasdaq fell 1.4%, the S&P 500 lost 1.01%, and the semiconductor index slid 1.6%, as geopolitical tensions pressured stocks and the market entered a bear market.
Why could open-weight AI expand Micron’s addressable market?
BofA feels open-weight models are a multiplier for memory demand, as they effectively shift infrastructure ownership from a handful of hyperscalers to thousands of individual users.
If users opt for self-hosting models such as Kimi or DeepSeek, they must provision for a tremendous amount of memory.
That switches up the economics of the market.
An enterprise deployment needs roughly 80 GB of HBM3E; a DeepSeek V3 cluster might require roughly 1.1 TB, and each Kimi K3 instance entails a 64-plus-accelerator system. Ten thousand customers making use of one closed API might share a limited number of model copies.
Ten thousand self-hosted deployments create ten thousand separate memory endpoints.
The benefit extends beyond HBM.
Hot weights sit in HBM; inactive experts spill into DDR5 or LPDDR5X, while cold KV cache and model files shift into enterprise NAND. Context windows between 128,000 and 1 million tokens can push KV-cache requirements above 40 GB per active session.
Cheaper pricing increases usage.
BofA says global token volume jumped from about 9% per week in 2026, while multi-agent workloads can generate nearly 200,000 output tokens per query versus 200 for basic chat.
What does Bank of America think Micron is worth?
According to Seeking Alpha, Micron stock is currently trading at $970.82, which means BofA’s $1,550 price target implies approximately 60% upside.
BofA used a sum-of-the-parts valuation, which includes the following:
- $1,040 per share for Micron’s traditional cyclical memory business, valued at 3 times estimated calendar-2028 book value
An AI HBM component valued at 31 times calendar-2028 earnings, in line with the median AI compute peer multiple.. - By subtraction, the HBM component accounts for roughly $510 of the $1,550 price target, although BofA doesn’t separately state that figure.
BofA also states that Micron’s CHIPS Act-related buyback restrictions are expected to expire around December 2026.
Against its projected $120 billion to $130 billion-plusfree cash flow outlook, the bank estimates a 40% payout policy to support $50 billion to $60 billion in share buybacks, equivalent to around 4.5% to 5.3% of the current market cap.
Despite its nosebleed valuation, Micron stock is still trading at 13.2 times forward non-GAAP earnings, 46% lower than the sector median and 82% lower than its 5-year average, according to Seeking Alpha.
Related: Microsoft CEO’s Anthropic criticism reveals bigger AI power struggle
